path toward faster partition pruning

From 1da6477fe698ddd8123d54274fb327dccb559043 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 8 Aug 2017 18:42:30 +0900
Subject: [PATCH 1/6] Teach pg_inherits.c a bit about partitioning

Both find_inheritance_children and find_all_inheritors now list
partitioned child tables before non-partitioned ones and return
the number of partitioned tables in an optional output argument

We also now store in pg_inherits, when adding a new child, if the
child is a partitioned table.

Per design idea from Robert Haas
---
 contrib/sepgsql/dml.c                  |   2 +-
 doc/src/sgml/catalogs.sgml             |  10 +++
 src/backend/catalog/partition.c        |   2 +-
 src/backend/catalog/pg_inherits.c      | 157 ++++++++++++++++++++++++++-------
 src/backend/commands/analyze.c         |   3 +-
 src/backend/commands/lockcmds.c        |   2 +-
 src/backend/commands/publicationcmds.c |   2 +-
 src/backend/commands/tablecmds.c       |  56 +++++++-----
 src/backend/commands/vacuum.c          |   3 +-
 src/backend/executor/execMain.c        |   3 +-
 src/backend/optimizer/prep/prepunion.c |   2 +-
 src/include/catalog/pg_inherits.h      |  20 ++++-
 src/include/catalog/pg_inherits_fn.h   |   5 +-
 13 files changed, 200 insertions(+), 67 deletions(-)

diff --git a/contrib/sepgsql/dml.c b/contrib/sepgsql/dml.c
index b643720e36..6fc279805c 100644
--- a/contrib/sepgsql/dml.c
+++ b/contrib/sepgsql/dml.c
@@ -333,7 +333,7 @@ sepgsql_dml_privileges(List *rangeTabls, bool abort_on_violation)
 		if (!rte->inh)
 			tableIds = list_make1_oid(rte->relid);
 		else
-			tableIds = find_all_inheritors(rte->relid, NoLock, NULL);
+			tableIds = find_all_inheritors(rte->relid, NoLock, NULL, NULL);
 
 		foreach(li, tableIds)
 		{
diff --git a/doc/src/sgml/catalogs.sgml b/doc/src/sgml/catalogs.sgml
index ef7054cf26..00ba2906c2 100644
--- a/doc/src/sgml/catalogs.sgml
+++ b/doc/src/sgml/catalogs.sgml
@@ -3894,6 +3894,16 @@ SCRAM-SHA-256$<replaceable>&lt;iteration count&gt;</>:<replaceable>&lt;salt&gt;<
        inherited columns are to be arranged.  The count starts at 1.
       </entry>
      </row>
+
+     <row>
+      <entry><structfield>inhchildpartitioned</structfield></entry>
+      <entry><type>bool</type></entry>
+      <entry></entry>
+      <entry>
+       This is <literal>true</> if the child table is a partitioned table,
+       <literal>false</> otherwise
+      </entry>
+     </row>
     </tbody>
    </tgroup>
   </table>
diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index c92756ecd5..fe8e60de14 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -178,7 +178,7 @@ RelationBuildPartitionDesc(Relation rel)
 		return;
 
 	/* Get partition oids from pg_inherits */
-	inhoids = find_inheritance_children(RelationGetRelid(rel), NoLock);
+	inhoids = find_inheritance_children(RelationGetRelid(rel), NoLock, NULL);
 
 	/* Collect bound spec nodes in a list */
 	i = 0;
diff --git a/src/backend/catalog/pg_inherits.c b/src/backend/catalog/pg_inherits.c
index 245a374fc9..5292ec8058 100644
--- a/src/backend/catalog/pg_inherits.c
+++ b/src/backend/catalog/pg_inherits.c
@@ -33,6 +33,8 @@
 #include "utils/syscache.h"
 #include "utils/tqual.h"
 
+static int32 inhchildinfo_cmp(const void *p1, const void *p2);
+
 /*
  * Entry of a hash table used in find_all_inheritors. See below.
  */
@@ -42,6 +44,30 @@ typedef struct SeenRelsEntry
 	ListCell   *numparents_cell;	/* corresponding list cell */
 } SeenRelsEntry;
 
+/* Information about one inheritance child table. */
+typedef struct InhChildInfo
+{
+	Oid			relid;
+	bool		is_partitioned;
+} InhChildInfo;
+
+#define OID_CMP(o1, o2) \
+		((o1) < (o2) ? -1 : ((o1) > (o2) ? 1 : 0));
+
+static int32
+inhchildinfo_cmp(const void *p1, const void *p2)
+{
+	InhChildInfo c1 = *((const InhChildInfo *) p1);
+	InhChildInfo c2 = *((const InhChildInfo *) p2);
+
+	if (c1.is_partitioned && !c2.is_partitioned)
+		return -1;
+	if (!c1.is_partitioned && c2.is_partitioned)
+		return 1;
+
+	return OID_CMP(c1.relid, c2.relid);
+}
+
 /*
  * find_inheritance_children
  *
@@ -54,7 +80,8 @@ typedef struct SeenRelsEntry
  * against possible DROPs of child relations.
  */
 List *
-find_inheritance_children(Oid parentrelId, LOCKMODE lockmode)
+find_inheritance_children(Oid parentrelId, LOCKMODE lockmode,
+						  int *num_partitioned_children)
 {
 	List	   *list = NIL;
 	Relation	relation;
@@ -62,9 +89,10 @@ find_inheritance_children(Oid parentrelId, LOCKMODE lockmode)
 	ScanKeyData key[1];
 	HeapTuple	inheritsTuple;
 	Oid			inhrelid;
-	Oid		   *oidarr;
-	int			maxoids,
-				numoids,
+	InhChildInfo *inhchildren;
+	int			maxchildren,
+				numchildren,
+				my_num_partitioned_children,
 				i;
 
 	/*
@@ -77,9 +105,10 @@ find_inheritance_children(Oid parentrelId, LOCKMODE lockmode)
 	/*
 	 * Scan pg_inherits and build a working array of subclass OIDs.
 	 */
-	maxoids = 32;
-	oidarr = (Oid *) palloc(maxoids * sizeof(Oid));
-	numoids = 0;
+	maxchildren = 32;
+	inhchildren = (InhChildInfo *) palloc(maxchildren * sizeof(InhChildInfo));
+	numchildren = 0;
+	my_num_partitioned_children = 0;
 
 	relation = heap_open(InheritsRelationId, AccessShareLock);
 
@@ -93,34 +122,45 @@ find_inheritance_children(Oid parentrelId, LOCKMODE lockmode)
 
 	while ((inheritsTuple = systable_getnext(scan)) != NULL)
 	{
-		inhrelid = ((Form_pg_inherits) GETSTRUCT(inheritsTuple))->inhrelid;
-		if (numoids >= maxoids)
+		Form_pg_inherits form = (Form_pg_inherits) GETSTRUCT(inheritsTuple);
+
+		if (numchildren >= maxchildren)
 		{
-			maxoids *= 2;
-			oidarr = (Oid *) repalloc(oidarr, maxoids * sizeof(Oid));
+			maxchildren *= 2;
+			inhchildren = (InhChildInfo *) repalloc(inhchildren,
+										maxchildren * sizeof(InhChildInfo));
 		}
-		oidarr[numoids++] = inhrelid;
+		inhchildren[numchildren].relid = form->inhrelid;
+		inhchildren[numchildren].is_partitioned = form->inhpartitioned;
+
+		if (form->inhpartitioned)
+			my_num_partitioned_children++;
+		numchildren++;
 	}
 
 	systable_endscan(scan);
 
 	heap_close(relation, AccessShareLock);
 
+	if (num_partitioned_children)
+		*num_partitioned_children = my_num_partitioned_children;
+
 	/*
 	 * If we found more than one child, sort them by OID.  This ensures
 	 * reasonably consistent behavior regardless of the vagaries of an
 	 * indexscan.  This is important since we need to be sure all backends
 	 * lock children in the same order to avoid needless deadlocks.
 	 */
-	if (numoids > 1)
-		qsort(oidarr, numoids, sizeof(Oid), oid_cmp);
+	if (numchildren > 1)
+		qsort(inhchildren, numchildren, sizeof(InhChildInfo),
+			  inhchildinfo_cmp);
 
 	/*
 	 * Acquire locks and build the result list.
 	 */
-	for (i = 0; i < numoids; i++)
+	for (i = 0; i < numchildren; i++)
 	{
-		inhrelid = oidarr[i];
+		inhrelid = inhchildren[i].relid;
 
 		if (lockmode != NoLock)
 		{
@@ -144,7 +184,7 @@ find_inheritance_children(Oid parentrelId, LOCKMODE lockmode)
 		list = lappend_oid(list, inhrelid);
 	}
 
-	pfree(oidarr);
+	pfree(inhchildren);
 
 	return list;
 }
@@ -159,19 +199,30 @@ find_inheritance_children(Oid parentrelId, LOCKMODE lockmode)
  *		given rel.
  *
  * The specified lock type is acquired on all child relations (but not on the
- * given rel; caller should already have locked it).  If lockmode is NoLock
- * then no locks are acquired, but caller must beware of race conditions
- * against possible DROPs of child relations.
+ * given rel; caller should already have locked it), unless
+ * lock_only_partitioned_children is specified, in which case, only the
+ * child relations that are partitioned tables are locked.  If lockmode is
+ * NoLock then no locks are acquired, but caller must beware of race
+ * conditions against possible DROPs of child relations.
+ *
+ * Returned list of OIDs is such that all the partitioned tables in the tree
+ * appear at the head of the list.  If num_partitioned_children is non-NULL,
+ * *num_partitioned_children returns the number of partitioned child table
+ * OIDs at the head of the list.
  */
 List *
-find_all_inheritors(Oid parentrelId, LOCKMODE lockmode, List **numparents)
+find_all_inheritors(Oid parentrelId, LOCKMODE lockmode,
+					List **numparents, int *num_partitioned_children)
 {
 	/* hash table for O(1) rel_oid -> rel_numparents cell lookup */
 	HTAB	   *seen_rels;
 	HASHCTL		ctl;
 	List	   *rels_list,
-			   *rel_numparents;
+			   *rel_numparents,
+			   *partitioned_rels_list,
+			   *other_rels_list;
 	ListCell   *l;
+	int			my_num_partitioned_children;
 
 	memset(&ctl, 0, sizeof(ctl));
 	ctl.keysize = sizeof(Oid);
@@ -185,31 +236,69 @@ find_all_inheritors(Oid parentrelId, LOCKMODE lockmode, List **numparents)
 
 	/*
 	 * We build a list starting with the given rel and adding all direct and
-	 * indirect children.  We can use a single list as both the record of
-	 * already-found rels and the agenda of rels yet to be scanned for more
-	 * children.  This is a bit tricky but works because the foreach() macro
-	 * doesn't fetch the next list element until the bottom of the loop.
+	 * indirect children.  We can use a single list (rels_list) as both the
+	 * record of already-found rels and the agenda of rels yet to be scanned
+	 * for more children.  This is a bit tricky but works because the foreach()
+	 * macro doesn't fetch the next list element until the bottom of the loop.
+	 *
+	 * partitioned_child_rels will contain the OIDs of the partitioned child
+	 * tables and other_rels_list will contain the OIDs of the non-partitioned
+	 * child tables.  Result list will be generated by concatening the two
+	 * lists together with partitioned_child_rels appearing first.
 	 */
 	rels_list = list_make1_oid(parentrelId);
+	partitioned_rels_list = list_make1_oid(parentrelId);
+	other_rels_list = NIL;
 	rel_numparents = list_make1_int(0);
 
+	my_num_partitioned_children = 0;
+
 	foreach(l, rels_list)
 	{
 		Oid			currentrel = lfirst_oid(l);
 		List	   *currentchildren;
-		ListCell   *lc;
+		ListCell   *lc,
+				   *first_nonpartitioned_child;
+		int			cur_num_partitioned_children = 0,
+					i;
 
 		/* Get the direct children of this rel */
-		currentchildren = find_inheritance_children(currentrel, lockmode);
+		currentchildren = find_inheritance_children(currentrel, lockmode,
+											&cur_num_partitioned_children);
+
+		my_num_partitioned_children += cur_num_partitioned_children;
+
+		/*
+		 * Append partitioned children to rels_list and partitioned_rels_list.
+		 * We know for sure that partitioned children don't need the
+		 * the de-duplication logic in the following loop, because partitioned
+		 * tables are not allowed to partiticipate in multiple inheritance.
+		 */
+		i = 0;
+		foreach(lc, currentchildren)
+		{
+			if (i < cur_num_partitioned_children)
+			{
+				Oid		child_oid = lfirst_oid(lc);
+
+				rels_list = lappend_oid(rels_list, child_oid);
+				partitioned_rels_list = lappend_oid(partitioned_rels_list,
+													child_oid);
+			}
+			else
+				break;
+			i++;
+		}
+		first_nonpartitioned_child = lc;
 
 		/*
 		 * Add to the queue only those children not already seen. This avoids
 		 * making duplicate entries in case of multiple inheritance paths from
 		 * the same parent.  (It'll also keep us from getting into an infinite
 		 * loop, though theoretically there can't be any cycles in the
-		 * inheritance graph anyway.)
+		 * inheritance graph anyway.)  Also, add them to the other_rels_list.
 		 */
-		foreach(lc, currentchildren)
+		for_each_cell(lc, first_nonpartitioned_child)
 		{
 			Oid			child_oid = lfirst_oid(lc);
 			bool		found;
@@ -225,6 +314,7 @@ find_all_inheritors(Oid parentrelId, LOCKMODE lockmode, List **numparents)
 			{
 				/* if it's not there, add it. expect 1 parent, initially. */
 				rels_list = lappend_oid(rels_list, child_oid);
+				other_rels_list = lappend_oid(other_rels_list, child_oid);
 				rel_numparents = lappend_int(rel_numparents, 1);
 				hash_entry->numparents_cell = rel_numparents->tail;
 			}
@@ -237,8 +327,13 @@ find_all_inheritors(Oid parentrelId, LOCKMODE lockmode, List **numparents)
 		list_free(rel_numparents);
 
 	hash_destroy(seen_rels);
+	list_free(rels_list);
+
+	if (num_partitioned_children)
+		*num_partitioned_children = my_num_partitioned_children;
 
-	return rels_list;
+	/* List partitioned child tables before non-partitioned ones. */
+	return list_concat(partitioned_rels_list, other_rels_list);
 }
 
 
diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c
index fbad13ea94..10cc2b8314 100644
--- a/src/backend/commands/analyze.c
+++ b/src/backend/commands/analyze.c
@@ -1282,7 +1282,8 @@ acquire_inherited_sample_rows(Relation onerel, int elevel,
 	 * the children.
 	 */
 	tableOIDs =
-		find_all_inheritors(RelationGetRelid(onerel), AccessShareLock, NULL);
+		find_all_inheritors(RelationGetRelid(onerel), AccessShareLock, NULL,
+							NULL);
 
 	/*
 	 * Check that there's at least one descendant, else fail.  This could
diff --git a/src/backend/commands/lockcmds.c b/src/backend/commands/lockcmds.c
index 9fe9e022b0..529f244f7e 100644
--- a/src/backend/commands/lockcmds.c
+++ b/src/backend/commands/lockcmds.c
@@ -112,7 +112,7 @@ LockTableRecurse(Oid reloid, LOCKMODE lockmode, bool nowait)
 	List	   *children;
 	ListCell   *lc;
 
-	children = find_inheritance_children(reloid, NoLock);
+	children = find_inheritance_children(reloid, NoLock, NULL);
 
 	foreach(lc, children)
 	{
diff --git a/src/backend/commands/publicationcmds.c b/src/backend/commands/publicationcmds.c
index 610cb499d2..64179ea3ef 100644
--- a/src/backend/commands/publicationcmds.c
+++ b/src/backend/commands/publicationcmds.c
@@ -516,7 +516,7 @@ OpenTableList(List *tables)
 			List	   *children;
 
 			children = find_all_inheritors(myrelid, ShareUpdateExclusiveLock,
-										   NULL);
+										   NULL, NULL);
 
 			foreach(child, children)
 			{
diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c
index 0f08245a67..4d686a6f71 100644
--- a/src/backend/commands/tablecmds.c
+++ b/src/backend/commands/tablecmds.c
@@ -299,10 +299,10 @@ static bool MergeCheckConstraint(List *constraints, char *name, Node *expr);
 static void MergeAttributesIntoExisting(Relation child_rel, Relation parent_rel);
 static void MergeConstraintsIntoExisting(Relation child_rel, Relation parent_rel);
 static void StoreCatalogInheritance(Oid relationId, List *supers,
-						bool child_is_partition);
+						bool child_is_partition, bool child_is_partitioned);
 static void StoreCatalogInheritance1(Oid relationId, Oid parentOid,
 						 int16 seqNumber, Relation inhRelation,
-						 bool child_is_partition);
+						 bool child_is_partition, bool child_is_partitioned);
 static int	findAttrByName(const char *attributeName, List *schema);
 static void AlterIndexNamespaces(Relation classRel, Relation rel,
 					 Oid oldNspOid, Oid newNspOid, ObjectAddresses *objsMoved);
@@ -753,7 +753,8 @@ DefineRelation(CreateStmt *stmt, char relkind, Oid ownerId,
 										  typaddress);
 
 	/* Store inheritance information for new rel. */
-	StoreCatalogInheritance(relationId, inheritOids, stmt->partbound != NULL);
+	StoreCatalogInheritance(relationId, inheritOids, stmt->partbound != NULL,
+							relkind == RELKIND_PARTITIONED_TABLE);
 
 	/*
 	 * We must bump the command counter to make the newly-created relation
@@ -1238,7 +1239,8 @@ ExecuteTruncate(TruncateStmt *stmt)
 			ListCell   *child;
 			List	   *children;
 
-			children = find_all_inheritors(myrelid, AccessExclusiveLock, NULL);
+			children = find_all_inheritors(myrelid, AccessExclusiveLock, NULL,
+										   NULL);
 
 			foreach(child, children)
 			{
@@ -2305,7 +2307,7 @@ MergeCheckConstraint(List *constraints, char *name, Node *expr)
  */
 static void
 StoreCatalogInheritance(Oid relationId, List *supers,
-						bool child_is_partition)
+						bool child_is_partition, bool child_is_partitioned)
 {
 	Relation	relation;
 	int16		seqNumber;
@@ -2336,7 +2338,7 @@ StoreCatalogInheritance(Oid relationId, List *supers,
 		Oid			parentOid = lfirst_oid(entry);
 
 		StoreCatalogInheritance1(relationId, parentOid, seqNumber, relation,
-								 child_is_partition);
+								 child_is_partition, child_is_partitioned);
 		seqNumber++;
 	}
 
@@ -2350,7 +2352,7 @@ StoreCatalogInheritance(Oid relationId, List *supers,
 static void
 StoreCatalogInheritance1(Oid relationId, Oid parentOid,
 						 int16 seqNumber, Relation inhRelation,
-						 bool child_is_partition)
+						 bool child_is_partition, bool child_is_partitioned)
 {
 	TupleDesc	desc = RelationGetDescr(inhRelation);
 	Datum		values[Natts_pg_inherits];
@@ -2365,6 +2367,8 @@ StoreCatalogInheritance1(Oid relationId, Oid parentOid,
 	values[Anum_pg_inherits_inhrelid - 1] = ObjectIdGetDatum(relationId);
 	values[Anum_pg_inherits_inhparent - 1] = ObjectIdGetDatum(parentOid);
 	values[Anum_pg_inherits_inhseqno - 1] = Int16GetDatum(seqNumber);
+	values[Anum_pg_inherits_inhpartitioned - 1] =
+									BoolGetDatum(child_is_partitioned);
 
 	memset(nulls, 0, sizeof(nulls));
 
@@ -2564,7 +2568,7 @@ renameatt_internal(Oid myrelid,
 		 * outside the inheritance hierarchy being processed.
 		 */
 		child_oids = find_all_inheritors(myrelid, AccessExclusiveLock,
-										 &child_numparents);
+										 &child_numparents, NULL);
 
 		/*
 		 * find_all_inheritors does the recursive search of the inheritance
@@ -2591,7 +2595,7 @@ renameatt_internal(Oid myrelid,
 		 * expected_parents will only be 0 if we are not already recursing.
 		 */
 		if (expected_parents == 0 &&
-			find_inheritance_children(myrelid, NoLock) != NIL)
+			find_inheritance_children(myrelid, NoLock, NULL) != NIL)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 					 errmsg("inherited column \"%s\" must be renamed in child tables too",
@@ -2774,7 +2778,7 @@ rename_constraint_internal(Oid myrelid,
 					   *li;
 
 			child_oids = find_all_inheritors(myrelid, AccessExclusiveLock,
-											 &child_numparents);
+											 &child_numparents, NULL);
 
 			forboth(lo, child_oids, li, child_numparents)
 			{
@@ -2790,7 +2794,7 @@ rename_constraint_internal(Oid myrelid,
 		else
 		{
 			if (expected_parents == 0 &&
-				find_inheritance_children(myrelid, NoLock) != NIL)
+				find_inheritance_children(myrelid, NoLock, NULL) != NIL)
 				ereport(ERROR,
 						(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 						 errmsg("inherited constraint \"%s\" must be renamed in child tables too",
@@ -4803,7 +4807,7 @@ ATSimpleRecursion(List **wqueue, Relation rel,
 		ListCell   *child;
 		List	   *children;
 
-		children = find_all_inheritors(relid, lockmode, NULL);
+		children = find_all_inheritors(relid, lockmode, NULL, NULL);
 
 		/*
 		 * find_all_inheritors does the recursive search of the inheritance
@@ -5212,7 +5216,7 @@ ATExecAddColumn(List **wqueue, AlteredTableInfo *tab, Relation rel,
 	 */
 	if (colDef->identity &&
 		recurse &&
-		find_inheritance_children(myrelid, NoLock) != NIL)
+		find_inheritance_children(myrelid, NoLock, NULL) != NIL)
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 				 errmsg("cannot recursively add identity column to table that has child tables")));
@@ -5418,7 +5422,8 @@ ATExecAddColumn(List **wqueue, AlteredTableInfo *tab, Relation rel,
 	 * routines, we have to do this one level of recursion at a time; we can't
 	 * use find_all_inheritors to do it in one pass.
 	 */
-	children = find_inheritance_children(RelationGetRelid(rel), lockmode);
+	children = find_inheritance_children(RelationGetRelid(rel), lockmode,
+										 NULL);
 
 	/*
 	 * If we are told not to recurse, there had better not be any child
@@ -6537,7 +6542,8 @@ ATExecDropColumn(List **wqueue, Relation rel, const char *colName,
 	 * routines, we have to do this one level of recursion at a time; we can't
 	 * use find_all_inheritors to do it in one pass.
 	 */
-	children = find_inheritance_children(RelationGetRelid(rel), lockmode);
+	children = find_inheritance_children(RelationGetRelid(rel), lockmode,
+										 NULL);
 
 	if (children)
 	{
@@ -6971,7 +6977,8 @@ ATAddCheckConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel,
 	 * routines, we have to do this one level of recursion at a time; we can't
 	 * use find_all_inheritors to do it in one pass.
 	 */
-	children = find_inheritance_children(RelationGetRelid(rel), lockmode);
+	children = find_inheritance_children(RelationGetRelid(rel), lockmode,
+										 NULL);
 
 	/*
 	 * Check if ONLY was specified with ALTER TABLE.  If so, allow the
@@ -7692,7 +7699,7 @@ ATExecValidateConstraint(Relation rel, char *constrName, bool recurse,
 			 */
 			if (!recursing && !con->connoinherit)
 				children = find_all_inheritors(RelationGetRelid(rel),
-											   lockmode, NULL);
+											   lockmode, NULL, NULL);
 
 			/*
 			 * For CHECK constraints, we must ensure that we only mark the
@@ -8575,7 +8582,8 @@ ATExecDropConstraint(Relation rel, const char *constrName,
 	 * use find_all_inheritors to do it in one pass.
 	 */
 	if (!is_no_inherit_constraint)
-		children = find_inheritance_children(RelationGetRelid(rel), lockmode);
+		children = find_inheritance_children(RelationGetRelid(rel), lockmode,
+											 NULL);
 	else
 		children = NIL;
 
@@ -8864,7 +8872,7 @@ ATPrepAlterColumnType(List **wqueue,
 		ListCell   *child;
 		List	   *children;
 
-		children = find_all_inheritors(relid, lockmode, NULL);
+		children = find_all_inheritors(relid, lockmode, NULL, NULL);
 
 		/*
 		 * find_all_inheritors does the recursive search of the inheritance
@@ -8915,7 +8923,8 @@ ATPrepAlterColumnType(List **wqueue,
 		}
 	}
 	else if (!recursing &&
-			 find_inheritance_children(RelationGetRelid(rel), NoLock) != NIL)
+			 find_inheritance_children(RelationGetRelid(rel),
+									   NoLock, NULL) != NIL)
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 				 errmsg("type of inherited column \"%s\" must be changed in child tables too",
@@ -11027,7 +11036,7 @@ ATExecAddInherit(Relation child_rel, RangeVar *parent, LOCKMODE lockmode)
 	 * We use weakest lock we can on child's children, namely AccessShareLock.
 	 */
 	children = find_all_inheritors(RelationGetRelid(child_rel),
-								   AccessShareLock, NULL);
+								   AccessShareLock, NULL, NULL);
 
 	if (list_member_oid(children, RelationGetRelid(parent_rel)))
 		ereport(ERROR,
@@ -11136,6 +11145,8 @@ CreateInheritance(Relation child_rel, Relation parent_rel)
 							 inhseqno + 1,
 							 catalogRelation,
 							 parent_rel->rd_rel->relkind ==
+							 RELKIND_PARTITIONED_TABLE,
+							 child_rel->rd_rel->relkind ==
 							 RELKIND_PARTITIONED_TABLE);
 
 	/* Now we're done with pg_inherits */
@@ -13696,7 +13707,8 @@ ATExecAttachPartition(List **wqueue, Relation rel, PartitionCmd *cmd)
 	 * weaker lock now and the stronger one only when needed.
 	 */
 	attachrel_children = find_all_inheritors(RelationGetRelid(attachrel),
-											 AccessExclusiveLock, NULL);
+											 AccessExclusiveLock, NULL,
+											 NULL);
 	if (list_member_oid(attachrel_children, RelationGetRelid(rel)))
 		ereport(ERROR,
 				(errcode(ERRCODE_DUPLICATE_TABLE),
diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c
index faa181207a..e2e5ffce42 100644
--- a/src/backend/commands/vacuum.c
+++ b/src/backend/commands/vacuum.c
@@ -430,7 +430,8 @@ get_rel_oids(Oid relid, const RangeVar *vacrel)
 		oldcontext = MemoryContextSwitchTo(vac_context);
 		if (include_parts)
 			oid_list = list_concat(oid_list,
-								   find_all_inheritors(relid, NoLock, NULL));
+								   find_all_inheritors(relid, NoLock, NULL,
+													   NULL));
 		else
 			oid_list = lappend_oid(oid_list, relid);
 		MemoryContextSwitchTo(oldcontext);
diff --git a/src/backend/executor/execMain.c b/src/backend/executor/execMain.c
index 23ed2c55b9..b63abba1e4 100644
--- a/src/backend/executor/execMain.c
+++ b/src/backend/executor/execMain.c
@@ -3278,7 +3278,8 @@ ExecSetupPartitionTupleRouting(Relation rel,
 	 * Get the information about the partition tree after locking all the
 	 * partitions.
 	 */
-	(void) find_all_inheritors(RelationGetRelid(rel), RowExclusiveLock, NULL);
+	(void) find_all_inheritors(RelationGetRelid(rel), RowExclusiveLock, NULL,
+							   NULL);
 	pds = RelationGetPartitionDispatchInfo(rel, num_parted, &leaf_parts);
 
 	/*
diff --git a/src/backend/optimizer/prep/prepunion.c b/src/backend/optimizer/prep/prepunion.c
index 2202ad9941..81865cad7d 100644
--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -1425,7 +1425,7 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 		lockmode = AccessShareLock;
 
 	/* Scan for all members of inheritance set, acquire needed locks */
-	inhOIDs = find_all_inheritors(parentOID, lockmode, NULL);
+	inhOIDs = find_all_inheritors(parentOID, lockmode, NULL, NULL);
 
 	/*
 	 * Check that there's at least one descendant, else treat as no-child
diff --git a/src/include/catalog/pg_inherits.h b/src/include/catalog/pg_inherits.h
index 26bfab5db6..9f59c017e7 100644
--- a/src/include/catalog/pg_inherits.h
+++ b/src/include/catalog/pg_inherits.h
@@ -30,9 +30,20 @@
 
 CATALOG(pg_inherits,2611) BKI_WITHOUT_OIDS
 {
+	/* OID of the child table. */
 	Oid			inhrelid;
+
+	/* OID of the parent table. */
 	Oid			inhparent;
+
+	/*
+	 * Sequence number (starting with 1) of this parent, if this child table
+	 * has multiple parents.
+	 */
 	int32		inhseqno;
+
+	/* true if the child is a partitioned table, false otherwise. */
+	bool		inhpartitioned;
 } FormData_pg_inherits;
 
 /* ----------------
@@ -46,10 +57,11 @@ typedef FormData_pg_inherits *Form_pg_inherits;
  *		compiler constants for pg_inherits
  * ----------------
  */
-#define Natts_pg_inherits				3
-#define Anum_pg_inherits_inhrelid		1
-#define Anum_pg_inherits_inhparent		2
-#define Anum_pg_inherits_inhseqno		3
+#define Natts_pg_inherits					4
+#define Anum_pg_inherits_inhrelid			1
+#define Anum_pg_inherits_inhparent			2
+#define Anum_pg_inherits_inhseqno			3
+#define Anum_pg_inherits_inhpartitioned		4
 
 /* ----------------
  *		pg_inherits has no initial contents
diff --git a/src/include/catalog/pg_inherits_fn.h b/src/include/catalog/pg_inherits_fn.h
index 7743388899..8f371acae7 100644
--- a/src/include/catalog/pg_inherits_fn.h
+++ b/src/include/catalog/pg_inherits_fn.h
@@ -17,9 +17,10 @@
 #include "nodes/pg_list.h"
 #include "storage/lock.h"
 
-extern List *find_inheritance_children(Oid parentrelId, LOCKMODE lockmode);
+extern List *find_inheritance_children(Oid parentrelId, LOCKMODE lockmode,
+						  int *num_partitioned_children);
 extern List *find_all_inheritors(Oid parentrelId, LOCKMODE lockmode,
-					List **parents);
+					List **parents, int *num_partitioned_children);
 extern bool has_subclass(Oid relationId);
 extern bool has_superclass(Oid relationId);
 extern bool typeInheritsFrom(Oid subclassTypeId, Oid superclassTypeId);
-- 
2.11.0

From b92ef9ec07841942ceedd173dd81e3286419202f Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Thu, 10 Aug 2017 17:59:18 +0900
Subject: [PATCH 2/6] Allow locking only partitioned children in partition tree

find_inheritance_childrem will still return the OIDs of the
non-partitioned children, but does not lock them if the caller asks
it so.

None of the callers pass 'true' yet though.
---
 contrib/sepgsql/dml.c                  |  3 ++-
 src/backend/catalog/partition.c        |  3 ++-
 src/backend/catalog/pg_inherits.c      | 20 ++++++++++++++++----
 src/backend/commands/analyze.c         |  4 ++--
 src/backend/commands/lockcmds.c        |  2 +-
 src/backend/commands/publicationcmds.c |  2 +-
 src/backend/commands/tablecmds.c       | 34 +++++++++++++++++-----------------
 src/backend/commands/vacuum.c          |  4 ++--
 src/backend/executor/execMain.c        |  4 ++--
 src/backend/optimizer/prep/prepunion.c |  2 +-
 src/include/catalog/pg_inherits_fn.h   |  2 ++
 11 files changed, 48 insertions(+), 32 deletions(-)

diff --git a/contrib/sepgsql/dml.c b/contrib/sepgsql/dml.c
index 6fc279805c..91f338f8bf 100644
--- a/contrib/sepgsql/dml.c
+++ b/contrib/sepgsql/dml.c
@@ -333,7 +333,8 @@ sepgsql_dml_privileges(List *rangeTabls, bool abort_on_violation)
 		if (!rte->inh)
 			tableIds = list_make1_oid(rte->relid);
 		else
-			tableIds = find_all_inheritors(rte->relid, NoLock, NULL, NULL);
+			tableIds = find_all_inheritors(rte->relid, NoLock, false,
+										   NULL, NULL);
 
 		foreach(li, tableIds)
 		{
diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index fe8e60de14..9645381fcb 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -178,7 +178,8 @@ RelationBuildPartitionDesc(Relation rel)
 		return;
 
 	/* Get partition oids from pg_inherits */
-	inhoids = find_inheritance_children(RelationGetRelid(rel), NoLock, NULL);
+	inhoids = find_inheritance_children(RelationGetRelid(rel), NoLock, false,
+										NULL);
 
 	/* Collect bound spec nodes in a list */
 	i = 0;
diff --git a/src/backend/catalog/pg_inherits.c b/src/backend/catalog/pg_inherits.c
index 5292ec8058..72420f65f1 100644
--- a/src/backend/catalog/pg_inherits.c
+++ b/src/backend/catalog/pg_inherits.c
@@ -74,13 +74,16 @@ inhchildinfo_cmp(const void *p1, const void *p2)
  * Returns a list containing the OIDs of all relations which
  * inherit *directly* from the relation with OID 'parentrelId'.
  *
- * The specified lock type is acquired on each child relation (but not on the
- * given rel; caller should already have locked it).  If lockmode is NoLock
- * then no locks are acquired, but caller must beware of race conditions
- * against possible DROPs of child relations.
+ * The specified lock type is acquired on each child relation, (but not on the
+ * given rel; caller should already have locked it), unless
+ * lock_only_partitioned_children is specified in which case only partitioned
+ * children are locked.  If lockmode is NoLock then no locks are acquired, but
+ * caller must beware of race conditions against possible DROPs of child
+ * relations.
  */
 List *
 find_inheritance_children(Oid parentrelId, LOCKMODE lockmode,
+						  bool lock_only_partitioned_children,
 						  int *num_partitioned_children)
 {
 	List	   *list = NIL;
@@ -162,6 +165,13 @@ find_inheritance_children(Oid parentrelId, LOCKMODE lockmode,
 	{
 		inhrelid = inhchildren[i].relid;
 
+		/* If requested, skip locking non-partitioned children. */
+		if (lock_only_partitioned_children && i >= *num_partitioned_children)
+		{
+			list = lappend_oid(list, inhrelid);
+			continue;
+		}
+
 		if (lockmode != NoLock)
 		{
 			/* Get the lock to synchronize against concurrent drop */
@@ -212,6 +222,7 @@ find_inheritance_children(Oid parentrelId, LOCKMODE lockmode,
  */
 List *
 find_all_inheritors(Oid parentrelId, LOCKMODE lockmode,
+					bool lock_only_partitioned_children,
 					List **numparents, int *num_partitioned_children)
 {
 	/* hash table for O(1) rel_oid -> rel_numparents cell lookup */
@@ -264,6 +275,7 @@ find_all_inheritors(Oid parentrelId, LOCKMODE lockmode,
 
 		/* Get the direct children of this rel */
 		currentchildren = find_inheritance_children(currentrel, lockmode,
+											lock_only_partitioned_children,
 											&cur_num_partitioned_children);
 
 		my_num_partitioned_children += cur_num_partitioned_children;
diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c
index 10cc2b8314..4bd374632f 100644
--- a/src/backend/commands/analyze.c
+++ b/src/backend/commands/analyze.c
@@ -1282,8 +1282,8 @@ acquire_inherited_sample_rows(Relation onerel, int elevel,
 	 * the children.
 	 */
 	tableOIDs =
-		find_all_inheritors(RelationGetRelid(onerel), AccessShareLock, NULL,
-							NULL);
+		find_all_inheritors(RelationGetRelid(onerel), AccessShareLock, false,
+							NULL, NULL);
 
 	/*
 	 * Check that there's at least one descendant, else fail.  This could
diff --git a/src/backend/commands/lockcmds.c b/src/backend/commands/lockcmds.c
index 529f244f7e..771aa11b1c 100644
--- a/src/backend/commands/lockcmds.c
+++ b/src/backend/commands/lockcmds.c
@@ -112,7 +112,7 @@ LockTableRecurse(Oid reloid, LOCKMODE lockmode, bool nowait)
 	List	   *children;
 	ListCell   *lc;
 
-	children = find_inheritance_children(reloid, NoLock, NULL);
+	children = find_inheritance_children(reloid, NoLock, false, NULL);
 
 	foreach(lc, children)
 	{
diff --git a/src/backend/commands/publicationcmds.c b/src/backend/commands/publicationcmds.c
index 64179ea3ef..4315028c66 100644
--- a/src/backend/commands/publicationcmds.c
+++ b/src/backend/commands/publicationcmds.c
@@ -516,7 +516,7 @@ OpenTableList(List *tables)
 			List	   *children;
 
 			children = find_all_inheritors(myrelid, ShareUpdateExclusiveLock,
-										   NULL, NULL);
+										   false, NULL, NULL);
 
 			foreach(child, children)
 			{
diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c
index 4d686a6f71..ef3869854a 100644
--- a/src/backend/commands/tablecmds.c
+++ b/src/backend/commands/tablecmds.c
@@ -1239,8 +1239,8 @@ ExecuteTruncate(TruncateStmt *stmt)
 			ListCell   *child;
 			List	   *children;
 
-			children = find_all_inheritors(myrelid, AccessExclusiveLock, NULL,
-										   NULL);
+			children = find_all_inheritors(myrelid, AccessExclusiveLock, false,
+										   NULL, NULL);
 
 			foreach(child, children)
 			{
@@ -2567,7 +2567,7 @@ renameatt_internal(Oid myrelid,
 		 * calls to renameatt() can determine whether there are any parents
 		 * outside the inheritance hierarchy being processed.
 		 */
-		child_oids = find_all_inheritors(myrelid, AccessExclusiveLock,
+		child_oids = find_all_inheritors(myrelid, AccessExclusiveLock, false,
 										 &child_numparents, NULL);
 
 		/*
@@ -2595,7 +2595,7 @@ renameatt_internal(Oid myrelid,
 		 * expected_parents will only be 0 if we are not already recursing.
 		 */
 		if (expected_parents == 0 &&
-			find_inheritance_children(myrelid, NoLock, NULL) != NIL)
+			find_inheritance_children(myrelid, NoLock, false, NULL) != NIL)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 					 errmsg("inherited column \"%s\" must be renamed in child tables too",
@@ -2778,7 +2778,7 @@ rename_constraint_internal(Oid myrelid,
 					   *li;
 
 			child_oids = find_all_inheritors(myrelid, AccessExclusiveLock,
-											 &child_numparents, NULL);
+											 false, &child_numparents, NULL);
 
 			forboth(lo, child_oids, li, child_numparents)
 			{
@@ -2794,7 +2794,7 @@ rename_constraint_internal(Oid myrelid,
 		else
 		{
 			if (expected_parents == 0 &&
-				find_inheritance_children(myrelid, NoLock, NULL) != NIL)
+				find_inheritance_children(myrelid, NoLock, false, NULL) != NIL)
 				ereport(ERROR,
 						(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 						 errmsg("inherited constraint \"%s\" must be renamed in child tables too",
@@ -4807,7 +4807,7 @@ ATSimpleRecursion(List **wqueue, Relation rel,
 		ListCell   *child;
 		List	   *children;
 
-		children = find_all_inheritors(relid, lockmode, NULL, NULL);
+		children = find_all_inheritors(relid, lockmode, false, NULL, NULL);
 
 		/*
 		 * find_all_inheritors does the recursive search of the inheritance
@@ -5216,7 +5216,7 @@ ATExecAddColumn(List **wqueue, AlteredTableInfo *tab, Relation rel,
 	 */
 	if (colDef->identity &&
 		recurse &&
-		find_inheritance_children(myrelid, NoLock, NULL) != NIL)
+		find_inheritance_children(myrelid, NoLock, false, NULL) != NIL)
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 				 errmsg("cannot recursively add identity column to table that has child tables")));
@@ -5423,7 +5423,7 @@ ATExecAddColumn(List **wqueue, AlteredTableInfo *tab, Relation rel,
 	 * use find_all_inheritors to do it in one pass.
 	 */
 	children = find_inheritance_children(RelationGetRelid(rel), lockmode,
-										 NULL);
+										 false, NULL);
 
 	/*
 	 * If we are told not to recurse, there had better not be any child
@@ -6543,7 +6543,7 @@ ATExecDropColumn(List **wqueue, Relation rel, const char *colName,
 	 * use find_all_inheritors to do it in one pass.
 	 */
 	children = find_inheritance_children(RelationGetRelid(rel), lockmode,
-										 NULL);
+										 false, NULL);
 
 	if (children)
 	{
@@ -6978,7 +6978,7 @@ ATAddCheckConstraint(List **wqueue, AlteredTableInfo *tab, Relation rel,
 	 * use find_all_inheritors to do it in one pass.
 	 */
 	children = find_inheritance_children(RelationGetRelid(rel), lockmode,
-										 NULL);
+										 false, NULL);
 
 	/*
 	 * Check if ONLY was specified with ALTER TABLE.  If so, allow the
@@ -7699,7 +7699,7 @@ ATExecValidateConstraint(Relation rel, char *constrName, bool recurse,
 			 */
 			if (!recursing && !con->connoinherit)
 				children = find_all_inheritors(RelationGetRelid(rel),
-											   lockmode, NULL, NULL);
+											   lockmode, false, NULL, NULL);
 
 			/*
 			 * For CHECK constraints, we must ensure that we only mark the
@@ -8583,7 +8583,7 @@ ATExecDropConstraint(Relation rel, const char *constrName,
 	 */
 	if (!is_no_inherit_constraint)
 		children = find_inheritance_children(RelationGetRelid(rel), lockmode,
-											 NULL);
+											 false, NULL);
 	else
 		children = NIL;
 
@@ -8872,7 +8872,7 @@ ATPrepAlterColumnType(List **wqueue,
 		ListCell   *child;
 		List	   *children;
 
-		children = find_all_inheritors(relid, lockmode, NULL, NULL);
+		children = find_all_inheritors(relid, lockmode, false, NULL, NULL);
 
 		/*
 		 * find_all_inheritors does the recursive search of the inheritance
@@ -8924,7 +8924,7 @@ ATPrepAlterColumnType(List **wqueue,
 	}
 	else if (!recursing &&
 			 find_inheritance_children(RelationGetRelid(rel),
-									   NoLock, NULL) != NIL)
+									   NoLock, false, NULL) != NIL)
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
 				 errmsg("type of inherited column \"%s\" must be changed in child tables too",
@@ -11036,7 +11036,7 @@ ATExecAddInherit(Relation child_rel, RangeVar *parent, LOCKMODE lockmode)
 	 * We use weakest lock we can on child's children, namely AccessShareLock.
 	 */
 	children = find_all_inheritors(RelationGetRelid(child_rel),
-								   AccessShareLock, NULL, NULL);
+								   AccessShareLock, false, NULL, NULL);
 
 	if (list_member_oid(children, RelationGetRelid(parent_rel)))
 		ereport(ERROR,
@@ -13707,7 +13707,7 @@ ATExecAttachPartition(List **wqueue, Relation rel, PartitionCmd *cmd)
 	 * weaker lock now and the stronger one only when needed.
 	 */
 	attachrel_children = find_all_inheritors(RelationGetRelid(attachrel),
-											 AccessExclusiveLock, NULL,
+											 AccessExclusiveLock, false, NULL,
 											 NULL);
 	if (list_member_oid(attachrel_children, RelationGetRelid(rel)))
 		ereport(ERROR,
diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c
index e2e5ffce42..70cd5721f3 100644
--- a/src/backend/commands/vacuum.c
+++ b/src/backend/commands/vacuum.c
@@ -430,8 +430,8 @@ get_rel_oids(Oid relid, const RangeVar *vacrel)
 		oldcontext = MemoryContextSwitchTo(vac_context);
 		if (include_parts)
 			oid_list = list_concat(oid_list,
-								   find_all_inheritors(relid, NoLock, NULL,
-													   NULL));
+								   find_all_inheritors(relid, NoLock, false,
+													   NULL, NULL));
 		else
 			oid_list = lappend_oid(oid_list, relid);
 		MemoryContextSwitchTo(oldcontext);
diff --git a/src/backend/executor/execMain.c b/src/backend/executor/execMain.c
index b63abba1e4..9ee0e03a3c 100644
--- a/src/backend/executor/execMain.c
+++ b/src/backend/executor/execMain.c
@@ -3278,8 +3278,8 @@ ExecSetupPartitionTupleRouting(Relation rel,
 	 * Get the information about the partition tree after locking all the
 	 * partitions.
 	 */
-	(void) find_all_inheritors(RelationGetRelid(rel), RowExclusiveLock, NULL,
-							   NULL);
+	(void) find_all_inheritors(RelationGetRelid(rel), RowExclusiveLock, false,
+							   NULL, NULL);
 	pds = RelationGetPartitionDispatchInfo(rel, num_parted, &leaf_parts);
 
 	/*
diff --git a/src/backend/optimizer/prep/prepunion.c b/src/backend/optimizer/prep/prepunion.c
index 81865cad7d..0d20ffa2f7 100644
--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -1425,7 +1425,7 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 		lockmode = AccessShareLock;
 
 	/* Scan for all members of inheritance set, acquire needed locks */
-	inhOIDs = find_all_inheritors(parentOID, lockmode, NULL, NULL);
+	inhOIDs = find_all_inheritors(parentOID, lockmode, false, NULL, NULL);
 
 	/*
 	 * Check that there's at least one descendant, else treat as no-child
diff --git a/src/include/catalog/pg_inherits_fn.h b/src/include/catalog/pg_inherits_fn.h
index 8f371acae7..e568d11e43 100644
--- a/src/include/catalog/pg_inherits_fn.h
+++ b/src/include/catalog/pg_inherits_fn.h
@@ -18,8 +18,10 @@
 #include "storage/lock.h"
 
 extern List *find_inheritance_children(Oid parentrelId, LOCKMODE lockmode,
+						  bool lock_only_partitioned_children,
 						  int *num_partitioned_children);
 extern List *find_all_inheritors(Oid parentrelId, LOCKMODE lockmode,
+					bool lock_only_partitioned_children,
 					List **parents, int *num_partitioned_children);
 extern bool has_subclass(Oid relationId);
 extern bool has_superclass(Oid relationId);
-- 
2.11.0

From 35d5316f88a295576fd2c43d84a8df33e3f48728 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 26 Jul 2017 14:42:47 +0900
Subject: [PATCH 3/6] WIP: Defer opening and locking partitions to
 set_append_rel_size

This will still create RT entries for the child tables in
expand_inherited_rtentry(), though not AppendRelInfos, because
they require locking and opening the relation.  Having all the
RT entries created in advance means that setup_simple_rel_arrays
knows the size of root->simple_rte_array and root->simpl_rel_array
to allocate.

expand_inherited_rtentry also allocates LeafPartitionInfos and
PartitionInfos for individual leaf partitions and partitioned child
table, resp.  All LPOs and POs thus created are stuffed into
PartitionRootInfo that is created for the parent.  PartitionRootInfo
was previously called PartitionedChildRelInfo.

When set_append_rel_size is called for the root parent, the whole
partition tree will be recursively processed, creating a
PartitionAppendInfo in each recursive step, which maps a given
parent table in the partition tree to its immediate partitions (also
only those satisfy the query's WHERE condition).  Once we have
PartitionAppendInfos for all the parents in the tree, we resume
set_append_rel_size() processing, creating RelOptInfos and
AppendRelInfos for the root parent's children and recursively doing
the same for its partitioned children and so on.
---
 src/backend/catalog/partition.c        |  20 ++
 src/backend/nodes/copyfuncs.c          |  17 --
 src/backend/nodes/equalfuncs.c         |  12 -
 src/backend/nodes/outfuncs.c           |  59 ++++-
 src/backend/optimizer/path/allpaths.c  | 389 +++++++++++++++++++++++++++++++--
 src/backend/optimizer/plan/planner.c   | 115 +++++++++-
 src/backend/optimizer/plan/setrefs.c   |  26 +++
 src/backend/optimizer/prep/prepunion.c | 300 +++++++++++++++----------
 src/backend/optimizer/util/plancat.c   |  37 ++++
 src/backend/optimizer/util/relnode.c   |  91 +++++++-
 src/backend/utils/cache/lsyscache.c    |  50 +++++
 src/include/catalog/partition.h        |   4 +
 src/include/nodes/nodes.h              |   5 +-
 src/include/nodes/relation.h           |  95 +++++++-
 src/include/optimizer/plancat.h        |   1 +
 src/include/optimizer/prep.h           |   3 +
 src/include/utils/lsyscache.h          |   2 +
 src/test/regress/expected/insert.out   |   4 +-
 18 files changed, 1029 insertions(+), 201 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 9645381fcb..a193f02551 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1141,6 +1141,26 @@ RelationGetPartitionDispatchInfo(Relation rel,
 	return pd;
 }
 
+/*
+ * get_partitions_for_keys
+ *		Returns the list of indexes (from pd->indexes) of the partitions that
+ *		will need to be scanned for the given scan keys.
+ *
+ * TODO: add the interface to pass the query scan keys and the logic to look
+ * up partitions using those keys.
+ */
+List *
+get_partitions_for_keys(PartitionDispatch pd)
+{
+	int		i;
+	List   *result = NIL;
+
+	for (i = 0; i < pd->partdesc->nparts; i++)
+		result = lappend_int(result, pd->indexes[i]);
+
+	return result;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/nodes/copyfuncs.c b/src/backend/nodes/copyfuncs.c
index f9ddf4ed76..4c888ec3dc 100644
--- a/src/backend/nodes/copyfuncs.c
+++ b/src/backend/nodes/copyfuncs.c
@@ -2251,20 +2251,6 @@ _copyAppendRelInfo(const AppendRelInfo *from)
 }
 
 /*
- * _copyPartitionedChildRelInfo
- */
-static PartitionedChildRelInfo *
-_copyPartitionedChildRelInfo(const PartitionedChildRelInfo *from)
-{
-	PartitionedChildRelInfo *newnode = makeNode(PartitionedChildRelInfo);
-
-	COPY_SCALAR_FIELD(parent_relid);
-	COPY_NODE_FIELD(child_rels);
-
-	return newnode;
-}
-
-/*
  * _copyPlaceHolderInfo
  */
 static PlaceHolderInfo *
@@ -4996,9 +4982,6 @@ copyObjectImpl(const void *from)
 		case T_AppendRelInfo:
 			retval = _copyAppendRelInfo(from);
 			break;
-		case T_PartitionedChildRelInfo:
-			retval = _copyPartitionedChildRelInfo(from);
-			break;
 		case T_PlaceHolderInfo:
 			retval = _copyPlaceHolderInfo(from);
 			break;
diff --git a/src/backend/nodes/equalfuncs.c b/src/backend/nodes/equalfuncs.c
index 8d92c03633..fb248f31f3 100644
--- a/src/backend/nodes/equalfuncs.c
+++ b/src/backend/nodes/equalfuncs.c
@@ -905,15 +905,6 @@ _equalAppendRelInfo(const AppendRelInfo *a, const AppendRelInfo *b)
 }
 
 static bool
-_equalPartitionedChildRelInfo(const PartitionedChildRelInfo *a, const PartitionedChildRelInfo *b)
-{
-	COMPARE_SCALAR_FIELD(parent_relid);
-	COMPARE_NODE_FIELD(child_rels);
-
-	return true;
-}
-
-static bool
 _equalPlaceHolderInfo(const PlaceHolderInfo *a, const PlaceHolderInfo *b)
 {
 	COMPARE_SCALAR_FIELD(phid);
@@ -3155,9 +3146,6 @@ equal(const void *a, const void *b)
 		case T_AppendRelInfo:
 			retval = _equalAppendRelInfo(a, b);
 			break;
-		case T_PartitionedChildRelInfo:
-			retval = _equalPartitionedChildRelInfo(a, b);
-			break;
 		case T_PlaceHolderInfo:
 			retval = _equalPlaceHolderInfo(a, b);
 			break;
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index 9ee3e23761..2480fd6429 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -2213,7 +2213,7 @@ _outPlannerInfo(StringInfo str, const PlannerInfo *node)
 	WRITE_NODE_FIELD(full_join_clauses);
 	WRITE_NODE_FIELD(join_info_list);
 	WRITE_NODE_FIELD(append_rel_list);
-	WRITE_NODE_FIELD(pcinfo_list);
+	WRITE_NODE_FIELD(prinfo_list);
 	WRITE_NODE_FIELD(rowMarks);
 	WRITE_NODE_FIELD(placeholder_list);
 	WRITE_NODE_FIELD(fkey_list);
@@ -2287,6 +2287,12 @@ _outRelOptInfo(StringInfo str, const RelOptInfo *node)
 	WRITE_NODE_FIELD(joininfo);
 	WRITE_BOOL_FIELD(has_eclass_joins);
 	WRITE_BITMAPSET_FIELD(top_parent_relids);
+	WRITE_INT_FIELD(num_parted);
+	/* don't bother printing partition_infos */
+	WRITE_INT_FIELD(num_leaf_parts);
+	/* don't bother printing leaf_part_infos */
+	WRITE_NODE_FIELD(live_partition_painfos);
+	WRITE_UINT_FIELD(root_parent_relid);
 }
 
 static void
@@ -2512,12 +2518,44 @@ _outAppendRelInfo(StringInfo str, const AppendRelInfo *node)
 }
 
 static void
-_outPartitionedChildRelInfo(StringInfo str, const PartitionedChildRelInfo *node)
+_outPartitionInfo(StringInfo str, const PartitionInfo *node)
 {
-	WRITE_NODE_TYPE("PARTITIONEDCHILDRELINFO");
+	WRITE_NODE_TYPE("PARTITIONINFO");
+
+	WRITE_BOOL_FIELD(is_other_temp);
+	WRITE_UINT_FIELD(relid);
+	/* Don't bother writing out the PartitionDispatch object */
+}
+
+static void
+_outLeafPartitionInfo(StringInfo str, const LeafPartitionInfo *node)
+{
+	WRITE_NODE_TYPE("LEAFPARTITIONINFO");
+
+	WRITE_BOOL_FIELD(is_other_temp);
+	WRITE_OID_FIELD(reloid);
+	WRITE_UINT_FIELD(relid);
+}
+
+static void
+_outPartitionAppendInfo(StringInfo str, const PartitionAppendInfo *node)
+{
+	WRITE_NODE_TYPE("PARTITIONAPPENDINFO");
+
+	WRITE_UINT_FIELD(parent_relid);
+	WRITE_NODE_FIELD(live_partition_relids);
+}
+
+static void
+_outPartitionRootInfo(StringInfo str, const PartitionRootInfo *node)
+{
+	WRITE_NODE_TYPE("PARTITIONROOTINFO");
 
 	WRITE_UINT_FIELD(parent_relid);
-	WRITE_NODE_FIELD(child_rels);
+	WRITE_NODE_FIELD(partition_infos);
+	WRITE_NODE_FIELD(partitioned_relids);
+	WRITE_NODE_FIELD(leaf_part_infos);
+	WRITE_NODE_FIELD(orig_leaf_part_oids);
 }
 
 static void
@@ -4045,8 +4083,17 @@ outNode(StringInfo str, const void *obj)
 			case T_AppendRelInfo:
 				_outAppendRelInfo(str, obj);
 				break;
-			case T_PartitionedChildRelInfo:
-				_outPartitionedChildRelInfo(str, obj);
+			case T_PartitionInfo:
+				_outPartitionInfo(str, obj);
+				break;
+			case T_LeafPartitionInfo:
+				_outLeafPartitionInfo(str, obj);
+				break;
+			case T_PartitionAppendInfo:
+				_outPartitionAppendInfo(str, obj);
+				break;
+			case T_PartitionRootInfo:
+				_outPartitionRootInfo(str, obj);
 				break;
 			case T_PlaceHolderInfo:
 				_outPlaceHolderInfo(str, obj);
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 2d7e1d84d0..c5c50e3b9d 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,6 +20,7 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
@@ -43,6 +44,8 @@
 #include "parser/parse_clause.h"
 #include "parser/parsetree.h"
 #include "rewrite/rewriteManip.h"
+#include "storage/lmgr.h"
+#include "utils/builtins.h"
 #include "utils/lsyscache.h"
 
 
@@ -845,6 +848,172 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_partitions_recurse
+ *		Find partitions of the partitioned table described in partinfo,
+ *		recursing for those partitions that are themselves partitioned tables
+ *
+ * rootrel is the root of the partition tree of which this table is a part.
+ * We create a PartitionAppendInfo for this partitioned table and append it to
+ * rootrel->live_partition_painfos.
+ *
+ * List of the leaf partitions of this table will be returned.
+ */
+static List *
+get_rel_partitions_recurse(RelOptInfo *rootrel,
+						   PartitionInfo *partinfo,
+						   PartitionInfo **all_partinfos,
+						   LeafPartitionInfo **leaf_part_infos)
+{
+	PartitionAppendInfo *painfo;
+	List   *indexes;
+	List   *result = NIL,
+		   *my_live_partitions = NIL;
+	ListCell *l;
+
+	/*
+	 * Create a PartitionAppendInfo to map this table to the child tables
+	 * that will be its Append children.
+	 */
+	painfo = makeNode(PartitionAppendInfo);
+	painfo->parent_relid = partinfo->relid;
+
+	/* They will all be under the root table's Append node. */
+	rootrel->live_partition_painfos = lappend(rootrel->live_partition_painfos,
+											  painfo);
+
+	/*
+	 * TODO: collect the keys by looking at the clauses in
+	 * rootrel->baserestrictinfo considering this table's partition keys.
+	 */
+
+	/* Ask partition.c which partitions it thinks match the keys. */
+	indexes = get_partitions_for_keys(partinfo->pd);
+
+	/* Collect leaf partitions in the result list and recurse for others. */
+	foreach(l, indexes)
+	{
+		int		index = lfirst_int(l);
+
+		if (index >= 0)
+		{
+			LeafPartitionInfo *lpinfo = leaf_part_infos[index];
+
+			if (!lpinfo->is_other_temp)
+			{
+				result = lappend_oid(result, lpinfo->reloid);
+				my_live_partitions = lappend_int(my_live_partitions,
+												 lpinfo->relid);
+			}
+		}
+		else
+		{
+			PartitionInfo *recurse_partinfo = all_partinfos[-index];
+			List		  *my_leaf_partitions;
+
+			if (!recurse_partinfo->is_other_temp)
+			{
+				my_live_partitions = lappend_int(my_live_partitions,
+												 recurse_partinfo->relid);
+				my_leaf_partitions = get_rel_partitions_recurse(rootrel,
+															recurse_partinfo,
+															all_partinfos,
+															leaf_part_infos);
+				result = list_concat(result, my_leaf_partitions);
+			}
+		}
+	}
+
+	painfo->live_partition_relids = my_live_partitions;
+
+	return result;
+}
+
+/*
+ * get_rel_partitions
+ *		Recursively find partitions of rel
+ */
+static List *
+get_rel_partitions(RelOptInfo *rel)
+{
+	return get_rel_partitions_recurse(rel,
+									  rel->partition_infos[0],
+									  rel->partition_infos,
+									  rel->leaf_part_infos);
+}
+
+/*
+ * find_rel_partitions
+ *		Find and lock partitions of rel relevant to this query
+ *
+ * Note that we only ever need to lock the leaf partitions, because the
+ * partitioned tables in the partition tree have already been locked.
+ */
+static void
+find_partitions_for_query(PlannerInfo *root, RelOptInfo *rel)
+{
+	List	   *leaf_part_oids = NIL;
+	ListCell   *l;
+	PlanRowMark *rc = NULL;
+	int		lockmode;
+	int		num_leaf_parts,
+			i;
+	Oid	   *leaf_part_oids_array;
+	PartitionRootInfo *prinfo = NULL;
+
+	/* Find partitions. */
+	Assert(rel->partition_infos != NULL);
+	leaf_part_oids = get_rel_partitions(rel);
+
+	/* Convert the list to an array and sort for binary searching later. */
+	num_leaf_parts = list_length(leaf_part_oids);
+	leaf_part_oids_array = (Oid *) palloc(num_leaf_parts * sizeof(Oid));
+	i = 0;
+	foreach(l, leaf_part_oids)
+	{
+		leaf_part_oids_array[i++] = lfirst_oid(l);
+	}
+	qsort(leaf_part_oids_array, num_leaf_parts, sizeof(Oid), oid_cmp);
+
+	/*
+	 * Now lock partitions.  Note that rel cannot be a result relation or we
+	 * wouldn't be here (inheritance_planner is where result relations go).
+	 */
+	rc = get_plan_rowmark(root->rowMarks, rel->relid);
+	if (rc && RowMarkRequiresRowShareLock(rc->markType))
+		lockmode = RowShareLock;
+	else
+		lockmode = AccessShareLock;
+
+	/*
+	 * We lock leaf partitions in the order in which find_all_inheritors
+	 * found them in expand_inherited_rtentry().  Find that list by locating
+	 * the PartitionRootInfo for this table.
+	 */
+	foreach(l, root->prinfo_list)
+	{
+		prinfo = lfirst(l);
+
+		if (rel->relid == prinfo->parent_relid)
+			break;
+	}
+	Assert(prinfo != NULL && rel->relid == prinfo->parent_relid);
+	foreach(l, prinfo->orig_leaf_part_oids)
+	{
+		Oid		relid = lfirst_oid(l);
+		Oid	   *test;
+
+		/* Will this leaf partition be scanned? */
+		test = (Oid *) bsearch(&relid,
+							   leaf_part_oids_array,
+							   num_leaf_parts,
+							   sizeof(Oid), oid_cmp);
+		/* Yep, so lock. */
+		if (test != NULL)
+			LockRelationOid(relid, lockmode);
+	}
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -866,6 +1035,158 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	double	   *parent_attrsizes;
 	int			nattrs;
 	ListCell   *l;
+	List	   *rel_appinfos = NIL;
+
+	/*
+	 * Collect a list child AppendRelInfo's, which in the non-partitioned
+	 * case will be found in root->append_rel_list.  In the partitioned
+	 * table's case, we didn't build any AppendRelInfo's yet.  We will
+	 * do the same after figuring out which of the table's child tables
+	 * (aka partitions) will need to be scanned for this query.
+	 */
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach(l, root->append_rel_list)
+		{
+			AppendRelInfo *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		Index		root_parent_relid;
+		List	   *live_partitions,
+				   *parent_vars;
+		Relation	parent;
+
+		/*
+		 * If this is a partitioned table root, we will determine all the
+		 * partitions in this partition tree that we need to scan for this
+		 * query.  Among those, partitions that have not yet been locked (viz.
+		 * the leaf partitions), will be.
+		 */
+		if (rel->partition_infos != NULL)
+		{
+			PartitionAppendInfo *painfo;
+
+			root_parent_relid = rti;
+
+			find_partitions_for_query(root, rel);
+			painfo = linitial(rel->live_partition_painfos);
+			Assert(rti == painfo->parent_relid);
+			live_partitions = painfo->live_partition_relids;
+
+			parent = rel->partition_infos[0]->pd->reldesc;
+		}
+		else
+		{
+			int		i;
+			RelOptInfo *rootrel;
+
+			root_parent_relid = rel->root_parent_relid;
+			rootrel = root->simple_rel_array[root_parent_relid];
+
+			/*
+			 * Just need to get hold of the PartitionAppendInfo via the root
+			 * parent's RelOptInfo.
+			 */
+			i = 0;
+			foreach(l, rootrel->live_partition_painfos)
+			{
+				PartitionAppendInfo *painfo = lfirst(l);
+
+				if (rti == painfo->parent_relid)
+				{
+					live_partitions = painfo->live_partition_relids;
+					break;
+				}
+
+				/* Skip to the index of this table's PartitionInfo. */
+				i++;
+			}
+
+			/*
+			 * For non-root parttioned tables, we already have a relcache
+			 * pointer that RelationGetPartitionDispatchInfo() acquired for
+			 * us.
+			 */
+			parent = rootrel->partition_infos[i]->pd->reldesc;
+		}
+
+		/*
+		 * Create an AppendRelInfo and a RelOptInfo for every candidate
+		 * partition.
+		 */
+		parent_vars = build_rel_vars(parent, rti);
+		foreach(l, live_partitions)
+		{
+			Index		childRTindex = lfirst_int(l);
+			RangeTblEntry *childrte = planner_rt_fetch(childRTindex, root);
+			Relation	child;
+			AppendRelInfo *appinfo;
+			RelOptInfo	  *childrel;
+
+			child = heap_open(childrte->relid, NoLock);	/* already locked! */
+			appinfo = makeNode(AppendRelInfo);
+			appinfo->parent_relid = rti;
+			appinfo->child_relid = childRTindex;
+			appinfo->parent_reltype = parent->rd_rel->reltype;
+			appinfo->child_reltype = child->rd_rel->reltype;
+			appinfo->translated_vars = map_partition_varattnos(parent_vars,
+															   rti,
+															   child, parent,
+															   NULL);
+			ChangeVarNodes((Node *) appinfo->translated_vars,
+						   rti, childRTindex, 0);
+			appinfo->parent_reloid = rte->relid;
+
+			/* For the main loop below that does per-child table processing. */
+			rel_appinfos = lappend(rel_appinfos, appinfo);
+
+			/*
+			 * While at it, also add the appinfo into root->append_rel_list,
+			 * so that any place that obtains a parent's children by looking
+			 * them up in that list are able to do so.
+			 */
+			root->append_rel_list = lappend(root->append_rel_list, appinfo);
+
+			/*
+			 * Translate the column permissions bitmaps to the child's attnums
+			 * (we have to build the translated_vars list before we can do
+			 * this). But if this is the parent table, leave copyObject's
+			 * result alone.
+			 *
+			 * Note: we need to do this even though the executor won't run any
+			 * permissions checks on the child RTE.  The
+			 * insertedCols/updatedCols bitmaps may be examined for
+			 * trigger-firing purposes.
+			 */
+			childrte->selectedCols = translate_col_privs(rte->selectedCols,
+													appinfo->translated_vars);
+			childrte->insertedCols = translate_col_privs(rte->insertedCols,
+													appinfo->translated_vars);
+			childrte->updatedCols = translate_col_privs(rte->updatedCols,
+													appinfo->translated_vars);
+
+			childrel = build_simple_rel(root, childRTindex, rel);
+			childrel->root_parent_relid = root_parent_relid;
+			Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
+
+			/* Copy the data that create_lateral_join_info() created */
+			Assert(childrel->direct_lateral_relids == NULL);
+			childrel->direct_lateral_relids = rel->direct_lateral_relids;
+			Assert(childrel->lateral_relids == NULL);
+			childrel->lateral_relids = rel->lateral_relids;
+			Assert(childrel->lateral_referencers == NULL);
+			childrel->lateral_referencers = rel->lateral_referencers;
+
+			root->total_table_pages += childrel->pages;
+			heap_close(child, NoLock);
+		}
+	}
 
 	Assert(IS_SIMPLE_REL(rel));
 
@@ -889,7 +1210,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -902,10 +1223,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -1211,24 +1528,61 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 	int			parentRTindex = rti;
 	List	   *live_childrels = NIL;
 	ListCell   *l;
+	List	   *append_rel_children = NIL;
+
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach(l, root->append_rel_list)
+		{
+			AppendRelInfo *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				append_rel_children = lappend_int(append_rel_children,
+												  appinfo->child_relid);
+		}
+	}
+	else
+	{
+		/* For a partitioned table, first find its PartitionAppendInfo */
+		if (rel->live_partition_painfos != NIL)
+		{
+			PartitionAppendInfo *painfo;
+
+			/* This is the root partitioned rel. */
+			painfo = linitial(rel->live_partition_painfos);
+			append_rel_children = painfo->live_partition_relids;
+		}
+		else
+		{
+			RelOptInfo *rootrel;
+
+			/* Non-root partitioned table.  Get it from the root rel. */
+			rootrel = root->simple_rel_array[rel->root_parent_relid];
+			foreach(l, rootrel->live_partition_painfos)
+			{
+				PartitionAppendInfo *painfo = lfirst(l);
+
+				if (rti == painfo->parent_relid)
+				{
+					append_rel_children = painfo->live_partition_relids;
+					break;
+				}
+			}
+		}
+	}
 
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, append_rel_children)
 	{
-		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
-		int			childRTindex;
+		int			childRTindex = lfirst_int(l);
 		RangeTblEntry *childRTE;
 		RelOptInfo *childrel;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		/* Re-locate the child RTE and RelOptInfo */
-		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 		childrel = root->simple_rel_array[childRTindex];
 
@@ -1289,7 +1643,14 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	RangeTblEntry *rte;
 
 	rte = planner_rt_fetch(rel->relid, root);
-	if (rte->relkind == RELKIND_PARTITIONED_TABLE)
+
+	/*
+	 * Note that get_partitioned_child_rels must be called only for root
+	 * partitioned tables and only those have rel->live_partition_painfos
+	 * set.
+	 */
+	if (rte->relkind == RELKIND_PARTITIONED_TABLE &&
+		rel->live_partition_painfos != NIL)
 	{
 		partitioned_rels = get_partitioned_child_rels(root, rel->relid);
 		/* The root partitioned table is included as a child rel */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index 966230256e..1a85c83c50 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -520,7 +520,7 @@ subquery_planner(PlannerGlobal *glob, Query *parse,
 	root->multiexpr_params = NIL;
 	root->eq_classes = NIL;
 	root->append_rel_list = NIL;
-	root->pcinfo_list = NIL;
+	root->prinfo_list = NIL;
 	root->rowMarks = NIL;
 	memset(root->upper_rels, 0, sizeof(root->upper_rels));
 	memset(root->upper_targets, 0, sizeof(root->upper_targets));
@@ -1056,6 +1056,102 @@ inheritance_planner(PlannerInfo *root)
 	Index		rti;
 	RangeTblEntry *parent_rte;
 	List	   *partitioned_rels = NIL;
+	List	   *rel_appinfos = NIL;
+	ListCell   *l;
+
+	parent_rte = rt_fetch(parentRTindex, root->parse->rtable);
+	if (parent_rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach(l, root->append_rel_list)
+		{
+			AppendRelInfo *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		PartitionRootInfo *prinfo = NULL;
+		Relation	parent;
+		List	   *parent_vars;
+
+		/* Find the PartitionRootInfo for this parent. */
+		foreach(l, root->prinfo_list)
+		{
+			prinfo = lfirst(l);
+
+			if (prinfo->parent_relid == parentRTindex)
+				break;
+		}
+		Assert(prinfo != NULL && prinfo->parent_relid == parentRTindex);
+
+		parent = heap_open(parent_rte->relid, NoLock);
+		parent_vars = build_rel_vars(parent, parentRTindex);
+		foreach(l, prinfo->leaf_part_infos)
+		{
+			LeafPartitionInfo *lpinfo = lfirst(l);
+			Index		childRTindex = lpinfo->relid;
+			RangeTblEntry *childrte = planner_rt_fetch(childRTindex, root);
+			Relation	child;
+			AppendRelInfo *appinfo;
+
+			if (childrte->relkind == RELKIND_PARTITIONED_TABLE)
+				continue;
+
+			/*
+			 * We'll need RowExclusiveLock, because just like the parent, each
+			 * child is a result relation.
+			 */
+			child = heap_open(childrte->relid, RowExclusiveLock);
+			appinfo = makeNode(AppendRelInfo);
+			appinfo->parent_relid = parentRTindex;
+			appinfo->child_relid = childRTindex;
+			appinfo->parent_reltype = parent->rd_rel->reltype;
+			appinfo->child_reltype = child->rd_rel->reltype;
+			appinfo->translated_vars = map_partition_varattnos(parent_vars,
+															   parentRTindex,
+															   child, parent,
+															   NULL);
+			ChangeVarNodes((Node *) appinfo->translated_vars,
+						   parentRTindex, childRTindex, 0);
+			appinfo->parent_reloid = RelationGetRelid(parent);
+
+			/* For the main loop below that does per-child table planning. */
+			rel_appinfos = lappend(rel_appinfos, appinfo);
+
+			/*
+			 * While at it, also add the appinfo into root->append_rel_list,
+			 * so that any places that obtain a parent's children by looking
+			 * them up in that list are able to do so.
+			 */
+			root->append_rel_list = lappend(root->append_rel_list, appinfo);
+
+			/*
+			 * Translate the column permissions bitmaps to the child's attnums
+			 * (we have to build the translated_vars list before we can do
+			 * this). But if this is the parent table, leave copyObject's
+			 * result alone.
+			 *
+			 * Note: we need to do this even though the executor won't run any
+			 * permissions checks on the child RTE.  The
+			 * insertedCols/updatedCols bitmaps may be examined for
+			 * trigger-firing purposes.
+			 */
+			childrte->selectedCols =
+								translate_col_privs(parent_rte->selectedCols,
+													appinfo->translated_vars);
+			childrte->insertedCols =
+								translate_col_privs(parent_rte->insertedCols,
+													appinfo->translated_vars);
+			childrte->updatedCols =
+								translate_col_privs(parent_rte->updatedCols,
+													appinfo->translated_vars);
+			heap_close(child, NoLock);
+		}
+		heap_close(parent, NoLock);
+	}
 
 	Assert(parse->commandType != CMD_INSERT);
 
@@ -1121,14 +1217,13 @@ inheritance_planner(PlannerInfo *root)
 	 * opposite in the case of non-partitioned inheritance parent as described
 	 * below.
 	 */
-	parent_rte = rt_fetch(parentRTindex, root->parse->rtable);
 	if (parent_rte->relkind == RELKIND_PARTITIONED_TABLE)
 		nominalRelation = parentRTindex;
 
 	/*
 	 * And now we can get on with generating a plan for each child table.
 	 */
-	foreach(lc, root->append_rel_list)
+	foreach(lc, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(lc);
 		PlannerInfo *subroot;
@@ -1136,10 +1231,6 @@ inheritance_planner(PlannerInfo *root)
 		RelOptInfo *sub_final_rel;
 		Path	   *subpath;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		/*
 		 * We need a working copy of the PlannerInfo so that we can control
 		 * propagation of information back to the main copy.
@@ -6076,7 +6167,7 @@ plan_cluster_use_sort(Oid tableOid, Oid indexOid)
  *		Returns a list of the RT indexes of the partitioned child relations
  *		with rti as the root parent RT index.
  *
- * Note: Only call this function on RTEs known to be partitioned tables.
+ * Note: Only call this function on RTEs known to be a root partitioned table.
  */
 List *
 get_partitioned_child_rels(PlannerInfo *root, Index rti)
@@ -6084,13 +6175,13 @@ get_partitioned_child_rels(PlannerInfo *root, Index rti)
 	List	   *result = NIL;
 	ListCell   *l;
 
-	foreach(l, root->pcinfo_list)
+	foreach(l, root->prinfo_list)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		PartitionRootInfo *prinfo = lfirst(l);
 
-		if (pc->parent_relid == rti)
+		if (prinfo->parent_relid == rti)
 		{
-			result = pc->child_rels;
+			result = prinfo->partitioned_relids;
 			break;
 		}
 	}
diff --git a/src/backend/optimizer/plan/setrefs.c b/src/backend/optimizer/plan/setrefs.c
index b0c9e94459..4666e446d7 100644
--- a/src/backend/optimizer/plan/setrefs.c
+++ b/src/backend/optimizer/plan/setrefs.c
@@ -15,7 +15,9 @@
  */
 #include "postgres.h"
 
+#include "access/heapam.h"
 #include "access/transam.h"
+#include "catalog/partition.h"
 #include "catalog/pg_type.h"
 #include "nodes/makefuncs.h"
 #include "nodes/nodeFuncs.h"
@@ -204,6 +206,10 @@ static bool extract_query_dependencies_walker(Node *node,
  * to process targetlist and qual expressions.  We can assume that the Plan
  * nodes were just built by the planner and are not multiply referenced, but
  * it's not so safe to assume that for expression tree nodes.
+ *
+ * Finally, we close some relcache references lintering in root.  They are
+ * those of the partitioned tables whose PartitionDispatch objects are
+ * referenced from within root->prinfo_list.
  */
 Plan *
 set_plan_references(PlannerInfo *root, Plan *plan)
@@ -238,6 +244,26 @@ set_plan_references(PlannerInfo *root, Plan *plan)
 		glob->finalrowmarks = lappend(glob->finalrowmarks, newrc);
 	}
 
+	/*
+	 * Close relcache references in PartitionDispatch objects referenced in
+	 * root.
+	 */
+	foreach(lc, root->prinfo_list)
+	{
+		PartitionRootInfo *prinfo = lfirst(lc);
+		ListCell *lc1;
+
+		foreach(lc1, prinfo->partition_infos)
+		{
+			PartitionInfo *pinfo = lfirst(lc1);
+
+			if (pinfo->pd->reldesc)
+				heap_close(pinfo->pd->reldesc, NoLock);
+			/* Shouldn't try to close again. XXX - hack? */
+			pinfo->pd->reldesc = NULL;
+		}
+	}
+
 	/* Now fix the Plan tree */
 	return set_plan_refs(root, plan, rtoffset);
 }
diff --git a/src/backend/optimizer/prep/prepunion.c b/src/backend/optimizer/prep/prepunion.c
index 0d20ffa2f7..01de2d778d 100644
--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -105,8 +105,6 @@ static void make_inh_translation_list(Relation oldrelation,
 						  Relation newrelation,
 						  Index newvarno,
 						  List **translated_vars);
-static Bitmapset *translate_col_privs(const Bitmapset *parent_privs,
-					List *translated_vars);
 static Node *adjust_appendrel_attrs_mutator(Node *node,
 							   adjust_appendrel_attrs_context *context);
 static Relids adjust_child_relids(Relids relids, int nappinfos,
@@ -1352,11 +1350,19 @@ expand_inherited_tables(PlannerInfo *root)
 
 /*
  * expand_inherited_rtentry
- *		Check whether a rangetable entry represents an inheritance set.
- *		If so, add entries for all the child tables to the query's
- *		rangetable, and build AppendRelInfo nodes for all the child tables
- *		and add them to root->append_rel_list.  If not, clear the entry's
- *		"inh" flag to prevent later code from looking for AppendRelInfos.
+ *		Perform actions necessary for applying this query to an inheritance
+ *		set if the rte represents one
+ *
+ * That includes adding entries for all the child tables to the query's
+ * rangetable.  Also, if this query requires a PlanRowMark, generate the same
+ * for each child table and append them to the planner's global list
+ * (root->rowMarks).  If the inheritance set is really a partitioned table,
+ * our work here is done.  If not, we also create AppendRelInfo nodes for
+ * all the child tables and add them to root->append_rel_list.
+ *
+ * If it turns out that the rte is not (or no longer) an inheritance set,
+ * clear the entry's "inh" flag to prevent later code from looking for
+ * AppendRelInfos.
  *
  * Note that the original RTE is considered to represent the whole
  * inheritance set.  The first of the generated RTEs is an RTE for the same
@@ -1381,9 +1387,14 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 	List	   *inhOIDs;
 	List	   *appinfos;
 	ListCell   *l;
-	bool		has_child;
-	PartitionedChildRelInfo *pcinfo;
 	List	   *partitioned_child_rels = NIL;
+	List	   *partition_infos = NIL;
+	List	   *leaf_part_infos = NIL;
+	List	   *orig_leaf_part_oids;
+	int			num_partitioned_children,
+				i;
+	PartitionDispatch *pds;
+	PartitionInfo *pinfo;
 
 	/* Does RT entry allow inheritance? */
 	if (!rte->inh)
@@ -1408,6 +1419,10 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 	 * relation named in the query.  However, for each child relation we add
 	 * to the query, we must obtain an appropriate lock, because this will be
 	 * the first use of those relations in the parse/rewrite/plan pipeline.
+	 * For a partitioned table, we defer locking non-partitioned child tables
+	 * (aka leaf partitions) to when we actually know that they will be
+	 * scanned for this query.  We do that by passing 'true' for
+	 * lock_only_partitioned_children.
 	 *
 	 * If the parent relation is the query's result relation, then we need
 	 * RowExclusiveLock.  Otherwise, if it's accessed FOR UPDATE/SHARE, we
@@ -1425,7 +1440,8 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 		lockmode = AccessShareLock;
 
 	/* Scan for all members of inheritance set, acquire needed locks */
-	inhOIDs = find_all_inheritors(parentOID, lockmode, false, NULL, NULL);
+	inhOIDs = find_all_inheritors(parentOID, lockmode, true, NULL,
+								  &num_partitioned_children);
 
 	/*
 	 * Check that there's at least one descendant, else treat as no-child
@@ -1460,28 +1476,43 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 	if (rte->relkind == RELKIND_PARTITIONED_TABLE)
 	{
 		List   *leaf_part_oids;
-		int		num_parted,
-				i;
-		PartitionDispatch *pds;
+		int		num_parted;
+		Relation	rootrel;
+
+		/*
+		 * Keep leaf partition OIDs around so that we can lock them in this
+		 * order when we eventually do it.
+		 */
+		orig_leaf_part_oids = list_copy_tail(inhOIDs,
+											 num_partitioned_children + 1);
 
-		/* Discard the original list. */
-		list_free(inhOIDs);
+		/* Discard the original inhOIDs list. */
 		inhOIDs = NIL;
 
-		/* Request partitioning information. */
-		pds = RelationGetPartitionDispatchInfo(oldrelation, &num_parted,
+		/*
+		 * Request partitioning information.  We don't pass oldrelation,
+		 * because we want to keep the relcache pointer in PartitionDispatch
+		 * open until much later, but we'll be closing oldrelation before
+		 * returning from this function.
+		 */
+		rootrel = heap_open(rte->relid, NoLock);
+		pds = RelationGetPartitionDispatchInfo(rootrel, &num_parted,
 											   &leaf_part_oids);
 
 		/*
-		 * First collect the partitioned child table OIDs, which includes the
-		 * root parent at the head.
+		 * We make a PartitionInfo object for every partitioned table in the
+		 * tree, including the root table.  Note that we create the root
+		 * table's PartitionInfo outside the loop, because inhOIDs will not
+		 * contain its OID.  Also add the original rti to
+		 * partitioned_child_rels.
 		 */
-		for (i = 0; i < num_parted; i++)
-		{
+		pinfo = makeNode(PartitionInfo);
+		pinfo->relid = rti;
+		pinfo->pd = pds[0];
+		partition_infos = list_make1(pinfo);
+		partitioned_child_rels = list_make1_int(rti);
+		for (i = 1; i < num_parted; i++)
 			inhOIDs = lappend_oid(inhOIDs, RelationGetRelid(pds[i]->reldesc));
-			if (pds[i]->reldesc != oldrelation)
-				heap_close(pds[i]->reldesc, NoLock);
-		}
 
 		/* Concatenate the leaf partition OIDs. */
 		inhOIDs = list_concat(inhOIDs, leaf_part_oids);
@@ -1489,20 +1520,16 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 
 	/* Scan the inheritance set and expand it */
 	appinfos = NIL;
-	has_child = false;
+	i = 1;
 	foreach(l, inhOIDs)
 	{
 		Oid			childOID = lfirst_oid(l);
 		Relation	newrelation;
-		RangeTblEntry *childrte;
-		Index		childRTindex;
+		RangeTblEntry *childrte = NULL;
+		Index		childRTindex = 0;
 		AppendRelInfo *appinfo;
-
-		/* Open rel if needed; we already have required locks */
-		if (childOID != parentOID)
-			newrelation = heap_open(childOID, NoLock);
-		else
-			newrelation = oldrelation;
+		bool		is_other_temp;
+		char		child_relkind = get_rel_relkind(childOID);
 
 		/*
 		 * It is possible that the parent table has children that are temp
@@ -1510,11 +1537,7 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 		 * (because of buffering issues), and the best thing to do seems to be
 		 * to silently ignore them.
 		 */
-		if (childOID != parentOID && RELATION_IS_OTHER_TEMP(newrelation))
-		{
-			heap_close(newrelation, lockmode);
-			continue;
-		}
+		is_other_temp = rel_is_other_temp(childOID);
 
 		/*
 		 * Build an RTE for the child, and attach to query's rangetable list.
@@ -1528,64 +1551,22 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 		 * The parent securityQuals will be propagated to children along with
 		 * other base restriction clauses, so we don't need to do it here.
 		 */
-		childrte = copyObject(rte);
-		childrte->relid = childOID;
-		childrte->relkind = newrelation->rd_rel->relkind;
-		childrte->inh = false;
-		childrte->requiredPerms = 0;
-		childrte->securityQuals = NIL;
-		parse->rtable = lappend(parse->rtable, childrte);
-		childRTindex = list_length(parse->rtable);
-
-		/*
-		 * Build an AppendRelInfo for this parent and child, unless the child
-		 * is a partitioned table.
-		 */
-		if (childrte->relkind != RELKIND_PARTITIONED_TABLE)
+		if (!is_other_temp)
 		{
-			/* Remember if we saw a real child. */
-			if (childOID != parentOID)
-				has_child = true;
-
-			appinfo = makeNode(AppendRelInfo);
-			appinfo->parent_relid = rti;
-			appinfo->child_relid = childRTindex;
-			appinfo->parent_reltype = oldrelation->rd_rel->reltype;
-			appinfo->child_reltype = newrelation->rd_rel->reltype;
-			make_inh_translation_list(oldrelation, newrelation, childRTindex,
-									  &appinfo->translated_vars);
-			appinfo->parent_reloid = parentOID;
-			appinfos = lappend(appinfos, appinfo);
-
-			/*
-			 * Translate the column permissions bitmaps to the child's attnums
-			 * (we have to build the translated_vars list before we can do
-			 * this). But if this is the parent table, leave copyObject's
-			 * result alone.
-			 *
-			 * Note: we need to do this even though the executor won't run any
-			 * permissions checks on the child RTE.  The
-			 * insertedCols/updatedCols bitmaps may be examined for
-			 * trigger-firing purposes.
-			 */
-			if (childOID != parentOID)
-			{
-				childrte->selectedCols = translate_col_privs(rte->selectedCols,
-															 appinfo->translated_vars);
-				childrte->insertedCols = translate_col_privs(rte->insertedCols,
-															 appinfo->translated_vars);
-				childrte->updatedCols = translate_col_privs(rte->updatedCols,
-															appinfo->translated_vars);
-			}
+			childrte = copyObject(rte);
+			childrte->relid = childOID;
+			childrte->relkind = get_rel_relkind(childOID);
+			childrte->inh = (childrte->relkind == RELKIND_PARTITIONED_TABLE);
+			childrte->requiredPerms = 0;
+			childrte->securityQuals = NIL;
+			parse->rtable = lappend(parse->rtable, childrte);
+			childRTindex = list_length(parse->rtable);
 		}
-		else
-			partitioned_child_rels = lappend_int(partitioned_child_rels,
-												 childRTindex);
 
 		/*
 		 * Build a PlanRowMark if parent is marked FOR UPDATE/SHARE.
 		 */
-		if (oldrc)
+		if (!is_other_temp && oldrc)
 		{
 			PlanRowMark *newrc = makeNode(PlanRowMark);
 
@@ -1606,12 +1587,89 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 			 */
 			newrc->isParent = (childrte->relkind == RELKIND_PARTITIONED_TABLE);
 
-			/* Include child's rowmark type in parent's allMarkTypes */
-			oldrc->allMarkTypes |= newrc->allMarkTypes;
 
 			root->rowMarks = lappend(root->rowMarks, newrc);
 		}
 
+		/*
+		 * No need to create AppendRelInfo for partitions at this point.  We
+		 * will create one if and when we know we'll need it.  The fact that
+		 * this is a child table of the parent table will be recorded in the
+		 * PartitionRootInfo that will be created for the parent table.
+		 */
+		if (rel_is_partition(childOID) &&
+			child_relkind != RELKIND_PARTITIONED_TABLE)
+		{
+			LeafPartitionInfo   *lpinfo = makeNode(LeafPartitionInfo);
+
+			lpinfo->is_other_temp = is_other_temp;
+			lpinfo->reloid = childOID;
+			lpinfo->relid = childRTindex;
+			leaf_part_infos = lappend(leaf_part_infos, lpinfo);
+			continue;
+		}
+
+		/* Create the PartitionInfo of this child partitioned table. */
+		if (child_relkind == RELKIND_PARTITIONED_TABLE)
+		{
+			PartitionInfo *pinfo = makeNode(PartitionInfo);
+
+			pinfo->is_other_temp = is_other_temp;
+			pinfo->relid = childRTindex;
+			pinfo->pd = pds[i++];
+			partition_infos = lappend(partition_infos, pinfo);
+
+			partitioned_child_rels = lappend_int(partitioned_child_rels,
+												 childRTindex);
+			continue;
+		}
+
+		if (is_other_temp)
+			continue;
+
+		/*
+		 * Getting here means this is a non-partitioned child table that is
+		 * not a partition.  Build an AppendRelInfo for the same to remember
+		 * the parent-child relationship.
+		 */
+
+		/* Open rel if needed, we already have required locks */
+		if (childOID != parentOID)
+			newrelation = heap_open(childOID, NoLock);
+		else
+			newrelation = oldrelation;
+
+		appinfo = makeNode(AppendRelInfo);
+		appinfo->parent_relid = rti;
+		appinfo->child_relid = childRTindex;
+		appinfo->parent_reltype = oldrelation->rd_rel->reltype;
+		appinfo->child_reltype = newrelation->rd_rel->reltype;
+		make_inh_translation_list(oldrelation, newrelation, childRTindex,
+								  &appinfo->translated_vars);
+		appinfo->parent_reloid = parentOID;
+		appinfos = lappend(appinfos, appinfo);
+
+		/*
+		 * Translate the column permissions bitmaps to the child's attnums
+		 * (we have to build the translated_vars list before we can do
+		 * this). But if this is the parent table, leave copyObject's
+		 * result alone.
+		 *
+		 * Note: we need to do this even though the executor won't run any
+		 * permissions checks on the child RTE.  The
+		 * insertedCols/updatedCols bitmaps may be examined for
+		 * trigger-firing purposes.
+		 */
+		if (childOID != parentOID)
+		{
+			childrte->selectedCols = translate_col_privs(rte->selectedCols,
+														 appinfo->translated_vars);
+			childrte->insertedCols = translate_col_privs(rte->insertedCols,
+														 appinfo->translated_vars);
+			childrte->updatedCols = translate_col_privs(rte->updatedCols,
+														appinfo->translated_vars);
+		}
+
 		/* Close child relations, but keep locks */
 		if (childOID != parentOID)
 			heap_close(newrelation, NoLock);
@@ -1620,35 +1678,49 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 	heap_close(oldrelation, NoLock);
 
 	/*
-	 * If all the children were temp tables or a partitioned parent did not
-	 * have any leaf partitions, pretend it's a non-inheritance situation; we
-	 * don't need Append node in that case.  The duplicate RTE we added for
-	 * the parent table is harmless, so we don't bother to get rid of it;
-	 * ditto for the useless PlanRowMark node.
+	 * We keep a list of objects in root, each of which maps a partitioned
+	 * parent RT index to a bunch of information about the partition tree
+	 * rooted at that parent.  The information includes a list of RT indexes
+	 * of partitioned tables appearing in the tree, a list of PartitionInfo
+	 * objects for each such partitioned table, a list of LeafPartitionInfo
+	 * objects for each leaf partition in tree, and finally a list containing
+	 * leaf partition OIDs in an order in which find_all_inheritors() returned
+	 * them.  The first of these is used when creating an Append or a
+	 * ModifyTable path for the parent to be copied verbatim into the path
+	 * (and subsequently the plan) so that it could be carried over to the
+	 * executor.  That list is the only place where the executor could find
+	 * partitioned child tables to lock them.
 	 */
-	if (!has_child)
+	if (rte->relkind == RELKIND_PARTITIONED_TABLE)
 	{
-		/* Clear flag before returning */
-		rte->inh = false;
+		PartitionRootInfo *prinfo = makeNode(PartitionRootInfo);
+
+		Assert(list_length(partition_infos) >= 1);
+		prinfo->parent_relid = rti;
+		prinfo->partitioned_relids = partitioned_child_rels;
+		prinfo->partition_infos = partition_infos;
+		prinfo->leaf_part_infos = leaf_part_infos;
+		prinfo->orig_leaf_part_oids = orig_leaf_part_oids;
+
+		root->prinfo_list = lappend(root->prinfo_list, prinfo);
+
+		/*
+		 * Our job here is done, because we didn't create any AppendRelInfos.
+		 */
 		return;
 	}
 
 	/*
-	 * We keep a list of objects in root, each of which maps a partitioned
-	 * parent RT index to the list of RT indexes of its partitioned child
-	 * tables.  When creating an Append or a ModifyTable path for the parent,
-	 * we copy the child RT index list verbatim to the path so that it could
-	 * be carried over to the executor so that the latter could identify the
-	 * partitioned child tables.
+	 * If all the children were temp tables, pretend it's a non-inheritance
+	 * situation; we don't need Append node in that case.  The duplicate
+	 * RTE we added for the parent table is harmless, so we don't bother to
+	 * get rid of it; ditto for the useless PlanRowMark node.
 	 */
-	if (partitioned_child_rels != NIL)
+	if (list_length(appinfos) < 2)
 	{
-		pcinfo = makeNode(PartitionedChildRelInfo);
-
-		Assert(rte->relkind == RELKIND_PARTITIONED_TABLE);
-		pcinfo->parent_relid = rti;
-		pcinfo->child_rels = partitioned_child_rels;
-		root->pcinfo_list = lappend(root->pcinfo_list, pcinfo);
+		/* Clear flag before returning */
+		rte->inh = false;
+		return;
 	}
 
 	/* Otherwise, OK to add to root->append_rel_list */
@@ -1769,7 +1841,7 @@ make_inh_translation_list(Relation oldrelation, Relation newrelation,
  * query is really only going to reference the inherited columns.  Instead
  * we set the per-column bits for all inherited columns.
  */
-static Bitmapset *
+Bitmapset *
 translate_col_privs(const Bitmapset *parent_privs,
 					List *translated_vars)
 {
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index a1ebd4acc8..3781a91b76 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1577,6 +1577,43 @@ build_physical_tlist(PlannerInfo *root, RelOptInfo *rel)
 }
 
 /*
+ * build_rel_vars
+ *
+ * Returns a list containing Var expressions corresponding to a relation's
+ * attributes.  Since the caller may already have the RangeTblEntry, we it
+ * pass the same instead of PlannerInfo to avoid finding it in the range
+ * table all over again.
+ */
+List *
+build_rel_vars(Relation relation, Index relid)
+{
+	AttrNumber	attrno;
+	int			numattrs;
+	List	   *result = NIL;
+
+	numattrs = RelationGetNumberOfAttributes(relation);
+	for (attrno = 1; attrno <= numattrs; attrno++)
+	{
+		Form_pg_attribute att_tup = TupleDescAttr(relation->rd_att,
+												  attrno - 1);
+
+		if (att_tup->attisdropped)
+			continue;
+
+		result = lappend(result,
+						 makeVar(relid,
+								 attrno,
+								 att_tup->atttypid,
+								 att_tup->atttypmod,
+								 att_tup->attcollation,
+								 0));
+
+	}
+
+	return result;
+}
+
+/*
  * build_index_tlist
  *
  * Build a targetlist representing the columns of the specified index.
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 8ad0b4a669..1bcda9254f 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -16,7 +16,9 @@
 
 #include <limits.h>
 
+#include "catalog/pg_class.h"
 #include "miscadmin.h"
+#include "nodes/relation.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -146,6 +148,15 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->baserestrict_min_security = UINT_MAX;
 	rel->joininfo = NIL;
 	rel->has_eclass_joins = false;
+	/* Set in build_simple_rel if rel is root partitioned table */
+	rel->num_parted = 0;
+	rel->partition_infos = NULL;
+	rel->num_leaf_parts = 0;
+	rel->leaf_part_infos = NULL;
+	/* Set in get_rel_partitions_recurse */
+	rel->live_partition_painfos = NIL;
+	/* Set in set_append_rel_size if rel is a partition. */
+	rel->root_parent_relid = 0;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -210,25 +221,83 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 										list_length(rte->securityQuals));
 
 	/*
-	 * If this rel is an appendrel parent, recurse to build "other rel"
-	 * RelOptInfos for its children.  They are "other rels" because they are
-	 * not in the main join tree, but we will need RelOptInfos to plan access
-	 * to them.
+	 * If this rel is an appendrel parent, generate additional information
+	 * based on whether the parent is a partitioned table or not.  For
+	 * regular parent tables, recurse to build "other rel" RelOptInfos for its
+	 * children.  They are "other rels" because they are not in the main join
+	 * tree, but we will need RelOptInfos to plan access to them.  For
+	 * partitioned parent tables, we do not yet create "other rel" RelOptInfos
+	 * for the children.  Instead, we set up some informations that will be
+	 * used in set_append_rel_size() to look up its partitions.
 	 */
 	if (rte->inh)
 	{
 		ListCell   *l;
 
-		foreach(l, root->append_rel_list)
+		if (rte->relkind == RELKIND_PARTITIONED_TABLE)
 		{
-			AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
+			PartitionRootInfo *prinfo = NULL;
+			LeafPartitionInfo **lpinfos;
+			int		i;
+
+			foreach(l, root->prinfo_list)
+			{
+				if (((PartitionRootInfo *) lfirst(l))->parent_relid == relid)
+				{
+					prinfo = lfirst(l);
+					break;
+				}
+			}
 
-			/* append_rel_list contains all append rels; ignore others */
-			if (appinfo->parent_relid != relid)
-				continue;
+			/*
+			 * Only the root partitioned tables have an entry in
+			 * root->prinfo_list.  For other partitioned table rels, we don't
+			 * need to set the following fields.
+			 */
+			if (prinfo == NULL)
+				return rel;
+
+			Assert(prinfo->parent_relid == relid);
+			rel->num_parted = list_length(prinfo->partition_infos);
+			rel->num_leaf_parts = list_length(prinfo->leaf_part_infos);
+			rel->partition_infos = (PartitionInfo **)
+											palloc0(rel->num_parted *
+												sizeof(PartitionInfo *));
+			lpinfos = (LeafPartitionInfo **) palloc0(rel->num_leaf_parts *
+												sizeof(LeafPartitionInfo *));
+			i = 0;
+			foreach(l, prinfo->partition_infos)
+			{
+				rel->partition_infos[i++] = lfirst(l);
+			}
+			i = 0;
+			foreach(l, prinfo->leaf_part_infos)
+			{
+				lpinfos[i++] = lfirst(l);
+			}
+			rel->leaf_part_infos = lpinfos;
+
+			/*
+			 * Don't build RelOptInfo for partitions yet; we don't know which
+			 * ones we'll need.  We did create RangeTblEntry's though, so we
+			 * have an empty slot in root->simple_rel_array that will be
+			 * filled eventually if the respective partition is chosen to be
+			 * scanned after all.
+			 */
+		}
+		else
+		{
+			foreach(l, root->append_rel_list)
+			{
+				AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
+
+				/* append_rel_list contains all append rels; ignore others */
+				if (appinfo->parent_relid != relid)
+					continue;
 
-			(void) build_simple_rel(root, appinfo->child_relid,
-									rel);
+				(void) build_simple_rel(root, appinfo->child_relid,
+										rel);
+			}
 		}
 	}
 
diff --git a/src/backend/utils/cache/lsyscache.c b/src/backend/utils/cache/lsyscache.c
index 82763f8013..ebbc3da985 100644
--- a/src/backend/utils/cache/lsyscache.c
+++ b/src/backend/utils/cache/lsyscache.c
@@ -1817,6 +1817,28 @@ get_rel_relkind(Oid relid)
 }
 
 /*
+ * rel_is_partition
+ *
+ *		Returns the relkind associated with a given relation.
+ */
+char
+rel_is_partition(Oid relid)
+{
+	HeapTuple	tp;
+	Form_pg_class reltup;
+	bool		result;
+
+	tp = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
+	if (!HeapTupleIsValid(tp))
+		elog(ERROR, "cache lookup failed for relation %u", relid);
+	reltup = (Form_pg_class) GETSTRUCT(tp);
+	result = reltup->relispartition;
+	ReleaseSysCache(tp);
+
+	return result;
+}
+
+/*
  * get_rel_tablespace
  *
  *		Returns the pg_tablespace OID associated with a given relation.
@@ -1865,6 +1887,34 @@ get_rel_persistence(Oid relid)
 	return result;
 }
 
+/*
+ * rel_is_other_temp
+ *
+ *		Returns whether a relation is a temp table from another session
+ */
+bool
+rel_is_other_temp(Oid relid)
+{
+	HeapTuple	tp;
+	Form_pg_class reltup;
+	bool		result = false;
+
+	tp = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
+	if (!HeapTupleIsValid(tp))
+		elog(ERROR, "cache lookup failed for relation %u", relid);
+	reltup = (Form_pg_class) GETSTRUCT(tp);
+
+	if (reltup->relpersistence == RELPERSISTENCE_TEMP &&
+		!isTempOrTempToastNamespace(reltup->relnamespace))
+	{
+		result = true;
+	}
+
+	ReleaseSysCache(tp);
+
+	return result;
+}
+
 
 /*				---------- TRANSFORM CACHE ----------						 */
 
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 1091dd572c..20fc3a89db 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -16,6 +16,7 @@
 #include "fmgr.h"
 #include "executor/tuptable.h"
 #include "nodes/execnodes.h"
+#include "nodes/relation.h"
 #include "parser/parse_node.h"
 #include "utils/rel.h"
 
@@ -93,4 +94,7 @@ extern int get_partition_for_tuple(PartitionTupleRoutingInfo **ptrinfos,
 						EState *estate,
 						PartitionTupleRoutingInfo **failed_at,
 						TupleTableSlot **failed_slot);
+
+/* Planner support stuff. */
+extern List *get_partitions_for_keys(PartitionDispatch pd);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/nodes.h b/src/include/nodes/nodes.h
index 27bd4f3363..e957615ac6 100644
--- a/src/include/nodes/nodes.h
+++ b/src/include/nodes/nodes.h
@@ -260,7 +260,10 @@ typedef enum NodeTag
 	T_PlaceHolderVar,
 	T_SpecialJoinInfo,
 	T_AppendRelInfo,
-	T_PartitionedChildRelInfo,
+	T_PartitionInfo,
+	T_LeafPartitionInfo,
+	T_PartitionAppendInfo,
+	T_PartitionRootInfo,
 	T_PlaceHolderInfo,
 	T_MinMaxAggInfo,
 	T_PlannerParamItem,
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index a39e59d8ac..a67a43b069 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -251,7 +251,7 @@ typedef struct PlannerInfo
 
 	List	   *append_rel_list;	/* list of AppendRelInfos */
 
-	List	   *pcinfo_list;	/* list of PartitionedChildRelInfos */
+	List	   *prinfo_list;	/* list of PartitionRootInfos */
 
 	List	   *rowMarks;		/* list of PlanRowMarks */
 
@@ -515,6 +515,9 @@ typedef enum RelOptKind
 /* Is the given relation an "other" relation? */
 #define IS_OTHER_REL(rel) ((rel)->reloptkind == RELOPT_OTHER_MEMBER_REL)
 
+typedef struct PartitionInfo PartitionInfo;
+typedef struct LeafPartitionInfo LeafPartitionInfo;
+
 typedef struct RelOptInfo
 {
 	NodeTag		type;
@@ -592,6 +595,23 @@ typedef struct RelOptInfo
 
 	/* used by "other" relations */
 	Relids		top_parent_relids;	/* Relids of topmost parents */
+
+	/* Fields set for "root" partitioned relations */
+	int		num_parted;			/* Number of entries in partition_infos */
+	PartitionInfo **partition_infos;
+	int		num_leaf_parts;		/* Number of entries in leaf_part_infos */
+	LeafPartitionInfo **leaf_part_infos;	/* LeafPartitionInfos */
+
+	/* Fields set for partitioned relations (list of PartitionAppendInfo's) */
+	List   *live_partition_painfos;
+
+	/* Fields set for partition otherrels */
+
+	/*
+	 * RT index of the root partitioned table in the the partition tree of
+	 * which this rel is a member.
+	 */
+	Index	root_parent_relid;
 } RelOptInfo;
 
 /*
@@ -2012,24 +2032,75 @@ typedef struct AppendRelInfo
 	Oid			parent_reloid;	/* OID of parent relation */
 } AppendRelInfo;
 
+/* Forward declarations, to avoid including other headers */
+typedef struct PartitionDispatchData *PartitionDispatch;
+
+/*
+ * PartitionInfo - information about partitioning of one partitioned table in
+ *				   a given partition tree
+ */
+typedef struct PartitionInfo
+{
+	NodeTag		type;
+
+	bool		is_other_temp;		/* If true, ignore the following fields */
+	Index				relid;		/* Ordinal position in the rangetable */
+	PartitionDispatch	pd;			/* Information about partitions */
+} PartitionInfo;
+
+/*
+ * LeafPartitionInfo - (OID, RT index) pair for one leaf partition
+ *
+ * Created when a leaf partition's RT entry is created in
+ * expand_inherited_rtentry().
+ */
+typedef struct LeafPartitionInfo
+{
+	NodeTag		type;
+
+	bool		is_other_temp;	/* If true, ignore the following fields. */
+	Oid			reloid;			/* OID */
+	Index		relid;			/* RT index */
+} LeafPartitionInfo;
+
 /*
- * For a partitioned table, this maps its RT index to the list of RT indexes
- * of the partitioned child tables in the partition tree.  We need to
- * separately store this information, because we do not create AppendRelInfos
- * for the partitioned child tables of a parent table, since AppendRelInfos
- * contain information that is unnecessary for the partitioned child tables.
- * The child_rels list must contain at least one element, because the parent
- * partitioned table is itself counted as a child.
+ * PartitionAppendInfo - list of child RT indexes for one partitioned table
+ *						 in a given partition tree
+ */
+typedef struct PartitionAppendInfo
+{
+	NodeTag		type;
+
+	Index		parent_relid;
+	List	   *live_partition_relids;	/* List of RT indexes */
+} PartitionAppendInfo;
+
+/*
+ * For a partitioned table, this maps its RT index to the information about
+ * the partition tree collected in expand_inherited_rtentry().
+ *
+ * That information includes a list of PartitionInfo nodes, one for each
+ * partitioned table in the partition tree, including for the table itself.
+ * Also included is a list of RT indexes of the entries for leaf partitions
+ * that are created at the same time by expand_inherited_rtentry().
+ *
+ * orig_leaf_part_oids contains the list of leaf partition OIDs as it was
+ * generated by find_all_inheritors().  We keep it around so that we can
+ * lock leaf partitions in that order when we actually do it.
  *
- * These structs are kept in the PlannerInfo node's pcinfo_list.
+ * PartitionRootInfo's for different partitioned tables in a query are placed
+ * in root->prinfo_list.
  */
-typedef struct PartitionedChildRelInfo
+typedef struct PartitionRootInfo
 {
 	NodeTag		type;
 
 	Index		parent_relid;
-	List	   *child_rels;
-} PartitionedChildRelInfo;
+	List	   *partition_infos;
+	List	   *partitioned_relids;
+	List	   *leaf_part_infos;
+	List	   *orig_leaf_part_oids;
+} PartitionRootInfo;
 
 /*
  * For each distinct placeholder expression generated during planning, we
diff --git a/src/include/optimizer/plancat.h b/src/include/optimizer/plancat.h
index 71f0faf938..e8e30f8f52 100644
--- a/src/include/optimizer/plancat.h
+++ b/src/include/optimizer/plancat.h
@@ -39,6 +39,7 @@ extern bool relation_excluded_by_constraints(PlannerInfo *root,
 								 RelOptInfo *rel, RangeTblEntry *rte);
 
 extern List *build_physical_tlist(PlannerInfo *root, RelOptInfo *rel);
+extern List *build_rel_vars(Relation relation, Index relid);
 
 extern bool has_unique_index(RelOptInfo *rel, AttrNumber attno);
 
diff --git a/src/include/optimizer/prep.h b/src/include/optimizer/prep.h
index 4be0afd566..d0af8dc7bc 100644
--- a/src/include/optimizer/prep.h
+++ b/src/include/optimizer/prep.h
@@ -16,6 +16,7 @@
 
 #include "nodes/plannodes.h"
 #include "nodes/relation.h"
+#include "utils/rel.h"
 
 
 /*
@@ -51,6 +52,8 @@ extern PlanRowMark *get_plan_rowmark(List *rowmarks, Index rtindex);
 extern RelOptInfo *plan_set_operations(PlannerInfo *root);
 
 extern void expand_inherited_tables(PlannerInfo *root);
+extern Bitmapset *translate_col_privs(const Bitmapset *parent_privs,
+					List *translated_vars);
 
 extern Node *adjust_appendrel_attrs(PlannerInfo *root, Node *node,
 					   int nappinfos, AppendRelInfo **appinfos);
diff --git a/src/include/utils/lsyscache.h b/src/include/utils/lsyscache.h
index 07208b56ce..b5b615a6fa 100644
--- a/src/include/utils/lsyscache.h
+++ b/src/include/utils/lsyscache.h
@@ -126,8 +126,10 @@ extern char *get_rel_name(Oid relid);
 extern Oid	get_rel_namespace(Oid relid);
 extern Oid	get_rel_type_id(Oid relid);
 extern char get_rel_relkind(Oid relid);
+extern bool rel_is_partition(Oid relid);
 extern Oid	get_rel_tablespace(Oid relid);
 extern char get_rel_persistence(Oid relid);
+extern bool rel_is_other_temp(Oid relid);
 extern Oid	get_transform_fromsql(Oid typid, Oid langid, List *trftypes);
 extern Oid	get_transform_tosql(Oid typid, Oid langid, List *trftypes);
 extern bool get_typisdefined(Oid typid);
diff --git a/src/test/regress/expected/insert.out b/src/test/regress/expected/insert.out
index a2d9469592..e159d62b66 100644
--- a/src/test/regress/expected/insert.out
+++ b/src/test/regress/expected/insert.out
@@ -278,12 +278,12 @@ select tableoid::regclass, * from list_parted;
 -------------+----+----
  part_aa_bb  | aA |   
  part_cc_dd  | cC |  1
- part_null   |    |  0
- part_null   |    |  1
  part_ee_ff1 | ff |  1
  part_ee_ff1 | EE |  1
  part_ee_ff2 | ff | 11
  part_ee_ff2 | EE | 10
+ part_null   |    |  0
+ part_null   |    |  1
 (8 rows)
 
 -- some more tests to exercise tuple-routing with multi-level partitioning
-- 
2.11.0

From f698fe0fe5a9454691304716711d08820e16e163 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/6] WIP: Interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 123 +++++++++++++++++++++++++++++-----------
 1 file changed, 90 insertions(+), 33 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index a193f02551..ed85eafc32 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -105,6 +105,30 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the user-defined
+ * partition bound of a given existing partition, while an instance of the
+ * following struct describes either a new partition bound being compared
+ * against existing bounds (is_bound is true in that case and either lbound
+ * or rbound is set), or a new tuple's partition key specified in datums
+ * (ndatums = number of partition key columns).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -131,14 +155,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 
 /*
  * RelationBuildPartitionDesc
@@ -684,10 +709,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -738,6 +769,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo && boundinfo->ndatums > 0 &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE);
@@ -757,8 +789,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -772,9 +807,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -2019,9 +2054,14 @@ get_partition_for_tuple(PartitionTupleRoutingInfo **ptrinfos,
 		{
 			/* Else bsearch in partdesc->boundinfo */
 			bool		equal = false;
+			PartitionBoundCmpArg	arg;
 
+			memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+			arg.is_bound = false;
+			arg.datums = values;
+			arg.ndatums = key->partnatts;
 			cur_offset = partition_bound_bsearch(key, partdesc->boundinfo,
-												 values, false, &equal);
+												 &arg, &equal);
 			switch (key->strategy)
 			{
 				case PARTITION_STRATEGY_LIST:
@@ -2219,12 +2259,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -2246,11 +2286,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -2258,17 +2298,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -2279,12 +2337,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -2298,20 +2357,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -2324,8 +2382,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From a7f5054cd9697e5f183ed5d0b674fa7235800bcb Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 5/6] WIP: partition.c interface additions for
 partition-pruning

Add new arguments to get_partitions_for_keys() to allow the caller
to specify bounding scan keys, along with other information about
the scan keys extracted from the query, such as NULL-ness of the
keys, inclusive-ness, etc.

Query planner side (the caller) still doesn't pass anything for
the new arguments of get_partitions_for_keys.  It will, once the
logic to extract the relevant scan keys will be implemented in
a subsequent commit.

...still using constraint exclusion...
---
 src/backend/catalog/partition.c       | 203 ++++++++++++++++++++++++++++++++--
 src/backend/optimizer/path/allpaths.c |  10 +-
 src/include/catalog/partition.h       |   5 +-
 3 files changed, 208 insertions(+), 10 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index ed85eafc32..afb85cbc37 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1181,17 +1181,204 @@ RelationGetPartitionDispatchInfo(Relation rel,
  *		Returns the list of indexes (from pd->indexes) of the partitions that
  *		will need to be scanned for the given scan keys.
  *
- * TODO: add the interface to pass the query scan keys and the logic to look
- * up partitions using those keys.
+ * minkeys represents the lower bound on the partition key of the records that
+ * the query will return, while maxkeys represents the upper bound.
  */
 List *
-get_partitions_for_keys(PartitionDispatch pd)
+get_partitions_for_keys(PartitionDispatch pd,
+						bool *key_is_null,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive)
 {
-	int		i;
+	int		i,
+			minoff,
+			minidx = -1,
+			maxoff,
+			maxidx = -1;
 	List   *result = NIL;
+	PartitionKey	partkey = pd->key;
+	PartitionDesc	partdesc = pd->partdesc;
+	PartitionBoundCmpArg	arg;
+	bool		is_equal;
 
-	for (i = 0; i < pd->partdesc->nparts; i++)
-		result = lappend_int(result, pd->indexes[i]);
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if partdesc->boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (key_is_null[i] && partdesc->boundinfo->null_index < 0)
+		{
+			minidx = INT_MAX;
+			maxidx = INT_MIN;
+			goto generate_partition_list;
+		}
+		else if (key_is_null[i])
+		{
+			minidx = maxidx = partdesc->boundinfo->null_index;
+			goto generate_partition_list;
+		}
+	}
+
+	if (n_minkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = minkeys;
+		arg.ndatums = n_minkeys;
+		minoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (min_inclusive)
+					minoff -= 1;
+				else
+					minoff += 1;
+				if (minoff < 0 ||
+					minoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+												 minoff, &arg);
+			} while (cmpval == 0);
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				if (is_equal)
+					minidx = partdesc->boundinfo->indexes[minoff];
+				else
+					minidx = INT_MAX;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Records returned by the query will be > bounds[minoff],
+				 * because min_scankey is >= bounds[minoff], that is, no
+				 * records of the partition at minoff will be returned.  Go
+				 * to the next bound.
+				 */
+				if (minoff < partdesc->boundinfo->ndatums - 1)
+					minoff += 1;
+
+				/*
+				 * Make sure to skip a gap.
+				 * Note: There are ndatums + 1 lots in the indexes array.
+				 */
+				if (partdesc->boundinfo->indexes[minoff] < 0 &&
+					partdesc->boundinfo->indexes[minoff + 1] >= 0)
+					minoff += 1;
+
+				/*
+				 * Make sure we return a valid partition's index.  It's
+				 * possible that no valid partition exists, that is, all
+				 * partition bounds were <= min_scankey.
+				 */
+				if (partdesc->boundinfo->indexes[minoff] < 0)
+					minidx = INT_MAX;
+				else
+					minidx = partdesc->boundinfo->indexes[minoff];
+				break;
+		}
+	}
+	else
+		minidx = 0;
+
+	if (n_maxkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = maxkeys;
+		arg.ndatums = n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (max_inclusive)
+					maxoff += 1;
+				else
+					maxoff -= 1;
+				if (maxoff < 0 ||
+					maxoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+											 maxoff, &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (max_inclusive)
+				maxoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				if (is_equal)
+					maxidx = partdesc->boundinfo->indexes[maxoff];
+				else
+					maxidx = INT_MIN;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Because bounds[maxoff] <= max_scankey, we may need to
+				 * to consider the next partition as well, in addition to
+				 * the partition at maxoff and earlier.
+				 */
+				if (!is_equal || max_inclusive)
+					maxoff += 1;
+
+				/* Make sure to skip a gap. */
+				if (partdesc->boundinfo->indexes[maxoff] < 0 && maxoff >= 1)
+					maxoff -= 1;
+
+				/*
+				 * Make sure we return a valid partition's index.  It's
+				 * possible that no valid partition exists, that is, all
+				 * partition bounds were > max_scankey.
+				 */
+				if (partdesc->boundinfo->indexes[maxoff] < 0)
+					maxidx = INT_MIN;
+				else
+					maxidx = partdesc->boundinfo->indexes[maxoff];
+
+				break;
+		}
+	}
+	else
+		maxidx = partdesc->nparts - 1;
+
+generate_partition_list:
+	for (i = minidx; i <= maxidx; i++)
+	{
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Multiple values may belong to the same partition, so make
+				 * sure we don't add the same partition index again.
+				 */
+				result = list_append_unique_int(result, pd->indexes[i]);
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				result = lappend_int(result, pd->indexes[i]);
+				break;
+		}
+	}
 
 	return result;
 }
@@ -2253,8 +2440,8 @@ partition_rbound_cmp(PartitionKey key,
 /*
  * partition_rbound_datum_cmp
  *
- * Return whether range bound (specified in rb_datums, rb_kind, and rb_lower)
- * is <, =, or > partition key of tuple (tuple_datums)
+ * Return whether range bound (specified in rb_datums, rb_kind) is <, =, or >
+ * partition key of tuple (tuple_datums)
  */
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index c5c50e3b9d..a5e217674b 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -869,6 +869,9 @@ get_rel_partitions_recurse(RelOptInfo *rootrel,
 	List   *result = NIL,
 		   *my_live_partitions = NIL;
 	ListCell *l;
+	bool	keyisnull[PARTITION_MAX_KEYS];
+	Datum	minkeys[PARTITION_MAX_KEYS],
+			maxkeys[PARTITION_MAX_KEYS];
 
 	/*
 	 * Create a PartitionAppendInfo to map this table to the child tables
@@ -885,9 +888,14 @@ get_rel_partitions_recurse(RelOptInfo *rootrel,
 	 * TODO: collect the keys by looking at the clauses in
 	 * rootrel->baserestrictinfo considering this table's partition keys.
 	 */
+	memset(keyisnull, false, sizeof(keyisnull));
+	memset(minkeys, 0, sizeof(minkeys));
+	memset(maxkeys, 0, sizeof(maxkeys));
 
 	/* Ask partition.c which partitions it thinks match the keys. */
-	indexes = get_partitions_for_keys(partinfo->pd);
+	indexes = get_partitions_for_keys(partinfo->pd, keyisnull,
+									  minkeys, 0, false,
+									  maxkeys, 0, false);
 
 	/* Collect leaf partitions in the result list and recurse for others. */
 	foreach(l, indexes)
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 20fc3a89db..fb15498f92 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -96,5 +96,8 @@ extern int get_partition_for_tuple(PartitionTupleRoutingInfo **ptrinfos,
 						TupleTableSlot **failed_slot);
 
 /* Planner support stuff. */
-extern List *get_partitions_for_keys(PartitionDispatch pd);
+extern List *get_partitions_for_keys(PartitionDispatch pd,
+						bool *key_is_null,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive);
 #endif							/* PARTITION_H */
-- 
2.11.0

From edd4e4c951166ce26eff88f1d9f8e9d9a2b19624 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 17:31:42 +0900
Subject: [PATCH 6/6] WIP: planner-side changes for partition-pruning

This implements the planner-side logic to extract bounding scan keys
to be passed to get_partitions_for_keys.  That is it will go through
rel->baserestrictinfo and match individual clauses to partition keys
and construct lower bound and upper bound tuples, which may cover only
a prefix of a multi-column partition key.

A bunch of smarts are still missing when mapping the clause operands
with keys.  For example, code to match a clause is specifed as
(constant op var) doesn't exist.  Also, redundant keys are not
eliminated, for example, a combination of clauses a = 10 and a > 1
will cause the later clause a > 1 taking over and resulting in
needless scanning of partitions containing values a > 1 and a < 10.

...constraint exclusion is no longer used...
---
 src/backend/catalog/partition.c         |  57 +++++
 src/backend/nodes/outfuncs.c            |   2 +
 src/backend/optimizer/path/allpaths.c   | 157 ++++++++++++-
 src/backend/optimizer/prep/prepunion.c  |   9 +
 src/backend/optimizer/util/plancat.c    |   4 +
 src/include/catalog/partition.h         |   2 +
 src/include/nodes/relation.h            |   2 +
 src/test/regress/expected/partition.out | 375 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      |  65 ++++++
 11 files changed, 668 insertions(+), 8 deletions(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index afb85cbc37..8e57d36449 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1177,6 +1177,63 @@ RelationGetPartitionDispatchInfo(Relation rel,
 }
 
 /*
+ * get_partition_keys
+ *		Returns a list of expressions matching the partition key columns
+ */
+List *
+get_partition_keys(PartitionDispatch pd, int varno)
+{
+	int		i;
+	PartitionKey	key = pd->key;
+	List   *result = NIL;
+	ListCell *lc;
+
+	lc = list_head(key->partexprs);
+	for (i = 0; i < key->partnatts; i++)
+	{
+		Expr *keyCol;
+
+		if (key->partattrs[i] != 0)
+		{
+			keyCol = (Expr *) makeVar(varno,
+									  key->partattrs[i],
+									  key->parttypid[i],
+									  key->parttypmod[i],
+									  key->parttypcoll[i],
+									  0);
+		}
+		else
+		{
+			if (lc == NULL)
+				elog(ERROR, "wrong number of partition key expressions");
+			keyCol = (Expr *) copyObject(lfirst(lc));
+			lc = lnext(lc);
+		}
+
+		result = lappend(result, keyCol);
+	}
+
+	return result;
+}
+
+/*
+ * get_partition_opfamilies
+ *		Get partitioning operator family OIDs for all keys
+ */
+List *
+get_partition_opfamilies(PartitionDispatch pd)
+{
+	List   *result = NIL;
+	PartitionKey key = pd->key;
+	int		i;
+
+	for (i = 0; i < key->partnatts; i++)
+		result = lappend_oid(result, key->partopfamily[i]);
+
+	return result;
+}
+
+/*
  * get_partitions_for_keys
  *		Returns the list of indexes (from pd->indexes) of the partitions that
  *		will need to be scanned for the given scan keys.
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index 2480fd6429..4d092489ba 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -2524,6 +2524,8 @@ _outPartitionInfo(StringInfo str, const PartitionInfo *node)
 
 	WRITE_BOOL_FIELD(is_other_temp);
 	WRITE_UINT_FIELD(relid);
+	WRITE_NODE_FIELD(keys);
+	WRITE_NODE_FIELD(keyopfamilies);
 	/* Don't bother writing out the PartitionDispatch object */
 }
 
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index a5e217674b..a55ede2faa 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -868,10 +868,24 @@ get_rel_partitions_recurse(RelOptInfo *rootrel,
 	List   *indexes;
 	List   *result = NIL,
 		   *my_live_partitions = NIL;
-	ListCell *l;
+	ListCell   *lc1,
+			   *lc2,
+			   *keyopfamilies_item;
+	int		keyPos;
+	List   *matchedclauses[PARTITION_MAX_KEYS];
 	bool	keyisnull[PARTITION_MAX_KEYS];
 	Datum	minkeys[PARTITION_MAX_KEYS],
 			maxkeys[PARTITION_MAX_KEYS];
+	bool	need_next_min,
+			need_next_max,
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_partkeys = list_length(partinfo->keys),
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			i;
 
 	/*
 	 * Create a PartitionAppendInfo to map this table to the child tables
@@ -885,22 +899,151 @@ get_rel_partitions_recurse(RelOptInfo *rootrel,
 											  painfo);
 
 	/*
-	 * TODO: collect the keys by looking at the clauses in
-	 * rootrel->baserestrictinfo considering this table's partition keys.
+	 * Match individual OpExprs in the query's restriction with individual
+	 * partition key columns.  There is one list per key.
 	 */
+	memset(matchedclauses, 0, sizeof(matchedclauses));
 	memset(keyisnull, false, sizeof(keyisnull));
+	keyPos = 0;
+	foreach(lc1, partinfo->keys)
+	{
+		Node   *partkey = lfirst(lc1);
+
+		foreach(lc2, rootrel->baserestrictinfo)
+		{
+			RestrictInfo   *rinfo = lfirst(lc2);
+			Expr		   *clause = rinfo->clause;
+
+			if (is_opclause(clause))
+			{
+				Node   *leftop = get_leftop(clause);
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+
+				if (equal(leftop, partkey))
+					matchedclauses[keyPos] = lappend(matchedclauses[keyPos],
+													 clause);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey) && nulltest->nulltesttype == IS_NULL)
+					keyisnull[keyPos] = true;
+			}
+		}
+
+		/* Onto finding clauses matching the next partition key. */
+		keyPos++;
+	}
+
+	/*
+	 * Determine the min keys and the max keys using btree semantics-based
+	 * interpretation of the clauses' operators.
+	 */
+
+	/*
+	 * XXX - There should be a step similar to _bt_preprocess_keys() here,
+	 * to eliminate any redundant scan keys for a given partition column.  For
+	 * example, among a <= 4 and a <= 5, we can only keep a <= 4 for being
+	 * more restrictive and discard a <= 5.  While doing that, we can also
+	 * check to see if there exists a contradictory combination of scan keys
+	 * that makes the query trivially false for all records in the table.
+	 */
+
 	memset(minkeys, 0, sizeof(minkeys));
 	memset(maxkeys, 0, sizeof(maxkeys));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+	need_next_min = true;
+	need_next_max = true;
+	keyopfamilies_item = list_head(partinfo->keyopfamilies);
+	for (i = 0; i < n_partkeys; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc1, matchedclauses[i])
+		{
+			Expr   *clause = lfirst(lc1);
+			Const  *rightop = (Const *) get_rightop(clause);
+			Oid		opno = ((OpExpr *) clause)->opno,
+					opfamily = lfirst_oid(keyopfamilies_item);
+			StrategyNumber	strategy;
+
+			strategy = get_op_opfamily_strategy(opno, opfamily);
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+					}
+					if (strategy == BTLessStrategyNumber)
+						need_next_max = false;
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+					}
+					if (strategy == BTGreaterStrategyNumber)
+						need_next_min = false;
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+					}
+					minkey_set[i] = true;
+					min_incl = true;
+
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+					}
+					maxkey_set[i] = true;
+					max_incl = true;
+					break;
+			}
+		}
+
+		keyopfamilies_item = lnext(keyopfamilies_item);
+	}
 
 	/* Ask partition.c which partitions it thinks match the keys. */
 	indexes = get_partitions_for_keys(partinfo->pd, keyisnull,
-									  minkeys, 0, false,
-									  maxkeys, 0, false);
+									  minkeys, n_minkeys, min_incl,
+									  maxkeys, n_maxkeys, max_incl);
 
 	/* Collect leaf partitions in the result list and recurse for others. */
-	foreach(l, indexes)
+	foreach(lc1, indexes)
 	{
-		int		index = lfirst_int(l);
+		int		index = lfirst_int(lc1);
 
 		if (index >= 0)
 		{
diff --git a/src/backend/optimizer/prep/prepunion.c b/src/backend/optimizer/prep/prepunion.c
index 01de2d778d..e5c60020b7 100644
--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -1509,6 +1509,8 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 		pinfo = makeNode(PartitionInfo);
 		pinfo->relid = rti;
 		pinfo->pd = pds[0];
+		pinfo->keys = get_partition_keys(pinfo->pd, rti);
+		pinfo->keyopfamilies = get_partition_opfamilies(pinfo->pd);
 		partition_infos = list_make1(pinfo);
 		partitioned_child_rels = list_make1_int(rti);
 		for (i = 1; i < num_parted; i++)
@@ -1617,6 +1619,13 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 			pinfo->is_other_temp = is_other_temp;
 			pinfo->relid = childRTindex;
 			pinfo->pd = pds[i++];
+
+			/* Convert so that expression contains oldrelation's attnos. */
+			pinfo->keys =
+				map_partition_varattnos(get_partition_keys(pinfo->pd, rti),
+										rti, oldrelation, pinfo->pd->reldesc,
+										NULL);
+			pinfo->keyopfamilies = get_partition_opfamilies(pinfo->pd);
 			partition_infos = lappend(partition_infos, pinfo);
 
 			partitioned_child_rels = lappend_int(partitioned_child_rels,
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 3781a91b76..fdcb77b16f 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1149,7 +1149,9 @@ get_relation_constraints(PlannerInfo *root,
 	Index		varno = rel->relid;
 	Relation	relation;
 	TupleConstr *constr;
+#ifdef	USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	List	   *pcqual;
+#endif
 
 	/*
 	 * We assume the relation has already been safely locked.
@@ -1235,6 +1237,7 @@ get_relation_constraints(PlannerInfo *root,
 		}
 	}
 
+#ifdef USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	/* Append partition predicates, if any */
 	pcqual = RelationGetPartitionQual(relation);
 	if (pcqual)
@@ -1252,6 +1255,7 @@ get_relation_constraints(PlannerInfo *root,
 
 		result = list_concat(result, pcqual);
 	}
+#endif
 
 	heap_close(relation, NoLock);
 
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index fb15498f92..d9ca8d8371 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -96,6 +96,8 @@ extern int get_partition_for_tuple(PartitionTupleRoutingInfo **ptrinfos,
 						TupleTableSlot **failed_slot);
 
 /* Planner support stuff. */
+extern List *get_partition_keys(PartitionDispatch pd, int varno);
+extern List *get_partition_opfamilies(PartitionDispatch pd);
 extern List *get_partitions_for_keys(PartitionDispatch pd,
 						bool *key_is_null,
 						Datum *minkeys, int n_minkeys, bool min_inclusive,
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index a67a43b069..0f9bcd81ed 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -2045,6 +2045,8 @@ typedef struct PartitionInfo
 
 	bool		is_other_temp;		/* If true, ignore the following fields */
 	Index				relid;		/* Ordinal position in the rangetable */
+	List			   *keys;		/* Expressions for partition keys */
+	List			   *keyopfamilies;	/* Operator family OID per key */
 	PartitionDispatch	pd;			/* Information about partitions */
 } PartitionInfo;
 
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..0839923fca
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,375 @@
+--
+-- Test partitioning planner code
+--
+create table rlpt (a int, b varchar) partition by range (a);
+create table rlpt1 partition of rlpt for values from (minvalue) to (1);
+create table rlpt2 partition of rlpt for values from (1) to (10);
+create table rlpt3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlpt3abcd partition of rlpt3 for values in ('ab', 'cd');
+create table rlpt3efgh partition of rlpt3 for values in ('ef', 'gh');
+create table rlpt3nullxy partition of rlpt3 for values in (null, 'xy');
+alter table rlpt attach partition rlpt3 for values from (15) to (20);
+create table rlpt4 partition of rlpt for values from (20) to (30);
+create table rlpt5 partition of rlpt for values from (31) to (maxvalue);
+explain (costs off) select * from rlpt where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlpt1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlpt where a <= 1;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlpt1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlpt2
+         Filter: (a <= 1)
+(5 rows)
+
+explain (costs off) select * from rlpt where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlpt2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlpt where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlpt2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlpt where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlpt1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlpt2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlpt3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlpt3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlpt3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlpt4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlpt5
+         Filter: ((a)::numeric = '1'::numeric)
+(15 rows)
+
+explain (costs off) select * from rlpt where a <= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on rlpt1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlpt2
+         Filter: (a <= 10)
+(5 rows)
+
+explain (costs off) select * from rlpt where a > 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlpt3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlpt3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlpt3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlpt4
+         Filter: (a > 10)
+   ->  Seq Scan on rlpt5
+         Filter: (a > 10)
+(11 rows)
+
+explain (costs off) select * from rlpt where a < 15;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlpt1
+         Filter: (a < 15)
+   ->  Seq Scan on rlpt2
+         Filter: (a < 15)
+(5 rows)
+
+explain (costs off) select * from rlpt where a <= 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlpt1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlpt2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlpt3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlpt3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlpt3nullxy
+         Filter: (a <= 15)
+(11 rows)
+
+explain (costs off) select * from rlpt where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlpt3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlpt4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlpt5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(7 rows)
+
+explain (costs off) select * from rlpt where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlpt3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlpt where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlpt3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlpt3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlpt3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(7 rows)
+
+explain (costs off) select * from rlpt where a is null;		/* while we're on nulls */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlpt where a > 30;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlpt5
+         Filter: (a > 30)
+(3 rows)
+
+explain (costs off) select * from rlpt where a = 30;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlpt where a <= 31;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlpt1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlpt2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlpt3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlpt3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlpt3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlpt4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlpt5
+         Filter: (a <= 31)
+(15 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, 0, 0) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, 0, 0);
+explain (costs off) select * from mc3p where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(9 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+(15 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+(5 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+(7 rows)
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p3
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p4
+         Filter: ((a > 1) AND (a = 10))
+(11 rows)
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlpt3 where b = 'ab' or b = 'ef';
+                               QUERY PLAN                               
+------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlpt3abcd
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlpt3efgh
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlpt3nullxy
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+(7 rows)
+
+drop table rlpt, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index eefdeeacae..e5089a7cee 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 76b0de30a7..6611662149 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..8ffe91b08f
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,65 @@
+--
+-- Test partitioning planner code
+--
+create table rlpt (a int, b varchar) partition by range (a);
+create table rlpt1 partition of rlpt for values from (minvalue) to (1);
+create table rlpt2 partition of rlpt for values from (1) to (10);
+
+create table rlpt3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlpt3abcd partition of rlpt3 for values in ('ab', 'cd');
+create table rlpt3efgh partition of rlpt3 for values in ('ef', 'gh');
+create table rlpt3nullxy partition of rlpt3 for values in (null, 'xy');
+alter table rlpt attach partition rlpt3 for values from (15) to (20);
+
+create table rlpt4 partition of rlpt for values from (20) to (30);
+create table rlpt5 partition of rlpt for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlpt where a < 1;
+explain (costs off) select * from rlpt where a <= 1;
+explain (costs off) select * from rlpt where a = 1;
+explain (costs off) select * from rlpt where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlpt where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlpt where a <= 10;
+explain (costs off) select * from rlpt where a > 10;
+explain (costs off) select * from rlpt where a < 15;
+explain (costs off) select * from rlpt where a <= 15;
+explain (costs off) select * from rlpt where a > 15 and b = 'ab';
+explain (costs off) select * from rlpt where a = 16 and b is null;
+explain (costs off) select * from rlpt where a = 16 and b is not null;
+explain (costs off) select * from rlpt where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlpt where a > 30;
+explain (costs off) select * from rlpt where a = 30;	/* empty */
+explain (costs off) select * from rlpt where a <= 31;
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, 0, 0) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, 0, 0);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlpt3 where b = 'ab' or b = 'ef';
+
+drop table rlpt, mc3p;
-- 
2.11.0

From 17dfaff62fe04cf18f5bba298974d42f92b597ef Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 6 Sep 2017 09:28:14 +0900
Subject: [PATCH 1/5] Expand partitioned inheritance in a non-flattened manner

...except when the partitioned table in question is the result rel
of the query.

This allows us perform scan and join planning for each sub-tree in a
given partition tree, with each sub-tree's root partitioned table
getting its own RelOptInfo.  Previously only the root of the whole
partition tree got a RelOptInfo, along with the leaf partitions,
with each leaf partition's AppendRelInfo pointing to the former as
its parent.
---
 src/backend/optimizer/path/allpaths.c  |  34 ++++++-
 src/backend/optimizer/plan/planner.c   |   3 +-
 src/backend/optimizer/prep/prepunion.c | 166 +++++++++++++++++++++------------
 src/backend/optimizer/util/plancat.c   |   7 +-
 4 files changed, 146 insertions(+), 64 deletions(-)

diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 2d7e1d84d0..6c3511bd47 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -1289,11 +1289,39 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	RangeTblEntry *rte;
 
 	rte = planner_rt_fetch(rel->relid, root);
+
+	/*
+	 * Get the partitioned_rels list from root->pcinfo_list after
+	 * confirming that rel is actually a root partitioned table.
+	 */
 	if (rte->relkind == RELKIND_PARTITIONED_TABLE)
 	{
-		partitioned_rels = get_partitioned_child_rels(root, rel->relid);
-		/* The root partitioned table is included as a child rel */
-		Assert(list_length(partitioned_rels) >= 1);
+		int		parent_relid;
+		bool	is_root_partitioned_table = false;
+
+		/*
+		 * Normally, only the root partitioned rel will be RELOPT_BASEREL
+		 * in a given partitione tree, except when the root table itself
+		 * is a child in the case of a UNION ALL query.
+		 */
+		if (!IS_OTHER_REL(rel))
+			is_root_partitioned_table = true;
+		else if (bms_get_singleton_member(rel->top_parent_relids,
+										  &parent_relid))
+		{
+			RelOptInfo *parent_rel;
+
+			parent_rel = root->simple_rel_array[parent_relid];
+			is_root_partitioned_table =
+						(parent_rel->rtekind != RTE_RELATION);
+		}
+
+		if (is_root_partitioned_table)
+		{
+			partitioned_rels = get_partitioned_child_rels(root, rel->relid);
+			/* The root partitioned table is included as a child rel */
+			Assert(list_length(partitioned_rels) >= 1);
+		}
 	}
 
 	/*
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index 966230256e..02662fad5d 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6076,7 +6076,8 @@ plan_cluster_use_sort(Oid tableOid, Oid indexOid)
  *		Returns a list of the RT indexes of the partitioned child relations
  *		with rti as the root parent RT index.
  *
- * Note: Only call this function on RTEs known to be partitioned tables.
+ * Note: Only call this function on RTEs known to be "root" partitioned
+ *		 tables.
  */
 List *
 get_partitioned_child_rels(PlannerInfo *root, Index rti)
diff --git a/src/backend/optimizer/prep/prepunion.c b/src/backend/optimizer/prep/prepunion.c
index ccf21453fd..433505948d 100644
--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -113,7 +113,8 @@ static void expand_single_inheritance_child(PlannerInfo *root,
 								Index parentRTindex, Relation parentrel,
 								PlanRowMark *parentrc, Relation childrel,
 								bool *has_child, List **appinfos,
-								List **partitioned_child_rels);
+								List **partitioned_child_rels,
+								RangeTblEntry **newrte, Index *newRTindex);
 static void make_inh_translation_list(Relation oldrelation,
 						  Relation newrelation,
 						  Index newvarno,
@@ -1380,8 +1381,8 @@ expand_inherited_tables(PlannerInfo *root)
  * regular inheritance, a parent RTE must always have at least two associated
  * AppendRelInfos: one corresponding to the parent table as a simple member of
  * inheritance set and one or more corresponding to the actual children.
- * Since a partitioned table is not scanned, it might have only one associated
- * AppendRelInfo.
+ * Since a partitioned table parent is itself not scanned, it might have only
+ * one associated AppendRelInfo.
  */
 static void
 expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
@@ -1473,13 +1474,10 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 	{
 		/*
 		 * If this table has partitions, recursively expand them in the order
-		 * in which they appear in the PartitionDesc.  But first, expand the
-		 * parent itself.
+		 * in which they appear in the PartitionDesc.  Also, start collecting
+		 * the RT indexes of the partitioned tables in the partition tree.
 		 */
-		expand_single_inheritance_child(root, rte, rti, oldrelation, oldrc,
-										oldrelation,
-										&has_child, &appinfos,
-										&partitioned_child_rels);
+		partitioned_child_rels = list_make1_int(rti);
 		expand_partitioned_rtentry(root, rte, rti, oldrelation, oldrc,
 									  RelationGetPartitionDesc(oldrelation),
 									  lockmode,
@@ -1516,10 +1514,14 @@ expand_inherited_rtentry(PlannerInfo *root, RangeTblEntry *rte, Index rti)
 				continue;
 			}
 
+			/*
+			 * Don't expect to find any partitioned tables in a regular
+			 * inheritance tree, so pass NULL for partitioned_child_rels here.
+			 */
 			expand_single_inheritance_child(root, rte, rti, oldrelation, oldrc,
 											newrelation,
 											&has_child, &appinfos,
-											&partitioned_child_rels);
+											NULL, NULL, NULL);
 
 			/* Close child relations, but keep locks */
 			if (childOID != parentOID)
@@ -1581,6 +1583,8 @@ expand_partitioned_rtentry(PlannerInfo *root, RangeTblEntry *parentrte,
 	{
 		Oid			childOID = partdesc->oids[i];
 		Relation	childrel;
+		RangeTblEntry *childrte;
+		Index		childRTindex;
 
 		/* Open rel; we already have required locks */
 		childrel = heap_open(childOID, NoLock);
@@ -1592,19 +1596,60 @@ expand_partitioned_rtentry(PlannerInfo *root, RangeTblEntry *parentrte,
 			continue;
 		}
 
-		expand_single_inheritance_child(root, parentrte, parentRTindex,
+		expand_single_inheritance_child(root,
+										parentrte, parentRTindex,
 										parentrel, parentrc, childrel,
 										has_child, appinfos,
-										partitioned_child_rels);
+										partitioned_child_rels,
+										&childrte, &childRTindex);
 
 		/* If this child is itself partitioned, recurse */
 		if (childrel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
-			expand_partitioned_rtentry(root, parentrte, parentRTindex,
-										  parentrel, parentrc,
-										  RelationGetPartitionDesc(childrel),
-										  lockmode,
-										  has_child, appinfos,
-										  partitioned_child_rels);
+		{
+			RangeTblEntry *new_parentrte;
+			Index		new_parentRTindex;
+			Relation	new_parentrel;
+
+			/*
+			 * For SELECT queries, it's desirable to perform scan and join
+			 * planning on the individual partition sub-trees, instead of
+			 * doing the same on the whole tree at once.  This allows to apply
+			 * techniques such as parition-pruning and/or partition-wise join
+			 * on the individual partition sub-trees.  For that to happen,
+			 * root parent of each sub-tree must get an RTE with inh set to
+			 * true, which must be already taken care of by
+			 * expand_single_inheritance_child().  Next, for each of the
+			 * children, we must record immediate parent as its parent in the
+			 * the child AppendRelInfo, instead of the root parent of the
+			 * whole tree.
+			 *
+			 * If parent is the query's result relation, inheritance_planner()
+			 * will expand the inheritance so as to apply the *whole* query to
+			 * each leaf partition, which means we cannot apply
+			 * partition-pruning and/or partition-wise join to a partitioned
+			 * result relation, meaning there is not much point in expanding
+			 * the tree hierarchically.
+			 */
+			if (parentRTindex == root->parse->resultRelation)
+			{
+				new_parentrte = parentrte;
+				new_parentRTindex = parentRTindex;
+				new_parentrel = parentrel;
+			}
+			else
+			{
+				new_parentrte = childrte;
+				new_parentRTindex = childRTindex;
+				new_parentrel = childrel;
+			}
+
+			expand_partitioned_rtentry(root, new_parentrte, new_parentRTindex,
+									   new_parentrel, parentrc,
+									   RelationGetPartitionDesc(childrel),
+									   lockmode,
+									   has_child, appinfos,
+									   partitioned_child_rels);
+		}
 
 		/* Close child relation, but keep locks */
 		heap_close(childrel, NoLock);
@@ -1619,13 +1664,17 @@ expand_partitioned_rtentry(PlannerInfo *root, RangeTblEntry *parentrte,
  * anything at all.  Otherwise, we'll set "has_child" to true, build a
  * RangeTblEntry and either a PartitionedChildRelInfo or AppendRelInfo as
  * appropriate, plus maybe a PlanRowMark.
+ *
+ * The newly created RT entry and its RT index are returned in *newrte and
+ * *newRTindex, respectively.
  */
 static void
 expand_single_inheritance_child(PlannerInfo *root, RangeTblEntry *parentrte,
 								Index parentRTindex, Relation parentrel,
 								PlanRowMark *parentrc, Relation childrel,
 								bool *has_child, List **appinfos,
-								List **partitioned_child_rels)
+								List **partitioned_child_rels,
+								RangeTblEntry **newrte, Index *newRTindex)
 {
 	Query	   *parse = root->parse;
 	Oid			parentOID = RelationGetRelid(parentrel);
@@ -1649,54 +1698,46 @@ expand_single_inheritance_child(PlannerInfo *root, RangeTblEntry *parentrte,
 	childrte = copyObject(parentrte);
 	childrte->relid = childOID;
 	childrte->relkind = childrel->rd_rel->relkind;
-	childrte->inh = false;
+	childrte->inh = (childrte->relkind == RELKIND_PARTITIONED_TABLE);
 	childrte->requiredPerms = 0;
 	childrte->securityQuals = NIL;
 	parse->rtable = lappend(parse->rtable, childrte);
 	childRTindex = list_length(parse->rtable);
 
+	/* Build an AppendRelInfo for this parent and child. */
+
+	/* Remember if we saw a real child. */
+	if (childOID != parentOID)
+		*has_child = true;
+
+	appinfo = makeNode(AppendRelInfo);
+	appinfo->parent_relid = parentRTindex;
+	appinfo->child_relid = childRTindex;
+	appinfo->parent_reltype = parentrel->rd_rel->reltype;
+	appinfo->child_reltype = childrel->rd_rel->reltype;
+	make_inh_translation_list(parentrel, childrel, childRTindex,
+							  &appinfo->translated_vars);
+	appinfo->parent_reloid = parentOID;
+	*appinfos = lappend(*appinfos, appinfo);
+
 	/*
-	 * Build an AppendRelInfo for this parent and child, unless the child is a
-	 * partitioned table.
+	 * Translate the column permissions bitmaps to the child's attnums (we
+	 * have to build the translated_vars list before we can do this). But
+	 * if this is the parent table, leave copyObject's result alone.
+	 *
+	 * Note: we need to do this even though the executor won't run any
+	 * permissions checks on the child RTE.  The insertedCols/updatedCols
+	 * bitmaps may be examined for trigger-firing purposes.
 	 */
-	if (childrte->relkind != RELKIND_PARTITIONED_TABLE)
+	if (childOID != parentOID)
 	{
-		/* Remember if we saw a real child. */
-		if (childOID != parentOID)
-			*has_child = true;
-
-		appinfo = makeNode(AppendRelInfo);
-		appinfo->parent_relid = parentRTindex;
-		appinfo->child_relid = childRTindex;
-		appinfo->parent_reltype = parentrel->rd_rel->reltype;
-		appinfo->child_reltype = childrel->rd_rel->reltype;
-		make_inh_translation_list(parentrel, childrel, childRTindex,
-								  &appinfo->translated_vars);
-		appinfo->parent_reloid = parentOID;
-		*appinfos = lappend(*appinfos, appinfo);
-
-		/*
-		 * Translate the column permissions bitmaps to the child's attnums (we
-		 * have to build the translated_vars list before we can do this). But
-		 * if this is the parent table, leave copyObject's result alone.
-		 *
-		 * Note: we need to do this even though the executor won't run any
-		 * permissions checks on the child RTE.  The insertedCols/updatedCols
-		 * bitmaps may be examined for trigger-firing purposes.
-		 */
-		if (childOID != parentOID)
-		{
-			childrte->selectedCols = translate_col_privs(parentrte->selectedCols,
-														 appinfo->translated_vars);
-			childrte->insertedCols = translate_col_privs(parentrte->insertedCols,
-														 appinfo->translated_vars);
-			childrte->updatedCols = translate_col_privs(parentrte->updatedCols,
-														appinfo->translated_vars);
-		}
+		childrte->selectedCols = translate_col_privs(parentrte->selectedCols,
+													 appinfo->translated_vars);
+		childrte->insertedCols = translate_col_privs(parentrte->insertedCols,
+													 appinfo->translated_vars);
+		childrte->updatedCols = translate_col_privs(parentrte->updatedCols,
+													appinfo->translated_vars);
 	}
-	else
-		*partitioned_child_rels = lappend_int(*partitioned_child_rels,
-											  childRTindex);
 
 	/*
 	 * Build a PlanRowMark if parent is marked FOR UPDATE/SHARE.
@@ -1726,6 +1767,15 @@ expand_single_inheritance_child(PlannerInfo *root, RangeTblEntry *parentrte,
 
 		root->rowMarks = lappend(root->rowMarks, childrc);
 	}
+
+	if (partitioned_child_rels &&
+		childrte->relkind == RELKIND_PARTITIONED_TABLE)
+		*partitioned_child_rels = lappend_int(*partitioned_child_rels,
+											  childRTindex);
+	if (newrte)
+		*newrte = childrte;
+	if (newRTindex)
+		*newRTindex = childRTindex;
 }
 
 /*
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index a1ebd4acc8..bfc05a1af5 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1402,8 +1402,11 @@ relation_excluded_by_constraints(PlannerInfo *root,
 	if (predicate_refuted_by(safe_restrictions, safe_restrictions, false))
 		return true;
 
-	/* Only plain relations have constraints */
-	if (rte->rtekind != RTE_RELATION || rte->inh)
+	/*
+	 * Only plain relations have constraints.  In addition, there can be
+	 * inheritance parent RTEs that are themselves partitions.
+	 */
+	if (rte->rtekind != RTE_RELATION || (rte->inh && !IS_OTHER_REL(rel)))
 		return false;
 
 	/*
-- 
2.11.0

From 0f1d944c23c0f9170ffe8553ef2d22754fa3aab7 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 17:31:42 +0900
Subject: [PATCH 2/5] WIP: planner-side changes for partition-pruning

Firstly, this adds a stub get_partitions_for_keys() in partition.c
with appropriate interface for the caller to specify bounding scan
keys, along with other information about the scan keys extracted
from the query, such as NULL-ness of the keys, inclusive-ness, etc.

More importantly, this implements the planner-side logic to extract
bounding scan keys to be passed to get_partitions_for_keys.  That is,
it will go through rel->baserestrictinfo and match individual clauses
to partition keys and construct lower bound and upper bound tuples,
which may cover only a prefix of a multi-column partition key.

A bunch of smarts are still missing when mapping the clause operands
with keys.  For example, code to match a clause is specifed as
(constant op var) doesn't exist.  Also, redundant keys are not
eliminated, for example, a combination of clauses a = 10 and a > 1
will cause the later clause a > 1 taking over and resulting in
needless scanning of partitions containing values a > 1 and a < 10.

...constraint exclusion is still used, because
get_partitions_for_keys is just a stub...
---
 src/backend/catalog/partition.c       |  42 +++++
 src/backend/optimizer/path/allpaths.c | 308 +++++++++++++++++++++++++++++-----
 src/backend/optimizer/util/plancat.c  | 120 +++++++++++++
 src/backend/optimizer/util/relnode.c  |  20 +++
 src/include/catalog/partition.h       |   8 +
 src/include/nodes/nodes.h             |   1 +
 src/include/nodes/relation.h          | 135 +++++++++++++++
 src/include/optimizer/plancat.h       |   2 +
 8 files changed, 595 insertions(+), 41 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 50162632f5..bb3009e5b3 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1138,6 +1138,48 @@ RelationGetPartitionDispatchInfo(Relation rel,
 	return pd;
 }
 
+/*
+ * get_partitions_for_keys
+ *		Returns the list of indexes of rel's partitions that will need to be
+ *		scanned given the bounding scan keys.
+ *
+ * Each value in the returned list can be used as an index into the oids array
+ * of the partition descriptor.
+ *
+ * Inputs:
+ *	keynullness contains between 0 and (key->partnatts - 1) values, each
+ *	telling what kind of NullTest has been applies to the corresponding
+ *	partition key column.  minkeys represents the lower bound on the partition
+ *	the key of the records that the query will return, while maxkeys
+ *	represents upper bound.  min_inclusive and max_inclusive tell whether the
+ *	bounds specified minkeys and maxkeys is inclusive, respectively.
+ *
+ * Other outputs:
+ *	*min_datum_index will return the index in boundinfo->datums of the first
+ *	datum that the query's bounding keys allow to be returned for the query.
+ *	Similarly, *max_datum_index.  *null_partition_chosen returns whether
+ *	the null partition will be scanned.
+ *
+ * TODO: Implement.
+ */
+List *
+get_partitions_for_keys(Relation rel,
+						NullTestType *keynullness,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive,
+						int *min_datum_index, int *max_datum_index,
+						bool *null_partition_chosen)
+{
+	List   *result = NIL;
+	int		i;
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+
+	for (i = 0; i < partdesc->nparts; i++)
+		result = lappend_int(result, i);
+
+	return result;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 6c3511bd47..97af646242 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,6 +20,7 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
@@ -845,6 +846,222 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the query clauses
+ *
+ * Returned list contains the AppendInfos of the chosen partitions.
+ */
+static List *
+get_rel_partitions(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
+{
+	Relation		parent = heap_open(rte->relid, NoLock);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(parent);
+	List   *indexes;
+	List   *result = NIL;
+	ListCell   *lc1,
+			   *lc2;
+	int		keyPos;
+	List   *matchedclauses[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	Datum	minkeys[PARTITION_MAX_KEYS],
+			maxkeys[PARTITION_MAX_KEYS];
+	bool	need_next_min,
+			need_next_max,
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_minkeys = 0,
+			n_maxkeys = 0,
+			i;
+
+	/*
+	 * Match individual OpExprs in the query's restriction with individual
+	 * partition key columns.  There is one list per key.
+	 */
+	memset(keynullness, -1, sizeof(keynullness));
+	memset(matchedclauses, 0, sizeof(matchedclauses));
+	keyPos = 0;
+	for (i = 0; i < rel->part_scheme->partnatts; i++)
+	{
+		Node   *partkey = linitial(rel->partexprs[i]);
+
+		foreach(lc2, rel->baserestrictinfo)
+		{
+			RestrictInfo   *rinfo = lfirst(lc2);
+			Expr		   *clause = rinfo->clause;
+
+			if (is_opclause(clause))
+			{
+				Node   *leftop = get_leftop(clause);
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					matchedclauses[keyPos] = lappend(matchedclauses[keyPos],
+													 clause);
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[keyPos] = IS_NOT_NULL;
+				}
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					keynullness[keyPos] = nulltest->nulltesttype;
+			}
+		}
+
+		/* Onto finding clauses matching the next partition key. */
+		keyPos++;
+	}
+
+	/*
+	 * Determine the min keys and the max keys using btree semantics-based
+	 * interpretation of the clauses' operators.
+	 */
+
+	/*
+	 * XXX - There should be a step similar to _bt_preprocess_keys() here,
+	 * to eliminate any redundant scan keys for a given partition column.  For
+	 * example, among a <= 4 and a <= 5, we can only keep a <= 4 for being
+	 * more restrictive and discard a <= 5.  While doing that, we can also
+	 * check to see if there exists a contradictory combination of scan keys
+	 * that makes the query trivially false for all records in the table.
+	 */
+	memset(minkeys, 0, sizeof(minkeys));
+	memset(maxkeys, 0, sizeof(maxkeys));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < rel->part_scheme->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc1, matchedclauses[i])
+		{
+			Expr   *clause = lfirst(lc1);
+			Const  *rightop = (Const *) get_rightop(clause);
+			Oid		opno = ((OpExpr *) clause)->opno,
+					opfamily = rel->part_scheme->partopfamily[i];
+			StrategyNumber	strategy;
+
+			strategy = get_op_opfamily_strategy(opno, opfamily);
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+					}
+					if (strategy == BTLessStrategyNumber)
+						need_next_max = false;
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+					}
+					if (strategy == BTGreaterStrategyNumber)
+						need_next_min = false;
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+					}
+					minkey_set[i] = true;
+					min_incl = true;
+
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+					}
+					maxkey_set[i] = true;
+					max_incl = true;
+					break;
+
+				/*
+				 * This might mean '<>', but we don't have anything for that
+				 * case yet.  Perhaps, handle that as key < const OR
+				 * key > const, once we have props needed for handling OR
+				 * clauses.
+				 */
+				default:
+					min_incl = max_incl = false;
+					break;
+			}
+		}
+	}
+
+	/* Ask partition.c which partitions it thinks match the keys. */
+	indexes = get_partitions_for_keys(parent, keynullness,
+									  minkeys, n_minkeys, min_incl,
+									  maxkeys, n_maxkeys, max_incl,
+									  &rel->painfo->min_datum_idx,
+									  &rel->painfo->max_datum_idx,
+									  &rel->painfo->contains_null_partition);
+
+	if (indexes != NIL)
+	{
+#ifdef USE_ASSERT_CHECKING
+		int		first_index,
+				last_index;
+		first_index = linitial_int(indexes);
+		last_index = llast_int(indexes);
+		Assert(first_index <= last_index ||
+			   rel->part_scheme->strategy != PARTITION_STRATEGY_RANGE);
+#endif
+
+		foreach(lc1, indexes)
+		{
+			int		partidx = lfirst_int(lc1);
+			AppendRelInfo *appinfo = rel->child_appinfos[partidx];
+#ifdef USE_ASSERT_CHECKING
+			RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+			Assert(partdesc->oids[partidx] == rte->relid);
+#endif
+			result = lappend(result, appinfo);
+		}
+	}
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->painfo->live_partition_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -859,6 +1076,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -869,6 +1087,24 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_rel_partitions(root, rel, rte);
+		Assert(rel->painfo != NULL);
+		rel->painfo->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -889,7 +1125,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -902,10 +1138,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -1114,6 +1346,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->painfo && rel->painfo)
+		{
+			rel->painfo->live_partitioned_rels =
+				list_concat(rel->painfo->live_partitioned_rels,
+					   list_copy(childrel->painfo->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1209,14 +1452,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->painfo->live_partition_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1289,40 +1547,8 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	RangeTblEntry *rte;
 
 	rte = planner_rt_fetch(rel->relid, root);
-
-	/*
-	 * Get the partitioned_rels list from root->pcinfo_list after
-	 * confirming that rel is actually a root partitioned table.
-	 */
-	if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-	{
-		int		parent_relid;
-		bool	is_root_partitioned_table = false;
-
-		/*
-		 * Normally, only the root partitioned rel will be RELOPT_BASEREL
-		 * in a given partitione tree, except when the root table itself
-		 * is a child in the case of a UNION ALL query.
-		 */
-		if (!IS_OTHER_REL(rel))
-			is_root_partitioned_table = true;
-		else if (bms_get_singleton_member(rel->top_parent_relids,
-										  &parent_relid))
-		{
-			RelOptInfo *parent_rel;
-
-			parent_rel = root->simple_rel_array[parent_relid];
-			is_root_partitioned_table =
-						(parent_rel->rtekind != RTE_RELATION);
-		}
-
-		if (is_root_partitioned_table)
-		{
-			partitioned_rels = get_partitioned_child_rels(root, rel->relid);
-			/* The root partitioned table is included as a child rel */
-			Assert(list_length(partitioned_rels) >= 1);
-		}
-	}
+	if (rte->relkind == RELKIND_PARTITIONED_TABLE && IS_SIMPLE_REL(rel))
+		partitioned_rels = rel->painfo->live_partitioned_rels;
 
 	/*
 	 * For every non-dummy child, remember the cheapest path.  Also, identify
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index bfc05a1af5..de50b5d86a 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -68,6 +68,8 @@ static List *get_relation_constraints(PlannerInfo *root,
 static List *build_index_tlist(PlannerInfo *root, IndexOptInfo *index,
 				  Relation heapRelation);
 static List *get_relation_statistics(RelOptInfo *rel, Relation relation);
+static void get_relation_partition_info(PlannerInfo *root, RelOptInfo *rel,
+				Relation relation);
 
 /*
  * get_relation_info -
@@ -420,6 +422,10 @@ get_relation_info(PlannerInfo *root, Oid relationObjectId, bool inhparent,
 	/* Collect info about relation's foreign keys, if relevant */
 	get_relation_foreign_keys(root, rel, relation, inhparent);
 
+	/* Collect partitioning info, if relevant. */
+	if (relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
+		get_relation_partition_info(root, rel, relation);
+
 	heap_close(relation, NoLock);
 
 	/*
@@ -1805,3 +1811,117 @@ has_row_triggers(PlannerInfo *root, Index rti, CmdType event)
 	heap_close(relation, NoLock);
 	return result;
 }
+
+static void
+get_relation_partition_info(PlannerInfo *root, RelOptInfo *rel,
+							Relation relation)
+{
+	int	i;
+	ListCell *l;
+	PartitionKey key = RelationGetPartitionKey(relation);
+	PartitionDesc partdesc = RelationGetPartitionDesc(relation);
+
+	rel->part_scheme = find_partition_scheme(root, relation);
+	rel->partexprs = (List **) palloc0(key->partnatts * sizeof(List *));
+
+	l = list_head(key->partexprs);
+	for (i = 0; i < key->partnatts; i++)
+	{
+		Expr *keyCol;
+
+		if (key->partattrs[i] != 0)
+		{
+			keyCol = (Expr *) makeVar(rel->relid,
+									  key->partattrs[i],
+									  key->parttypid[i],
+									  key->parttypmod[i],
+									  key->parttypcoll[i],
+									  0);
+		}
+		else
+		{
+			if (l == NULL)
+				elog(ERROR, "wrong number of partition key expressions");
+			keyCol = (Expr *) copyObject(lfirst(l));
+			l = lnext(l);
+		}
+
+		rel->partexprs[i] = list_make1(keyCol);
+	}
+
+	/* Values are filled in build_simple_rel(). */
+	rel->child_appinfos = (AppendRelInfo **) palloc0(partdesc->nparts *
+													 sizeof(AppendRelInfo *));
+
+	/*
+	 * A PartitionAppendInfo to map this table to its immediate partitions
+	 * that will be scanned by this query.  At the same time, it records the
+	 * table's partitioning properties reflecting any partition-pruning that
+	 * might occur to satisfy the query.  Rest of the fields are set in
+	 * get_rel_partitions() and set_append_rel_size().
+	 */
+	rel->painfo = makeNode(PartitionAppendInfo);
+	rel->painfo->boundinfo = partdesc->boundinfo;
+}
+
+/*
+ * find_partition_scheme
+ *
+ * The function returns a canonical partition scheme which exactly matches the
+ * partitioning scheme of the given relation if one exists in the list of
+ * canonical partitioning schemes maintained in PlannerInfo. If none of the
+ * existing partitioning schemes match, the function creates a canonical
+ * partition scheme and adds it to the list.
+ *
+ * For an unpartitioned table or for a multi-level partitioned table it returns
+ * NULL. See comments in the function for more details.
+ */
+PartitionScheme
+find_partition_scheme(PlannerInfo *root, Relation relation)
+{
+	ListCell   *lc;
+	PartitionKey key = RelationGetPartitionKey(relation);
+	char		strategy = key->strategy;
+	int			partnatts = key->partnatts;
+	PartitionScheme part_scheme = NULL;
+
+	/* Search for a matching partition scheme and return if found one. */
+	foreach(lc, root->partition_schemes)
+	{
+		part_scheme = lfirst(lc);
+
+		/* Match various partitioning attributes. */
+		if (strategy != part_scheme->strategy ||
+			partnatts != part_scheme->partnatts ||
+			memcmp(key->parttypid, part_scheme->parttypid,
+				   sizeof(Oid) * partnatts) != 0 ||
+			memcmp(key->parttypmod, part_scheme->parttypmod,
+				   sizeof(int32) * partnatts) != 0 ||
+			memcmp(key->partcollation, part_scheme->partcollation,
+				   sizeof(Oid) * partnatts) != 0 ||
+			memcmp(key->partopfamily, part_scheme->partopfamily,
+				   sizeof(Oid) * partnatts) != 0 ||
+			memcmp(key->partopcintype, part_scheme->partopcintype,
+				   sizeof(Oid) * partnatts) != 0)
+			continue;
+
+		/* Found a matching partition scheme. */
+		return part_scheme;
+	}
+
+	/* Did not find matching partition scheme. Create one. */
+	part_scheme = (PartitionScheme) palloc0(sizeof(PartitionSchemeData));
+
+	part_scheme->strategy = strategy;
+	part_scheme->partnatts = partnatts;
+	part_scheme->parttypid = key->parttypid;
+	part_scheme->parttypmod = key->parttypmod;
+	part_scheme->partcollation = key->partcollation;
+	part_scheme->partopfamily = key->partopfamily;
+	part_scheme->partopcintype = key->partopcintype;
+
+	/* Add the partitioning scheme to PlannerInfo. */
+	root->partition_schemes = lappend(root->partition_schemes, part_scheme);
+
+	return part_scheme;
+}
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 8ad0b4a669..390d3b4956 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -17,6 +17,7 @@
 #include <limits.h>
 
 #include "miscadmin.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -163,6 +164,11 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	else
 		rel->top_parent_relids = NULL;
 
+	rel->child_appinfos = NULL;
+	rel->part_scheme = NULL;
+	rel->partexprs = NULL;
+	rel->painfo = NULL;
+
 	/* Check type of rtable entry */
 	switch (rte->rtekind)
 	{
@@ -218,7 +224,18 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	if (rte->inh)
 	{
 		ListCell   *l;
+		AppendRelInfo **child_appinfos = NULL;
+		int			i;
 
+		if (rte->relkind == RELKIND_PARTITIONED_TABLE)
+		{
+			Assert(rel->part_scheme != NULL);
+			Assert(rel->child_appinfos != NULL);
+			Assert(rel->painfo != NULL);
+			child_appinfos = rel->child_appinfos;
+		}
+
+		i = 0;
 		foreach(l, root->append_rel_list)
 		{
 			AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
@@ -229,6 +246,9 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 
 			(void) build_simple_rel(root, appinfo->child_relid,
 									rel);
+
+			if (child_appinfos)
+				child_appinfos[i++] = appinfo;
 		}
 	}
 
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 2283c675e9..fd16494909 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -99,4 +99,12 @@ extern int get_partition_for_tuple(PartitionDispatch *pd,
 						EState *estate,
 						PartitionDispatchData **failed_at,
 						TupleTableSlot **failed_slot);
+
+/* Planner support stuff. */
+extern List *get_partitions_for_keys(Relation rel,
+						NullTestType *keynullness,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive,
+						int *min_datum_index, int *max_datum_index,
+						bool *null_partition_chosen);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/nodes.h b/src/include/nodes/nodes.h
index 27bd4f3363..63196a1211 100644
--- a/src/include/nodes/nodes.h
+++ b/src/include/nodes/nodes.h
@@ -261,6 +261,7 @@ typedef enum NodeTag
 	T_SpecialJoinInfo,
 	T_AppendRelInfo,
 	T_PartitionedChildRelInfo,
+	T_PartitionAppendInfo,
 	T_PlaceHolderInfo,
 	T_MinMaxAggInfo,
 	T_PlannerParamItem,
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index a39e59d8ac..2b535984a7 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -266,6 +266,8 @@ typedef struct PlannerInfo
 	List	   *distinct_pathkeys;	/* distinctClause pathkeys, if any */
 	List	   *sort_pathkeys;	/* sortClause pathkeys, if any */
 
+	List	   *partition_schemes;	/* List of PartitionScheme objects. */
+
 	List	   *initial_rels;	/* RelOptInfos we are now trying to join */
 
 	/* Use fetch_upper_rel() to get any particular upper rel */
@@ -326,6 +328,48 @@ typedef struct PlannerInfo
 	((root)->simple_rte_array ? (root)->simple_rte_array[rti] : \
 	 rt_fetch(rti, (root)->parse->rtable))
 
+/*
+ * Partitioning scheme
+ *     Structure to hold partitioning scheme for a given relation.
+ *
+ * Multiple relations may be partitioned in the same way. The relations
+ * resulting from joining such relations may be partitioned in the same way as
+ * the joining relations. Similarly, relations derived from such relations by
+ * grouping, sorting may be partitioned in the same way as the underlying scan
+ * relations. All such relations partitioned in the same way share the
+ * partitioning scheme.
+ *
+ * PlannerInfo stores a list of distinct "canonical" partitioning schemes.
+ * RelOptInfo of a partitioned relation holds the pointer to "canonical"
+ * partitioning scheme.
+ *
+ * We store opclass declared input data types instead of partition key
+ * datatypes since those are the ones used to compare partition bounds instead
+ * of actual partition key data types. Since partition key data types and the
+ * opclass declared input data types are expected to be binary compatible (per
+ * ResolveOpClass()), both of those should have same byval and length
+ * properties.
+ *
+ * The structure caches information about partition key data type to be used
+ * while matching partition bounds. While comparing partition schemes we don't
+ * need to compare this information as it should be same when opclass declared
+ * input data types are same for two partitioned relations.
+ */
+typedef struct PartitionSchemeData
+{
+	char		strategy;		/* Partitioning strategy */
+	int16		partnatts;		/* Number of partitioning attributes */
+
+	/* The following arrays each have partnatts members. */
+	Oid		   *parttypid;		/* Type OIDs */
+	int32	   *parttypmod;		/* Typemod values */
+	Oid		   *partcollation;	/* Partitioning collation */
+	Oid		   *partopfamily;	/* Operator family OIDs */
+	Oid		   *partopcintype;	/* Operator class-declared input type OIDs */
+} PartitionSchemeData;
+
+typedef struct PartitionSchemeData *PartitionScheme;
+
 
 /*----------
  * RelOptInfo
@@ -515,6 +559,9 @@ typedef enum RelOptKind
 /* Is the given relation an "other" relation? */
 #define IS_OTHER_REL(rel) ((rel)->reloptkind == RELOPT_OTHER_MEMBER_REL)
 
+typedef struct AppendRelInfo AppendRelInfo;
+typedef struct PartitionAppendInfo PartitionAppendInfo;
+
 typedef struct RelOptInfo
 {
 	NodeTag		type;
@@ -592,6 +639,48 @@ typedef struct RelOptInfo
 
 	/* used by "other" relations */
 	Relids		top_parent_relids;	/* Relids of topmost parents */
+
+	/* Fields set for partitioned relations */
+
+	/*
+	 * Information about the partitioning attributes, such as the number of
+	 * attributes, arrays containing per-attribute type/tpymod, partitioning
+	 * collation, operator family OIDs, etc.
+	 */
+	PartitionScheme		part_scheme;
+
+	/*
+	 * Following contains the exact identities of the individual partitioning
+	 * attributes.  For example, if the attribute is a table's column, then
+	 * it will be represented herein by a Var node for the same.  This is
+	 * structured as an array of Lists with part_scheme->partnatts members,
+	 * with each list containing the expression(s) corresponding to the ith
+	 * partitioning attribute (0 <= i < part_schem->partnatts) of this rel.
+	 * For baserels, there is just a single expression in each slot (the ith
+	 * list) of the array, because it corresponds to just one table.  But for
+	 * a joinrel, there will be as many expressions as there are tables
+	 * involved in that joinrel.  We have to do it that way, because in the
+	 * joinrel case, the same corresponding partitioning attribute may have
+	 * different identities in different tables involved in the join; for
+	 * example, a Var node's varno will differ and so might varattnos.
+	 */
+	List			  **partexprs;
+
+	/* AppendRelInfos of *all* partitions of the table. */
+	AppendRelInfo **child_appinfos;
+
+	/*
+	 * For a partitioned relation, the following represents the identities
+	 * of its live partition (their RT indexes) and some informations about
+	 * the bounds that the live partitions satisfy.
+	 */
+	PartitionAppendInfo *painfo;
+
+	/*
+	 * RT index of the root partitioned table in the the partition tree of
+	 * which this rel is a member.
+	 */
+	Index	root_parent_relid;
 } RelOptInfo;
 
 /*
@@ -2031,6 +2120,52 @@ typedef struct PartitionedChildRelInfo
 	List	   *child_rels;
 } PartitionedChildRelInfo;
 
+/* Forward declarations, to avoid including other headers */
+typedef struct PartitionDispatchData *PartitionDispatch;
+typedef struct PartitionBoundInfoData *PartitionBoundInfo;
+typedef struct PartitionKeyData *PartitionKey;
+
+/*
+ * PartitionAppendInfo - Properties of partitions contained in the Append path
+ *						 of a given partitioned table
+ */
+typedef struct PartitionAppendInfo
+{
+	NodeTag		type;
+
+	/*
+	 * List of AppendRelInfos of the table's partitions that satisfy a given
+	 * query.
+	 */
+	List	   *live_partition_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
+
+	/*
+	 * The following simply copies the pointer to boundinfo in the table's
+	 * PartitionDesc.
+	 */
+	PartitionBoundInfo boundinfo;
+
+	/*
+	 * Indexes in the boundinfo->datums array of the smallest and the largest
+	 * value of the partition key that the query allows.  They are set by
+	 * calling get_partitions_for_keys().
+	 */
+	int		min_datum_idx;
+	int		max_datum_idx;
+
+	/*
+	 * Does this Append contain the null-accepting partition, if one exists
+	 * and is allowed by the query's quals.
+	 */
+	bool	contains_null_partition;
+} PartitionAppendInfo;
+
 /*
  * For each distinct placeholder expression generated during planning, we
  * store a PlaceHolderInfo node in the PlannerInfo node's placeholder_list.
diff --git a/src/include/optimizer/plancat.h b/src/include/optimizer/plancat.h
index 71f0faf938..c45db074c6 100644
--- a/src/include/optimizer/plancat.h
+++ b/src/include/optimizer/plancat.h
@@ -56,5 +56,7 @@ extern Selectivity join_selectivity(PlannerInfo *root,
 				 SpecialJoinInfo *sjinfo);
 
 extern bool has_row_triggers(PlannerInfo *root, Index rti, CmdType event);
+extern PartitionScheme find_partition_scheme(PlannerInfo *root,
+				Relation relation);
 
 #endif							/* PLANCAT_H */
-- 
2.11.0

From 1921fc38dee9bd89f42f89012dd1d57ead4dc951 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 3/5] WIP: Interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 123 +++++++++++++++++++++++++++++-----------
 1 file changed, 90 insertions(+), 33 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index bb3009e5b3..14876f8ea3 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -105,6 +105,30 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the user-defined
+ * partition bound of a given existing partition, while an instance of the
+ * following struct describes either a new partition bound being compared
+ * against existing bounds (is_bound is true in that case and either lbound
+ * or rbound is set), or a new tuple's partition key specified in datums
+ * (ndatums = number of partition key columns).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -131,14 +155,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 
 /*
  * RelationBuildPartitionDesc
@@ -663,10 +688,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -717,6 +748,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo && boundinfo->ndatums > 0 &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE);
@@ -736,8 +768,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -751,9 +786,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -2037,9 +2072,14 @@ get_partition_for_tuple(PartitionDispatch *pd,
 		{
 			/* Else bsearch in partdesc->boundinfo */
 			bool		equal = false;
+			PartitionBoundCmpArg	arg;
 
+			memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+			arg.is_bound = false;
+			arg.datums = values;
+			arg.ndatums = key->partnatts;
 			cur_offset = partition_bound_bsearch(key, partdesc->boundinfo,
-												 values, false, &equal);
+												 &arg, &equal);
 			switch (key->strategy)
 			{
 				case PARTITION_STRATEGY_LIST:
@@ -2237,12 +2277,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -2264,11 +2304,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -2276,17 +2316,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -2297,12 +2355,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -2316,20 +2375,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -2342,8 +2400,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 240d2d65fbb4af5959df8e2e1ae576bf46440d4d Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 4/5] WIP: Implement get_partitions_for_keys()

Disable constraint exclusion that occurs using internal partition
constraints, so that it's apparent what the new partition-pruning
code still needs to do to able to create a plan matching the plain
the the traditional constraint exclusion based partition-pruning
would result in.
---
 src/backend/catalog/partition.c      | 193 ++++++++++++++++++++++++++++++++++-
 src/backend/optimizer/util/plancat.c |   4 +
 2 files changed, 192 insertions(+), 5 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 14876f8ea3..617900f62f 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1194,8 +1194,6 @@ RelationGetPartitionDispatchInfo(Relation rel,
  *	datum that the query's bounding keys allow to be returned for the query.
  *	Similarly, *max_datum_index.  *null_partition_chosen returns whether
  *	the null partition will be scanned.
- *
- * TODO: Implement.
  */
 List *
 get_partitions_for_keys(Relation rel,
@@ -1205,12 +1203,197 @@ get_partitions_for_keys(Relation rel,
 						int *min_datum_index, int *max_datum_index,
 						bool *null_partition_chosen)
 {
+	int		i,
+			minoff,
+			maxoff;
 	List   *result = NIL;
-	int		i;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
 	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+	int		null_partition_idx = partdesc->boundinfo->null_index;
+
+	*null_partition_chosen = false;
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if partdesc->boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keynullness[i] == IS_NULL)
+		{
+			if (null_partition_idx >= 0)
+			{
+				*null_partition_chosen = true;
+				result = list_make1_int(null_partition_idx);
+			}
+			else
+				result = NIL;
+
+			return result;
+		}
+	}
+
+	if (n_minkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = minkeys;
+		arg.ndatums = n_minkeys;
+		minoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (min_inclusive)
+					minoff -= 1;
+				else
+					minoff += 1;
+				if (minoff < 0 ||
+					minoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+												 minoff, &arg);
+			} while (cmpval == 0);
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !min_inclusive)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Records returned by the query will be > bounds[minoff],
+				 * because min_scankey is >= bounds[minoff], that is, no
+				 * records of the partition at minoff will be returned.  Go
+				 * to the next bound.
+				 */
+				if (minoff < partdesc->boundinfo->ndatums - 1)
+					minoff += 1;
+
+				/*
+				 * Make sure to skip a gap.
+				 * Note: There are ndatums + 1 lots in the indexes array.
+				 */
+				if (partdesc->boundinfo->indexes[minoff] < 0 &&
+					partdesc->boundinfo->indexes[minoff + 1] >= 0)
+					minoff += 1;
+				break;
+		}
+	}
+	else
+		minoff = 0;
+
+	if (n_maxkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = maxkeys;
+		arg.ndatums = n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
 
-	for (i = 0; i < partdesc->nparts; i++)
-		result = lappend_int(result, i);
+			is_equal = false;
+
+			do
+			{
+				if (max_inclusive)
+					maxoff += 1;
+				else
+					maxoff -= 1;
+				if (maxoff < 0 ||
+					maxoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+											 maxoff, &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (max_inclusive)
+				maxoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !max_inclusive)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Because bounds[maxoff] <= max_scankey, we may need to
+				 * to consider the next partition as well, in addition to
+				 * the partition at maxoff and earlier.
+				 */
+				if (!is_equal || max_inclusive)
+					maxoff += 1;
+
+				/* Make sure to skip a gap. */
+				if (partdesc->boundinfo->indexes[maxoff] < 0 && maxoff >= 1)
+					maxoff -= 1;
+				break;
+		}
+	}
+	else
+		maxoff = partdesc->boundinfo->ndatums - 1;
+
+	*min_datum_index = minoff;
+	*max_datum_index = maxoff;
+
+	switch (partkey->strategy)
+	{
+		case PARTITION_STRATEGY_LIST:
+			for (i = minoff; i <= maxoff; i++)
+			{
+				int		partition_idx = partdesc->boundinfo->indexes[i];
+
+				/*
+				 * Multiple values may belong to the same partition, so make
+				 * sure we don't add the same partition index again.
+				 */
+				result = list_append_unique_int(result, partition_idx);
+			}
+
+			/* If no bounding keys exist, include the null partition too. */
+			if (null_partition_idx >= 0 &&
+				(keynullness[0] == -1 || keynullness[0] != IS_NOT_NULL))
+			{
+				*null_partition_chosen = true;
+				result = list_append_unique_int(result, null_partition_idx);
+			}
+
+			break;
+
+		case PARTITION_STRATEGY_RANGE:
+			for (i = minoff; i <= maxoff; i++)
+			{
+				int		partition_idx = partdesc->boundinfo->indexes[i];
+
+				if (partition_idx >= 0)
+					result = lappend_int(result, partition_idx);
+			}
+			break;
+	}
 
 	return result;
 }
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index de50b5d86a..ec51d89caa 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1155,7 +1155,9 @@ get_relation_constraints(PlannerInfo *root,
 	Index		varno = rel->relid;
 	Relation	relation;
 	TupleConstr *constr;
+#ifdef USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	List	   *pcqual;
+#endif
 
 	/*
 	 * We assume the relation has already been safely locked.
@@ -1241,6 +1243,7 @@ get_relation_constraints(PlannerInfo *root,
 		}
 	}
 
+#ifdef USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	/* Append partition predicates, if any */
 	pcqual = RelationGetPartitionQual(relation);
 	if (pcqual)
@@ -1258,6 +1261,7 @@ get_relation_constraints(PlannerInfo *root,
 
 		result = list_concat(result, pcqual);
 	}
+#endif
 
 	heap_close(relation, NoLock);
 
-- 
2.11.0

From a8808856cb8a6e4553c72b8c0418a69b2bb1aa47 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 5/5] Add more tests for the new partitioning-related planning
 code

---
 src/test/regress/expected/partition.out | 449 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      |  80 ++++++
 4 files changed, 531 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..ae5f59e3a6
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,449 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_null
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(3 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+(7 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+(15 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+(5 rows)
+
+explain (costs off) select * from rlp where a > 10;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+(11 rows)
+
+explain (costs off) select * from rlp where a < 15;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+(5 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+(11 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(7 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where a > 30;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+(15 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, 0, 0) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, 0, 0);
+explain (costs off) select * from mc3p where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(9 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+(15 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+(5 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+(7 rows)
+
+-- XXX - '<>' clauses cannot be handled yet
+explain (costs off) select * from lp where a <> 'a';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'a'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'a'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'a'::bpchar)
+(7 rows)
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p3
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p4
+         Filter: ((a > 1) AND (a = 10))
+(11 rows)
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlp3 where b = 'ab' or b = 'ef';
+                               QUERY PLAN                               
+------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+(7 rows)
+
+drop table lp, rlp, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 2fd3f2b1b1..2eb81fcf41 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 76b0de30a7..6611662149 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..423ac3726f
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,80 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* empty */
+explain (costs off) select * from rlp where a <= 31;
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, 0, 0) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, 0, 0);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+
+-- XXX - '<>' clauses cannot be handled yet
+explain (costs off) select * from lp where a <> 'a';
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlp3 where b = 'ab' or b = 'ef';
+
+drop table lp, rlp, mc3p;
-- 
2.11.0

From ff9ccd8df6555cfca31e54e22293ac1613db327c Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 1/5] Some optimizer data structures for partitioned rels

Nobody uses it though.
---
 src/backend/optimizer/util/plancat.c | 120 +++++++++++++++++++++++++++++++
 src/backend/optimizer/util/relnode.c |  20 ++++++
 src/include/nodes/nodes.h            |   1 +
 src/include/nodes/relation.h         | 135 +++++++++++++++++++++++++++++++++++
 src/include/optimizer/plancat.h      |   2 +
 5 files changed, 278 insertions(+)

diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index a1ebd4acc8..f7e3a1df5f 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -68,6 +68,8 @@ static List *get_relation_constraints(PlannerInfo *root,
 static List *build_index_tlist(PlannerInfo *root, IndexOptInfo *index,
 				  Relation heapRelation);
 static List *get_relation_statistics(RelOptInfo *rel, Relation relation);
+static void get_relation_partition_info(PlannerInfo *root, RelOptInfo *rel,
+				Relation relation);
 
 /*
  * get_relation_info -
@@ -420,6 +422,10 @@ get_relation_info(PlannerInfo *root, Oid relationObjectId, bool inhparent,
 	/* Collect info about relation's foreign keys, if relevant */
 	get_relation_foreign_keys(root, rel, relation, inhparent);
 
+	/* Collect partitioning info, if relevant. */
+	if (relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
+		get_relation_partition_info(root, rel, relation);
+
 	heap_close(relation, NoLock);
 
 	/*
@@ -1802,3 +1808,117 @@ has_row_triggers(PlannerInfo *root, Index rti, CmdType event)
 	heap_close(relation, NoLock);
 	return result;
 }
+
+static void
+get_relation_partition_info(PlannerInfo *root, RelOptInfo *rel,
+							Relation relation)
+{
+	int	i;
+	ListCell *l;
+	PartitionKey key = RelationGetPartitionKey(relation);
+	PartitionDesc partdesc = RelationGetPartitionDesc(relation);
+
+	rel->part_scheme = find_partition_scheme(root, relation);
+	rel->partexprs = (List **) palloc0(key->partnatts * sizeof(List *));
+
+	l = list_head(key->partexprs);
+	for (i = 0; i < key->partnatts; i++)
+	{
+		Expr *keyCol;
+
+		if (key->partattrs[i] != 0)
+		{
+			keyCol = (Expr *) makeVar(rel->relid,
+									  key->partattrs[i],
+									  key->parttypid[i],
+									  key->parttypmod[i],
+									  key->parttypcoll[i],
+									  0);
+		}
+		else
+		{
+			if (l == NULL)
+				elog(ERROR, "wrong number of partition key expressions");
+			keyCol = (Expr *) copyObject(lfirst(l));
+			l = lnext(l);
+		}
+
+		rel->partexprs[i] = list_make1(keyCol);
+	}
+
+	/* Values are filled in build_simple_rel(). */
+	rel->child_appinfos = (AppendRelInfo **) palloc0(partdesc->nparts *
+													 sizeof(AppendRelInfo *));
+
+	/*
+	 * A PartitionAppendInfo to map this table to its immediate partitions
+	 * that will be scanned by this query.  At the same time, it records the
+	 * table's partitioning properties reflecting any partition-pruning that
+	 * might occur to satisfy the query.  Rest of the fields are set in
+	 * get_rel_partitions() and set_append_rel_size().
+	 */
+	rel->painfo = makeNode(PartitionAppendInfo);
+	rel->painfo->boundinfo = partdesc->boundinfo;
+}
+
+/*
+ * find_partition_scheme
+ *
+ * The function returns a canonical partition scheme which exactly matches the
+ * partitioning scheme of the given relation if one exists in the list of
+ * canonical partitioning schemes maintained in PlannerInfo. If none of the
+ * existing partitioning schemes match, the function creates a canonical
+ * partition scheme and adds it to the list.
+ *
+ * For an unpartitioned table or for a multi-level partitioned table it returns
+ * NULL. See comments in the function for more details.
+ */
+PartitionScheme
+find_partition_scheme(PlannerInfo *root, Relation relation)
+{
+	ListCell   *lc;
+	PartitionKey key = RelationGetPartitionKey(relation);
+	char		strategy = key->strategy;
+	int			partnatts = key->partnatts;
+	PartitionScheme part_scheme = NULL;
+
+	/* Search for a matching partition scheme and return if found one. */
+	foreach(lc, root->partition_schemes)
+	{
+		part_scheme = lfirst(lc);
+
+		/* Match various partitioning attributes. */
+		if (strategy != part_scheme->strategy ||
+			partnatts != part_scheme->partnatts ||
+			memcmp(key->parttypid, part_scheme->parttypid,
+				   sizeof(Oid) * partnatts) != 0 ||
+			memcmp(key->parttypmod, part_scheme->parttypmod,
+				   sizeof(int32) * partnatts) != 0 ||
+			memcmp(key->partcollation, part_scheme->partcollation,
+				   sizeof(Oid) * partnatts) != 0 ||
+			memcmp(key->partopfamily, part_scheme->partopfamily,
+				   sizeof(Oid) * partnatts) != 0 ||
+			memcmp(key->partopcintype, part_scheme->partopcintype,
+				   sizeof(Oid) * partnatts) != 0)
+			continue;
+
+		/* Found a matching partition scheme. */
+		return part_scheme;
+	}
+
+	/* Did not find matching partition scheme. Create one. */
+	part_scheme = (PartitionScheme) palloc0(sizeof(PartitionSchemeData));
+
+	part_scheme->strategy = strategy;
+	part_scheme->partnatts = partnatts;
+	part_scheme->parttypid = key->parttypid;
+	part_scheme->parttypmod = key->parttypmod;
+	part_scheme->partcollation = key->partcollation;
+	part_scheme->partopfamily = key->partopfamily;
+	part_scheme->partopcintype = key->partopcintype;
+
+	/* Add the partitioning scheme to PlannerInfo. */
+	root->partition_schemes = lappend(root->partition_schemes, part_scheme);
+
+	return part_scheme;
+}
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index c7b2695ebb..f0973b83b9 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -17,6 +17,7 @@
 #include <limits.h>
 
 #include "miscadmin.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -163,6 +164,11 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	else
 		rel->top_parent_relids = NULL;
 
+	rel->child_appinfos = NULL;
+	rel->part_scheme = NULL;
+	rel->partexprs = NULL;
+	rel->painfo = NULL;
+
 	/* Check type of rtable entry */
 	switch (rte->rtekind)
 	{
@@ -218,7 +224,18 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	if (rte->inh)
 	{
 		ListCell   *l;
+		AppendRelInfo **child_appinfos = NULL;
+		int			i;
 
+		if (rte->relkind == RELKIND_PARTITIONED_TABLE)
+		{
+			Assert(rel->part_scheme != NULL);
+			Assert(rel->child_appinfos != NULL);
+			Assert(rel->painfo != NULL);
+			child_appinfos = rel->child_appinfos;
+		}
+
+		i = 0;
 		foreach(l, root->append_rel_list)
 		{
 			AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
@@ -229,6 +246,9 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 
 			(void) build_simple_rel(root, appinfo->child_relid,
 									rel);
+
+			if (child_appinfos)
+				child_appinfos[i++] = appinfo;
 		}
 	}
 
diff --git a/src/include/nodes/nodes.h b/src/include/nodes/nodes.h
index 27bd4f3363..63196a1211 100644
--- a/src/include/nodes/nodes.h
+++ b/src/include/nodes/nodes.h
@@ -261,6 +261,7 @@ typedef enum NodeTag
 	T_SpecialJoinInfo,
 	T_AppendRelInfo,
 	T_PartitionedChildRelInfo,
+	T_PartitionAppendInfo,
 	T_PlaceHolderInfo,
 	T_MinMaxAggInfo,
 	T_PlannerParamItem,
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index d50ff55681..0f4996b424 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -266,6 +266,8 @@ typedef struct PlannerInfo
 	List	   *distinct_pathkeys;	/* distinctClause pathkeys, if any */
 	List	   *sort_pathkeys;	/* sortClause pathkeys, if any */
 
+	List	   *partition_schemes;	/* List of PartitionScheme objects. */
+
 	List	   *initial_rels;	/* RelOptInfos we are now trying to join */
 
 	/* Use fetch_upper_rel() to get any particular upper rel */
@@ -326,6 +328,48 @@ typedef struct PlannerInfo
 	((root)->simple_rte_array ? (root)->simple_rte_array[rti] : \
 	 rt_fetch(rti, (root)->parse->rtable))
 
+/*
+ * Partitioning scheme
+ *     Structure to hold partitioning scheme for a given relation.
+ *
+ * Multiple relations may be partitioned in the same way. The relations
+ * resulting from joining such relations may be partitioned in the same way as
+ * the joining relations. Similarly, relations derived from such relations by
+ * grouping, sorting may be partitioned in the same way as the underlying scan
+ * relations. All such relations partitioned in the same way share the
+ * partitioning scheme.
+ *
+ * PlannerInfo stores a list of distinct "canonical" partitioning schemes.
+ * RelOptInfo of a partitioned relation holds the pointer to "canonical"
+ * partitioning scheme.
+ *
+ * We store opclass declared input data types instead of partition key
+ * datatypes since those are the ones used to compare partition bounds instead
+ * of actual partition key data types. Since partition key data types and the
+ * opclass declared input data types are expected to be binary compatible (per
+ * ResolveOpClass()), both of those should have same byval and length
+ * properties.
+ *
+ * The structure caches information about partition key data type to be used
+ * while matching partition bounds. While comparing partition schemes we don't
+ * need to compare this information as it should be same when opclass declared
+ * input data types are same for two partitioned relations.
+ */
+typedef struct PartitionSchemeData
+{
+	char		strategy;		/* Partitioning strategy */
+	int16		partnatts;		/* Number of partitioning attributes */
+
+	/* The following arrays each have partnatts members. */
+	Oid		   *parttypid;		/* Type OIDs */
+	int32	   *parttypmod;		/* Typemod values */
+	Oid		   *partcollation;	/* Partitioning collation */
+	Oid		   *partopfamily;	/* Operator family OIDs */
+	Oid		   *partopcintype;	/* Operator class-declared input type OIDs */
+} PartitionSchemeData;
+
+typedef struct PartitionSchemeData *PartitionScheme;
+
 
 /*----------
  * RelOptInfo
@@ -515,6 +559,9 @@ typedef enum RelOptKind
 /* Is the given relation an "other" relation? */
 #define IS_OTHER_REL(rel) ((rel)->reloptkind == RELOPT_OTHER_MEMBER_REL)
 
+typedef struct AppendRelInfo AppendRelInfo;
+typedef struct PartitionAppendInfo PartitionAppendInfo;
+
 typedef struct RelOptInfo
 {
 	NodeTag		type;
@@ -592,6 +639,42 @@ typedef struct RelOptInfo
 
 	/* used by "other" relations */
 	Relids		top_parent_relids;	/* Relids of topmost parents */
+
+	/* Fields set for partitioned relations */
+
+	/*
+	 * Information about the partitioning attributes, such as the number of
+	 * attributes, arrays containing per-attribute type/tpymod, partitioning
+	 * collation, operator family OIDs, etc.
+	 */
+	PartitionScheme		part_scheme;
+
+	/*
+	 * Following contains the exact identities of the individual partitioning
+	 * attributes.  For example, if the attribute is a table's column, then
+	 * it will be represented herein by a Var node for the same.  This is
+	 * structured as an array of Lists with part_scheme->partnatts members,
+	 * with each list containing the expression(s) corresponding to the ith
+	 * partitioning attribute (0 <= i < part_schem->partnatts) of this rel.
+	 * For baserels, there is just a single expression in each slot (the ith
+	 * list) of the array, because it corresponds to just one table.  But for
+	 * a joinrel, there will be as many expressions as there are tables
+	 * involved in that joinrel.  We have to do it that way, because in the
+	 * joinrel case, the same corresponding partitioning attribute may have
+	 * different identities in different tables involved in the join; for
+	 * example, a Var node's varno will differ and so might varattnos.
+	 */
+	List			  **partexprs;
+
+	/* AppendRelInfos of *all* partitions of the table. */
+	AppendRelInfo **child_appinfos;
+
+	/*
+	 * For a partitioned relation, the following represents the identities
+	 * of its live partition (their RT indexes) and some informations about
+	 * the bounds that the live partitions satisfy.
+	 */
+	PartitionAppendInfo *painfo;
 } RelOptInfo;
 
 /*
@@ -2031,6 +2114,58 @@ typedef struct PartitionedChildRelInfo
 	List	   *child_rels;
 } PartitionedChildRelInfo;
 
+/* Forward declarations, to avoid including other headers */
+typedef struct PartitionDispatchData *PartitionDispatch;
+typedef struct PartitionBoundInfoData *PartitionBoundInfo;
+typedef struct PartitionKeyData *PartitionKey;
+
+/*
+ * PartitionAppendInfo - Properties of partitions contained in the Append path
+ *						 of a given partitioned table
+ */
+typedef struct PartitionAppendInfo
+{
+	NodeTag		type;
+
+	/*
+	 * List of AppendRelInfos of the table's partitions that satisfy a given
+	 * query.
+	 */
+	List	   *live_partition_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
+
+	/*
+	 * The following simply copies the pointer to boundinfo in the table's
+	 * PartitionDesc.
+	 */
+	PartitionBoundInfo boundinfo;
+
+	/*
+	 * Indexes in the boundinfo->datums array of the smallest and the largest
+	 * value of the partition key that the query allows.  They are set by
+	 * calling get_partitions_for_keys().
+	 */
+	int		min_datum_idx;
+	int		max_datum_idx;
+
+	/*
+	 * Does this Append contain the null-accepting partition, if one exists
+	 * and is allowed by the query's quals.
+	 */
+	bool	contains_null_partition;
+
+	/*
+	 * Does this Append contain the default partition, if one exists and is
+	 * allowed by the query's quals.
+	 */
+	bool contains_default_partition;
+} PartitionAppendInfo;
+
 /*
  * For each distinct placeholder expression generated during planning, we
  * store a PlaceHolderInfo node in the PlannerInfo node's placeholder_list.
diff --git a/src/include/optimizer/plancat.h b/src/include/optimizer/plancat.h
index 71f0faf938..c45db074c6 100644
--- a/src/include/optimizer/plancat.h
+++ b/src/include/optimizer/plancat.h
@@ -56,5 +56,7 @@ extern Selectivity join_selectivity(PlannerInfo *root,
 				 SpecialJoinInfo *sjinfo);
 
 extern bool has_row_triggers(PlannerInfo *root, Index rti, CmdType event);
+extern PartitionScheme find_partition_scheme(PlannerInfo *root,
+				Relation relation);
 
 #endif							/* PLANCAT_H */
-- 
2.11.0

From 24bc6a3428730e24b04b2b1282960e45ffeb0467 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 17:31:42 +0900
Subject: [PATCH 2/5] WIP: planner-side changes for partition-pruning

Firstly, this adds a stub get_partitions_for_keys() in partition.c
with appropriate interface for the caller to specify bounding scan
keys, along with other information about the scan keys extracted
from the query, such as NULL-ness of the keys, inclusive-ness, etc.

More importantly, this implements the planner-side logic to extract
bounding scan keys to be passed to get_partitions_for_keys.  That is,
it will go through rel->baserestrictinfo and match individual clauses
to partition keys and construct lower bound and upper bound tuples,
which may cover only a prefix of a multi-column partition key.

A bunch of smarts are still missing when mapping the clause operands
with keys.  For example, code to match a clause is specifed as
(constant op var) doesn't exist.  Also, redundant keys are not
eliminated, for example, a combination of clauses a = 10 and a > 1
will cause the later clause a > 1 taking over and resulting in
needless scanning of partitions containing values a > 1 and a < 10.

...constraint exclusion is still used, because
get_partitions_for_keys is just a stub...
---
 src/backend/catalog/partition.c       |  43 ++++
 src/backend/optimizer/path/allpaths.c | 358 ++++++++++++++++++++++++++++++----
 src/include/catalog/partition.h       |   9 +
 3 files changed, 371 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 1ab6dba7ae..ccf8a1fa67 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1335,6 +1335,49 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_for_keys
+ *		Returns the list of indexes of rel's partitions that will need to be
+ *		scanned given the bounding scan keys.
+ *
+ * Each value in the returned list can be used as an index into the oids array
+ * of the partition descriptor.
+ *
+ * Inputs:
+ *	keynullness contains between 0 and (key->partnatts - 1) values, each
+ *	telling what kind of NullTest has been applies to the corresponding
+ *	partition key column.  minkeys represents the lower bound on the partition
+ *	the key of the records that the query will return, while maxkeys
+ *	represents upper bound.  min_inclusive and max_inclusive tell whether the
+ *	bounds specified minkeys and maxkeys is inclusive, respectively.
+ *
+ * Other outputs:
+ *	*min_datum_index will return the index in boundinfo->datums of the first
+ *	datum that the query's bounding keys allow to be returned for the query.
+ *	Similarly, *max_datum_index.  *null_partition_chosen returns whether
+ *	the null partition will be scanned.
+ *
+ * TODO: Implement.
+ */
+List *
+get_partitions_for_keys(Relation rel,
+						NullTestType *keynullness,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive,
+						int *min_datum_index, int *max_datum_index,
+						bool *null_partition_chosen,
+						bool *default_partition_chosen)
+{
+	List   *result = NIL;
+	int		i;
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+
+	for (i = 0; i < partdesc->nparts; i++)
+		result = lappend_int(result, i);
+
+	return result;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 5b746a906a..6e5efe98f9 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,6 +20,7 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
@@ -846,6 +847,251 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the query clauses
+ *
+ * Returned list contains the AppendInfos of the chosen partitions.
+ */
+static List *
+get_rel_partitions(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
+{
+	Relation		parent = heap_open(rte->relid, NoLock);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(parent);
+	List   *indexes;
+	List   *result = NIL;
+	ListCell   *lc1,
+			   *lc2;
+	int		keyPos;
+	List   *matchedclauses[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	Datum	minkeys[PARTITION_MAX_KEYS],
+			maxkeys[PARTITION_MAX_KEYS];
+	bool	need_next_min,
+			need_next_max,
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_minkeys = 0,
+			n_maxkeys = 0,
+			i;
+
+	/*
+	 * Match individual OpExprs in the query's restriction with individual
+	 * partition key columns.  There is one list per key.
+	 */
+	memset(keynullness, -1, sizeof(keynullness));
+	memset(matchedclauses, 0, sizeof(matchedclauses));
+	keyPos = 0;
+	for (i = 0; i < rel->part_scheme->partnatts; i++)
+	{
+		Node   *partkey = linitial(rel->partexprs[i]);
+
+		foreach(lc2, rel->baserestrictinfo)
+		{
+			RestrictInfo   *rinfo = lfirst(lc2);
+			Expr		   *clause = rinfo->clause;
+
+			if (is_opclause(clause))
+			{
+				Node   *leftop = get_leftop(clause),
+					   *rightop = get_rightop(clause);
+				Oid		expr_op = ((OpExpr *) clause)->opno;
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = (Node *) ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					matchedclauses[keyPos] = lappend(matchedclauses[keyPos],
+													 clause);
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[keyPos] = IS_NOT_NULL;
+				}
+				else if (equal(rightop, partkey))
+				{
+					Oid		commutator = get_commutator(expr_op);
+
+					if (OidIsValid(commutator))
+					{
+						OpExpr   *commutated_expr;
+
+						/*
+						 * Generate a commutated copy of the expression, but
+						 * try to make it look valid, because we only want
+						 * it to put the constant operand in a place that the
+						 * following code knows as the only place to find it.
+						 */
+						commutated_expr = (OpExpr *) copyObject(clause);
+						commutated_expr->opno = commutator;	/* really? */
+						commutated_expr->args = list_make2(rightop, leftop);
+						matchedclauses[keyPos] =
+											lappend(matchedclauses[keyPos],
+													commutated_expr);
+						/* A strict operator implies NOT NULL argument. */
+						keynullness[keyPos] = IS_NOT_NULL;
+					}
+				}
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					keynullness[keyPos] = nulltest->nulltesttype;
+			}
+		}
+
+		/* Onto finding clauses matching the next partition key. */
+		keyPos++;
+	}
+
+	/*
+	 * Determine the min keys and the max keys using btree semantics-based
+	 * interpretation of the clauses' operators.
+	 */
+
+	/*
+	 * XXX - There should be a step similar to _bt_preprocess_keys() here,
+	 * to eliminate any redundant scan keys for a given partition column.  For
+	 * example, among a <= 4 and a <= 5, we can only keep a <= 4 for being
+	 * more restrictive and discard a <= 5.  While doing that, we can also
+	 * check to see if there exists a contradictory combination of scan keys
+	 * that makes the query trivially false for all records in the table.
+	 */
+	memset(minkeys, 0, sizeof(minkeys));
+	memset(maxkeys, 0, sizeof(maxkeys));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < rel->part_scheme->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc1, matchedclauses[i])
+		{
+			Expr   *clause = lfirst(lc1);
+			Const  *rightop = (Const *) get_rightop(clause);
+			Oid		opno = ((OpExpr *) clause)->opno,
+					opfamily = rel->part_scheme->partopfamily[i];
+			StrategyNumber	strategy;
+
+			strategy = get_op_opfamily_strategy(opno, opfamily);
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+					}
+					if (strategy == BTLessStrategyNumber)
+						need_next_max = false;
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+					}
+					if (strategy == BTGreaterStrategyNumber)
+						need_next_min = false;
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+					}
+					minkey_set[i] = true;
+					min_incl = true;
+
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+					}
+					maxkey_set[i] = true;
+					max_incl = true;
+					break;
+
+				/*
+				 * This might mean '<>', but we don't have anything for that
+				 * case yet.  Perhaps, handle that as key < const OR
+				 * key > const, once we have props needed for handling OR
+				 * clauses.
+				 */
+				default:
+					min_incl = max_incl = false;
+					break;
+			}
+		}
+	}
+
+	/* Ask partition.c which partitions it thinks match the keys. */
+	indexes = get_partitions_for_keys(parent, keynullness,
+									  minkeys, n_minkeys, min_incl,
+									  maxkeys, n_maxkeys, max_incl,
+									  &rel->painfo->min_datum_idx,
+									  &rel->painfo->max_datum_idx,
+									  &rel->painfo->contains_null_partition,
+								  &rel->painfo->contains_default_partition);
+
+	if (indexes != NIL)
+	{
+#ifdef USE_ASSERT_CHECKING
+		int		first_index,
+				last_index;
+		first_index = linitial_int(indexes);
+		last_index = llast_int(indexes);
+		Assert(first_index <= last_index ||
+			   rel->part_scheme->strategy != PARTITION_STRATEGY_RANGE);
+#endif
+
+		foreach(lc1, indexes)
+		{
+			int		partidx = lfirst_int(lc1);
+			AppendRelInfo *appinfo = rel->child_appinfos[partidx];
+#ifdef USE_ASSERT_CHECKING
+			RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+			Assert(partdesc->oids[partidx] == rte->relid);
+#endif
+			result = lappend(result, appinfo);
+		}
+	}
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->painfo->live_partition_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1106,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1120,24 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_rel_partitions(root, rel, rte);
+		Assert(rel->painfo != NULL);
+		rel->painfo->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1158,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1171,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -1118,6 +1379,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->painfo && rel->painfo)
+		{
+			rel->painfo->live_partitioned_rels =
+				list_concat(rel->painfo->live_partitioned_rels,
+					   list_copy(childrel->painfo->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1213,14 +1485,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->painfo->live_partition_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1291,33 +1578,31 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
 	/*
-	 * A root partition will already have a PartitionedChildRelInfo, and a
-	 * non-root partitioned table doesn't need one, because its Append paths
-	 * will get flattened into the parent anyway.  For a subquery RTE, no
-	 * PartitionedChildRelInfo exists; we collect all partitioned_rels
-	 * associated with any child.  (This assumes that we don't need to look
-	 * through multiple levels of subquery RTEs; if we ever do, we could
-	 * create a PartitionedChildRelInfo with the accumulated list of
-	 * partitioned_rels which would then be found when populated our parent
-	 * rel with paths.  For the present, that appears to be unnecessary.)
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
 	 */
 	rte = planner_rt_fetch(rel->relid, root);
-	switch (rte->rtekind)
-	{
-		case RTE_RELATION:
-			if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-				partitioned_rels =
-					get_partitioned_child_rels(root, rel->relid);
-			break;
-		case RTE_SUBQUERY:
-			build_partitioned_rels = true;
-			break;
-		default:
-			elog(ERROR, "unexpcted rtekind: %d", (int) rte->rtekind);
-	}
+	if (rte->rtekind == RTE_RELATION &&
+		rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->painfo->live_partitioned_rels;
 
 	/*
 	 * For every non-dummy child, remember the cheapest path.  Also, identify
@@ -1330,17 +1615,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->painfo)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+								   childrel->painfo->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 454a940a23..4f27d3018e 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -99,6 +99,7 @@ extern int get_partition_for_tuple(PartitionDispatch *pd,
 						EState *estate,
 						PartitionDispatchData **failed_at,
 						TupleTableSlot **failed_slot);
+
 extern Oid	get_default_oid_from_partdesc(PartitionDesc partdesc);
 extern Oid	get_default_partition_oid(Oid parentId);
 extern void update_default_partition_oid(Oid parentId, Oid defaultPartId);
@@ -106,4 +107,12 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* Planner support stuff. */
+extern List *get_partitions_for_keys(Relation rel,
+						NullTestType *keynullness,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive,
+						int *min_datum_index, int *max_datum_index,
+						bool *null_partition_chosen,
+						bool *default_partition_chosen);
 #endif							/* PARTITION_H */
-- 
2.11.0

From a5725aca1168b6539ea591a886775a7f3d170e8d Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 3/5] WIP: Interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index ccf8a1fa67..0133748234 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -111,6 +111,30 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the user-defined
+ * partition bound of a given existing partition, while an instance of the
+ * following struct describes either a new partition bound being compared
+ * against existing bounds (is_bound is true in that case and either lbound
+ * or rbound is set), or a new tuple's partition key specified in datums
+ * (ndatums = number of partition key columns).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -139,14 +163,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -755,10 +780,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -809,6 +840,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -830,8 +862,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -845,9 +880,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -2380,12 +2415,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -2397,6 +2435,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -2427,12 +2466,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -2629,12 +2669,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -2656,11 +2696,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -2668,17 +2708,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -2689,12 +2747,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -2708,20 +2767,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -2734,8 +2792,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 28778969ecddb2a9c3f31ff2ed119a342e97fb2d Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 4/5] WIP: Implement get_partitions_for_keys()

Disable constraint exclusion that occurs using internal partition
constraints, so that it's apparent what the new partition-pruning
code still needs to do to able to create a plan matching the plain
the the traditional constraint exclusion based partition-pruning
would result in.
---
 src/backend/catalog/partition.c      | 210 ++++++++++++++++++++++++++++++++++-
 src/backend/optimizer/util/plancat.c |   4 +
 2 files changed, 209 insertions(+), 5 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 0133748234..9d4b7c1a7f 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1391,8 +1391,6 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
  *	datum that the query's bounding keys allow to be returned for the query.
  *	Similarly, *max_datum_index.  *null_partition_chosen returns whether
  *	the null partition will be scanned.
- *
- * TODO: Implement.
  */
 List *
 get_partitions_for_keys(Relation rel,
@@ -1403,12 +1401,214 @@ get_partitions_for_keys(Relation rel,
 						bool *null_partition_chosen,
 						bool *default_partition_chosen)
 {
+	int		i,
+			minoff,
+			maxoff;
 	List   *result = NIL;
-	int		i;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
 	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundCmpArg	arg;
+	bool	is_equal,
+			scan_default = false;
+	int		null_partition_idx = partdesc->boundinfo->null_index;
 
-	for (i = 0; i < partdesc->nparts; i++)
-		result = lappend_int(result, i);
+	*null_partition_chosen = false;
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if partdesc->boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keynullness[i] == IS_NULL)
+		{
+			if (null_partition_idx >= 0)
+			{
+				*null_partition_chosen = true;
+				result = list_make1_int(null_partition_idx);
+			}
+			else
+				result = NIL;
+
+			return result;
+		}
+	}
+
+	/*
+	 * If query provides no quals, don't forget to scan the default partition.
+	 */
+	if (n_minkeys == 0 && n_maxkeys == 0)
+		scan_default = true;
+
+	if (n_minkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = minkeys;
+		arg.ndatums = n_minkeys;
+		minoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (min_inclusive)
+					minoff -= 1;
+				else
+					minoff += 1;
+				if (minoff < 0 ||
+					minoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+												 minoff, &arg);
+			} while (cmpval == 0);
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !min_inclusive)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Records returned by the query will be > bounds[minoff],
+				 * because min_scankey is >= bounds[minoff], that is, no
+				 * records of the partition at minoff will be returned.  Go
+				 * to the next bound.
+				 */
+				if (minoff < partdesc->boundinfo->ndatums - 1)
+					minoff += 1;
+
+				/*
+				 * Make sure to skip a gap.
+				 * Note: There are ndatums + 1 lots in the indexes array.
+				 */
+				if (partdesc->boundinfo->indexes[minoff] < 0 &&
+					partdesc->boundinfo->indexes[minoff + 1] >= 0)
+					minoff += 1;
+				break;
+		}
+	}
+	else
+		minoff = 0;
+
+	if (n_maxkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = maxkeys;
+		arg.ndatums = n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (max_inclusive)
+					maxoff += 1;
+				else
+					maxoff -= 1;
+				if (maxoff < 0 ||
+					maxoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+											 maxoff, &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (max_inclusive)
+				maxoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !max_inclusive)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Because bounds[maxoff] <= max_scankey, we may need to
+				 * to consider the next partition as well, in addition to
+				 * the partition at maxoff and earlier.
+				 */
+				if (!is_equal || max_inclusive)
+					maxoff += 1;
+
+				/* Make sure to skip a gap. */
+				if (partdesc->boundinfo->indexes[maxoff] < 0 && maxoff >= 1)
+					maxoff -= 1;
+				break;
+		}
+	}
+	else
+		maxoff = partdesc->boundinfo->ndatums - 1;
+
+	*min_datum_index = minoff;
+	*max_datum_index = maxoff;
+
+	switch (partkey->strategy)
+	{
+		case PARTITION_STRATEGY_LIST:
+			for (i = minoff; i <= maxoff; i++)
+			{
+				int		partition_idx = partdesc->boundinfo->indexes[i];
+
+				/*
+				 * Multiple values may belong to the same partition, so make
+				 * sure we don't add the same partition index again.
+				 */
+				result = list_append_unique_int(result, partition_idx);
+			}
+
+			/* If no bounding keys exist, include the null partition too. */
+			if (null_partition_idx >= 0 &&
+				(keynullness[0] == -1 || keynullness[0] != IS_NOT_NULL))
+			{
+				*null_partition_chosen = true;
+				result = list_append_unique_int(result, null_partition_idx);
+			}
+
+			break;
+
+		case PARTITION_STRATEGY_RANGE:
+			for (i = minoff; i <= maxoff; i++)
+			{
+				int		partition_idx = partdesc->boundinfo->indexes[i];
+
+				/*
+				 * If a valid partition exists for this range, add its
+				 * index, if not, the default partition (if any) would be
+				 * covering that range, so request to include the same.
+				 */
+				if (partition_idx >= 0)
+					result = lappend_int(result, partition_idx);
+				else
+					scan_default = true;
+			}
+			break;
+	}
+
+	if (scan_default && partdesc->boundinfo->default_index >= 0)
+		result = lappend_int(result, partdesc->boundinfo->default_index);
 
 	return result;
 }
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index f7e3a1df5f..26ea2b4162 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1155,7 +1155,9 @@ get_relation_constraints(PlannerInfo *root,
 	Index		varno = rel->relid;
 	Relation	relation;
 	TupleConstr *constr;
+#ifdef USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	List	   *pcqual;
+#endif
 
 	/*
 	 * We assume the relation has already been safely locked.
@@ -1241,6 +1243,7 @@ get_relation_constraints(PlannerInfo *root,
 		}
 	}
 
+#ifdef USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	/* Append partition predicates, if any */
 	pcqual = RelationGetPartitionQual(relation);
 	if (pcqual)
@@ -1258,6 +1261,7 @@ get_relation_constraints(PlannerInfo *root,
 
 		result = list_concat(result, pcqual);
 	}
+#endif
 
 	heap_close(relation, NoLock);
 
-- 
2.11.0

From a8660afe1d147dcaa5ff46a8ed4faf366242d4d0 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 5/5] Add more tests for the new partitioning-related planning
 code

---
 src/test/regress/expected/partition.out | 465 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      |  82 ++++++
 4 files changed, 549 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..db92368fc5
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,465 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_null
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(3 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+(7 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+(15 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+(5 rows)
+
+explain (costs off) select * from rlp where a > 10;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+(11 rows)
+
+explain (costs off) select * from rlp where a < 15;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+(5 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+(11 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(7 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where a > 30;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+(15 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, 0, 0) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, 0, 0);
+explain (costs off) select * from mc3p where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(9 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+(15 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+(5 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+(7 rows)
+
+-- XXX - '<>' clauses cannot be handled yet
+explain (costs off) select * from lp where a <> 'a';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'a'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'a'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'a'::bpchar)
+(7 rows)
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p3
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p4
+         Filter: ((a > 1) AND (a = 10))
+(11 rows)
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlp3 where b = 'ab' or b = 'ef';
+                               QUERY PLAN                               
+------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+(7 rows)
+
+drop table lp, rlp, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 2fd3f2b1b1..2eb81fcf41 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 76b0de30a7..6611662149 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..616ad95611
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,82 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* empty */
+explain (costs off) select * from rlp where a <= 31;
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, 0, 0) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, 0, 0);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+
+-- XXX - '<>' clauses cannot be handled yet
+explain (costs off) select * from lp where a <> 'a';
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlp3 where b = 'ab' or b = 'ef';
+
+drop table lp, rlp, mc3p;
-- 
2.11.0

diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 6e5efe9..2bb3641 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -846,50 +846,28 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->fdwroutine->GetForeignPaths(root, rel, rte->relid);
 }
 
-/*
- * get_rel_partitions
- *		Return the list of partitions of rel that pass the query clauses
- *
- * Returned list contains the AppendInfos of the chosen partitions.
- */
-static List *
-get_rel_partitions(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
+static void
+get_matching_clause(RelOptInfo *rel, List *clauses, List **matchedclauses,
+					NullTestType *keynullness)
 {
-	Relation		parent = heap_open(rte->relid, NoLock);
-	PartitionDesc	partdesc = RelationGetPartitionDesc(parent);
-	List   *indexes;
-	List   *result = NIL;
-	ListCell   *lc1,
-			   *lc2;
+	ListCell   *lc;
 	int		keyPos;
-	List   *matchedclauses[PARTITION_MAX_KEYS];
-	NullTestType keynullness[PARTITION_MAX_KEYS];
-	Datum	minkeys[PARTITION_MAX_KEYS],
-			maxkeys[PARTITION_MAX_KEYS];
-	bool	need_next_min,
-			need_next_max,
-			minkey_set[PARTITION_MAX_KEYS],
-			maxkey_set[PARTITION_MAX_KEYS],
-			min_incl,
-			max_incl;
-	int		n_minkeys = 0,
-			n_maxkeys = 0,
-			i;
+	int		i;
 
 	/*
 	 * Match individual OpExprs in the query's restriction with individual
 	 * partition key columns.  There is one list per key.
 	 */
-	memset(keynullness, -1, sizeof(keynullness));
-	memset(matchedclauses, 0, sizeof(matchedclauses));
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType));
+	memset(matchedclauses, 0, PARTITION_MAX_KEYS * sizeof(List*));
 	keyPos = 0;
 	for (i = 0; i < rel->part_scheme->partnatts; i++)
 	{
 		Node   *partkey = linitial(rel->partexprs[i]);
 
-		foreach(lc2, rel->baserestrictinfo)
+		foreach(lc, clauses)
 		{
-			RestrictInfo   *rinfo = lfirst(lc2);
+			RestrictInfo   *rinfo = lfirst(lc);
 			Expr		   *clause = rinfo->clause;
 
 			if (is_opclause(clause))
@@ -948,6 +926,37 @@ get_rel_partitions(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 		/* Onto finding clauses matching the next partition key. */
 		keyPos++;
 	}
+}
+
+/*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the query clauses
+ *
+ * Returned list contains the AppendInfos of the chosen partitions.
+ */
+static List *
+get_rel_partitions(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
+{
+	Relation		parent = heap_open(rte->relid, NoLock);
+	List   *indexes;
+	List   *result = NIL;
+	ListCell   *lc1;
+	List   *matchedclauses[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	Datum	minkeys[PARTITION_MAX_KEYS],
+			maxkeys[PARTITION_MAX_KEYS];
+	bool	need_next_min,
+			need_next_max,
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_minkeys = 0,
+			n_maxkeys = 0,
+			i;
+
+	get_matching_clause(rel, rel->baserestrictinfo, matchedclauses,
+						keynullness);
 
 	/*
 	 * Determine the min keys and the max keys using btree semantics-based

From f20aebcad9b089434ba60cd4439fa1a9d55091b8 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 1/4] WIP: planner-side changes for partition-pruning

Firstly, this adds a stub get_partitions_for_keys() in partition.c
with appropriate interface for the caller to specify bounding scan
keys, along with other information about the scan keys extracted
from the query, such as NULL-ness of the keys, inclusive-ness, etc.

More importantly, this implements the planner-side logic to extract
bounding scan keys to be passed to get_partitions_for_keys.  That is,
it will go through rel->baserestrictinfo and match individual clauses
to partition keys and construct lower bound and upper bound tuples,
which may cover only a prefix of a multi-column partition key.

A bunch of smarts are still missing when mapping the clause operands
with keys.  For example, code to match a clause is specifed as
(constant op var) doesn't exist.  Also, redundant keys are not
eliminated, for example, a combination of clauses a = 10 and a > 1
will cause the later clause a > 1 taking over and resulting in
needless scanning of partitions containing values a > 1 and a < 10.

...constraint exclusion is still used, because
get_partitions_for_keys is just a stub...

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  43 ++++
 src/backend/optimizer/path/allpaths.c | 390 ++++++++++++++++++++++++++++++----
 src/backend/optimizer/util/plancat.c  |  10 +
 src/backend/optimizer/util/relnode.c  |   7 +
 src/include/catalog/partition.h       |   9 +
 src/include/nodes/nodes.h             |   1 +
 src/include/nodes/relation.h          |  66 ++++++
 7 files changed, 487 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 1ab6dba7ae..ccf8a1fa67 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1335,6 +1335,49 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_for_keys
+ *		Returns the list of indexes of rel's partitions that will need to be
+ *		scanned given the bounding scan keys.
+ *
+ * Each value in the returned list can be used as an index into the oids array
+ * of the partition descriptor.
+ *
+ * Inputs:
+ *	keynullness contains between 0 and (key->partnatts - 1) values, each
+ *	telling what kind of NullTest has been applies to the corresponding
+ *	partition key column.  minkeys represents the lower bound on the partition
+ *	the key of the records that the query will return, while maxkeys
+ *	represents upper bound.  min_inclusive and max_inclusive tell whether the
+ *	bounds specified minkeys and maxkeys is inclusive, respectively.
+ *
+ * Other outputs:
+ *	*min_datum_index will return the index in boundinfo->datums of the first
+ *	datum that the query's bounding keys allow to be returned for the query.
+ *	Similarly, *max_datum_index.  *null_partition_chosen returns whether
+ *	the null partition will be scanned.
+ *
+ * TODO: Implement.
+ */
+List *
+get_partitions_for_keys(Relation rel,
+						NullTestType *keynullness,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive,
+						int *min_datum_index, int *max_datum_index,
+						bool *null_partition_chosen,
+						bool *default_partition_chosen)
+{
+	List   *result = NIL;
+	int		i;
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+
+	for (i = 0; i < partdesc->nparts; i++)
+		result = lappend_int(result, i);
+
+	return result;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index a7866a99e0..47f88d1a8f 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,6 +20,7 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
@@ -135,6 +136,10 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static void match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 List **matchedclauses,
+						 NullTestType *keynullness);
 
 
 /*
@@ -846,6 +851,279 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * match_clauses_to_partkey
+ *		Match the clauses in the list with partition's partition key
+ *
+ * Matched clauses are returned matchedclauses, which is an array with
+ * partnatts members, where each member is a list of clauses matched to the
+ * respective partition key.  keynullness array also contains partnatts
+ * members where each member corresponds to the type of the NullTest
+ * encountered for a given partition key.
+ */
+void
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 List **matchedclauses,
+						 NullTestType *keynullness)
+{
+	ListCell   *lc;
+	int		keyPos;
+	int		i;
+
+	/*
+	 * Match individual OpExprs in the query's restriction with individual
+	 * partition key columns.  There is one list per key.
+	 */
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType));
+	memset(matchedclauses, 0, PARTITION_MAX_KEYS * sizeof(List*));
+	keyPos = 0;
+	for (i = 0; i < rel->part_scheme->partnatts; i++)
+	{
+		Node   *partkey = linitial(rel->partexprs[i]);
+
+		foreach(lc, clauses)
+		{
+			RestrictInfo   *rinfo = lfirst(lc);
+			Expr		   *clause = rinfo->clause;
+
+			if (is_opclause(clause))
+			{
+				Node   *leftop = get_leftop(clause),
+					   *rightop = get_rightop(clause);
+				Oid		expr_op = ((OpExpr *) clause)->opno;
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = (Node *) ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					matchedclauses[keyPos] = lappend(matchedclauses[keyPos],
+													 clause);
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[keyPos] = IS_NOT_NULL;
+				}
+				else if (equal(rightop, partkey))
+				{
+					Oid		commutator = get_commutator(expr_op);
+
+					if (OidIsValid(commutator))
+					{
+						OpExpr   *commutated_expr;
+
+						/*
+						 * Generate a commutated copy of the expression, but
+						 * try to make it look valid, because we only want
+						 * it to put the constant operand in a place that the
+						 * following code knows as the only place to find it.
+						 */
+						commutated_expr = (OpExpr *) copyObject(clause);
+						commutated_expr->opno = commutator;	/* really? */
+						commutated_expr->args = list_make2(rightop, leftop);
+						matchedclauses[keyPos] =
+											lappend(matchedclauses[keyPos],
+													commutated_expr);
+						/* A strict operator implies NOT NULL argument. */
+						keynullness[keyPos] = IS_NOT_NULL;
+					}
+				}
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					keynullness[keyPos] = nulltest->nulltesttype;
+			}
+		}
+
+		/* Onto finding clauses matching the next partition key. */
+		keyPos++;
+	}
+}
+
+/*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the query clauses
+ *
+ * Returned list contains the AppendInfos of the chosen partitions.
+ */
+static List *
+get_rel_partitions(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
+{
+	Relation		parent = heap_open(rte->relid, NoLock);
+	List   *indexes;
+	List   *result = NIL;
+	ListCell   *lc1;
+	List   *matchedclauses[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	Datum	minkeys[PARTITION_MAX_KEYS],
+			maxkeys[PARTITION_MAX_KEYS];
+	bool	need_next_min,
+			need_next_max,
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_minkeys = 0,
+			n_maxkeys = 0,
+			i;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.
+	 */
+	match_clauses_to_partkey(rel,
+							 rel->baserestrictinfo,
+							 matchedclauses,
+							 keynullness);
+
+	/*
+	 * Determine the min keys and the max keys using btree semantics-based
+	 * interpretation of the clauses' operators.
+	 */
+
+	/*
+	 * XXX - There should be a step similar to _bt_preprocess_keys() here,
+	 * to eliminate any redundant scan keys for a given partition column.  For
+	 * example, among a <= 4 and a <= 5, we can only keep a <= 4 for being
+	 * more restrictive and discard a <= 5.  While doing that, we can also
+	 * check to see if there exists a contradictory combination of scan keys
+	 * that makes the query trivially false for all records in the table.
+	 */
+	memset(minkeys, 0, sizeof(minkeys));
+	memset(maxkeys, 0, sizeof(maxkeys));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < rel->part_scheme->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc1, matchedclauses[i])
+		{
+			Expr   *clause = lfirst(lc1);
+			Const  *rightop = (Const *) get_rightop(clause);
+			Oid		opno = ((OpExpr *) clause)->opno,
+					opfamily = rel->part_scheme->partopfamily[i];
+			StrategyNumber	strategy;
+
+			strategy = get_op_opfamily_strategy(opno, opfamily);
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+					}
+					if (strategy == BTLessStrategyNumber)
+						need_next_max = false;
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+					}
+					if (strategy == BTGreaterStrategyNumber)
+						need_next_min = false;
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkeys[i] = rightop->constvalue;
+						if (!minkey_set[i])
+							n_minkeys++;
+					}
+					minkey_set[i] = true;
+					min_incl = true;
+
+					if (need_next_max)
+					{
+						maxkeys[i] = rightop->constvalue;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+					}
+					maxkey_set[i] = true;
+					max_incl = true;
+					break;
+
+				/*
+				 * This might mean '<>', but we don't have anything for that
+				 * case yet.  Perhaps, handle that as key < const OR
+				 * key > const, once we have props needed for handling OR
+				 * clauses.
+				 */
+				default:
+					min_incl = max_incl = false;
+					break;
+			}
+		}
+	}
+
+	/* Ask partition.c which partitions it thinks match the keys. */
+	indexes = get_partitions_for_keys(parent, keynullness,
+									  minkeys, n_minkeys, min_incl,
+									  maxkeys, n_maxkeys, max_incl,
+									  &rel->painfo->min_datum_idx,
+									  &rel->painfo->max_datum_idx,
+									  &rel->painfo->contains_null_partition,
+								  &rel->painfo->contains_default_partition);
+
+	if (indexes != NIL)
+	{
+#ifdef USE_ASSERT_CHECKING
+		int		first_index,
+				last_index;
+		first_index = linitial_int(indexes);
+		last_index = llast_int(indexes);
+		Assert(first_index <= last_index ||
+			   rel->part_scheme->strategy != PARTITION_STRATEGY_RANGE);
+#endif
+
+		foreach(lc1, indexes)
+		{
+			int		partidx = lfirst_int(lc1);
+			AppendRelInfo *appinfo = rel->part_appinfos[partidx];
+#ifdef USE_ASSERT_CHECKING
+			RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+			PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+			Assert(partdesc->oids[partidx] == rte->relid);
+#endif
+			result = lappend(result, appinfo);
+		}
+	}
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->painfo->live_partition_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1138,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1152,24 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_rel_partitions(root, rel, rte);
+		Assert(rel->painfo != NULL);
+		rel->painfo->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1190,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1203,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -1118,6 +1411,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->painfo && rel->painfo)
+		{
+			rel->painfo->live_partitioned_rels =
+				list_concat(rel->painfo->live_partitioned_rels,
+					   list_copy(childrel->painfo->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1213,14 +1517,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->painfo->live_partition_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1291,33 +1610,31 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
 	/*
-	 * A root partition will already have a PartitionedChildRelInfo, and a
-	 * non-root partitioned table doesn't need one, because its Append paths
-	 * will get flattened into the parent anyway.  For a subquery RTE, no
-	 * PartitionedChildRelInfo exists; we collect all partitioned_rels
-	 * associated with any child.  (This assumes that we don't need to look
-	 * through multiple levels of subquery RTEs; if we ever do, we could
-	 * create a PartitionedChildRelInfo with the accumulated list of
-	 * partitioned_rels which would then be found when populated our parent
-	 * rel with paths.  For the present, that appears to be unnecessary.)
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
 	 */
 	rte = planner_rt_fetch(rel->relid, root);
-	switch (rte->rtekind)
-	{
-		case RTE_RELATION:
-			if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-				partitioned_rels =
-					get_partitioned_child_rels(root, rel->relid);
-			break;
-		case RTE_SUBQUERY:
-			build_partitioned_rels = true;
-			break;
-		default:
-			elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-	}
+	if (rte->rtekind == RTE_RELATION &&
+		rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->painfo->live_partitioned_rels;
 
 	/*
 	 * For every non-dummy child, remember the cheapest path.  Also, identify
@@ -1330,17 +1647,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->painfo)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+								   childrel->painfo->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index cac46bedf9..49578c0684 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1833,6 +1833,16 @@ set_relation_partition_info(PlannerInfo *root, RelOptInfo *rel,
 	rel->boundinfo = partdesc->boundinfo;
 	rel->nparts = partdesc->nparts;
 	rel->partexprs = build_baserel_partition_key_exprs(relation, rel->relid);
+
+	/*
+	 * A PartitionAppendInfo to map this table to its immediate partitions
+	 * that will be scanned by this query.  At the same time, it records the
+	 * table's partitioning properties reflecting any partition-pruning that
+	 * might've occurred to satisfy the query.  Rest of the fields are set in
+	 * get_rel_partitions() and set_append_rel_size().
+	 */
+	rel->painfo = makeNode(PartitionAppendInfo);
+	rel->painfo->boundinfo = partdesc->boundinfo;
 }
 
 /*
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 077e89ae43..24c8442eae 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -17,6 +17,7 @@
 #include <limits.h>
 
 #include "miscadmin.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -151,6 +152,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->boundinfo = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
+	rel->painfo = NULL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -227,8 +229,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -252,6 +258,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 454a940a23..4f27d3018e 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -99,6 +99,7 @@ extern int get_partition_for_tuple(PartitionDispatch *pd,
 						EState *estate,
 						PartitionDispatchData **failed_at,
 						TupleTableSlot **failed_slot);
+
 extern Oid	get_default_oid_from_partdesc(PartitionDesc partdesc);
 extern Oid	get_default_partition_oid(Oid parentId);
 extern void update_default_partition_oid(Oid parentId, Oid defaultPartId);
@@ -106,4 +107,12 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* Planner support stuff. */
+extern List *get_partitions_for_keys(Relation rel,
+						NullTestType *keynullness,
+						Datum *minkeys, int n_minkeys, bool min_inclusive,
+						Datum *maxkeys, int n_maxkeys, bool max_inclusive,
+						int *min_datum_index, int *max_datum_index,
+						bool *null_partition_chosen,
+						bool *default_partition_chosen);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/nodes.h b/src/include/nodes/nodes.h
index 27bd4f3363..63196a1211 100644
--- a/src/include/nodes/nodes.h
+++ b/src/include/nodes/nodes.h
@@ -261,6 +261,7 @@ typedef enum NodeTag
 	T_SpecialJoinInfo,
 	T_AppendRelInfo,
 	T_PartitionedChildRelInfo,
+	T_PartitionAppendInfo,
 	T_PlaceHolderInfo,
 	T_MinMaxAggInfo,
 	T_PlannerParamItem,
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index 48e6012f7f..6dfa28bd1b 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -524,6 +524,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs - Partition key expressions
  *
@@ -561,6 +562,9 @@ typedef enum RelOptKind
 /* Is the given relation an "other" relation? */
 #define IS_OTHER_REL(rel) ((rel)->reloptkind == RELOPT_OTHER_MEMBER_REL)
 
+typedef struct AppendRelInfo AppendRelInfo;
+typedef struct PartitionAppendInfo PartitionAppendInfo;
+
 typedef struct RelOptInfo
 {
 	NodeTag		type;
@@ -643,9 +647,19 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
 	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
 									 * stored in the same order of bounds */
 	List	  **partexprs;		/* Partition key expressions. */
+
+	/*
+	 * For a partitioned relation, the following represents the identities
+	 * of its live partitions (their appinfos) and some informations about
+	 * the bounds that the live partitions satisfy.
+	 */
+	PartitionAppendInfo *painfo;
 } RelOptInfo;
 
 /*
@@ -2085,6 +2099,58 @@ typedef struct PartitionedChildRelInfo
 	List	   *child_rels;
 } PartitionedChildRelInfo;
 
+/* Forward declarations, to avoid including other headers */
+typedef struct PartitionDispatchData *PartitionDispatch;
+typedef struct PartitionBoundInfoData *PartitionBoundInfo;
+typedef struct PartitionKeyData *PartitionKey;
+
+/*
+ * PartitionAppendInfo - Properties of partitions contained in the Append path
+ *						 of a given partitioned table
+ */
+typedef struct PartitionAppendInfo
+{
+	NodeTag		type;
+
+	/*
+	 * List of AppendRelInfos of the table's partitions that satisfy a given
+	 * query.
+	 */
+	List	   *live_partition_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
+
+	/*
+	 * The following simply copies the pointer to boundinfo in the table's
+	 * PartitionDesc.
+	 */
+	PartitionBoundInfo boundinfo;
+
+	/*
+	 * Indexes in the boundinfo->datums array of the smallest and the largest
+	 * value of the partition key that the query allows.  They are set by
+	 * calling get_partitions_for_keys().
+	 */
+	int		min_datum_idx;
+	int		max_datum_idx;
+
+	/*
+	 * Does this Append contain the null-accepting partition, if one exists
+	 * and is allowed by the query's quals.
+	 */
+	bool	contains_null_partition;
+
+	/*
+	 * Does this Append contain the default partition, if one exists and is
+	 * allowed by the query's quals.
+	 */
+	bool contains_default_partition;
+} PartitionAppendInfo;
+
 /*
  * For each distinct placeholder expression generated during planning, we
  * store a PlaceHolderInfo node in the PlannerInfo node's placeholder_list.
-- 
2.11.0

From 2830316f6d0f2e74a3e7960290df3a75aad10720 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 2/4] WIP: Interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index ccf8a1fa67..0133748234 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -111,6 +111,30 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the user-defined
+ * partition bound of a given existing partition, while an instance of the
+ * following struct describes either a new partition bound being compared
+ * against existing bounds (is_bound is true in that case and either lbound
+ * or rbound is set), or a new tuple's partition key specified in datums
+ * (ndatums = number of partition key columns).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -139,14 +163,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -755,10 +780,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -809,6 +840,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -830,8 +862,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -845,9 +880,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -2380,12 +2415,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -2397,6 +2435,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -2427,12 +2466,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -2629,12 +2669,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -2656,11 +2696,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -2668,17 +2708,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -2689,12 +2747,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -2708,20 +2767,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -2734,8 +2792,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From b39fe19609c4cdec4961455540cef50beebce33a Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/4] WIP: Implement get_partitions_for_keys()

Disable constraint exclusion that occurs using internal partition
constraints, so that it's apparent what the new partition-pruning
code still needs to do to able to create a plan matching the plain
the the traditional constraint exclusion based partition-pruning
would result in.
---
 src/backend/catalog/partition.c      | 210 ++++++++++++++++++++++++++++++++++-
 src/backend/optimizer/util/plancat.c |   4 +
 2 files changed, 209 insertions(+), 5 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 0133748234..9d4b7c1a7f 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1391,8 +1391,6 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
  *	datum that the query's bounding keys allow to be returned for the query.
  *	Similarly, *max_datum_index.  *null_partition_chosen returns whether
  *	the null partition will be scanned.
- *
- * TODO: Implement.
  */
 List *
 get_partitions_for_keys(Relation rel,
@@ -1403,12 +1401,214 @@ get_partitions_for_keys(Relation rel,
 						bool *null_partition_chosen,
 						bool *default_partition_chosen)
 {
+	int		i,
+			minoff,
+			maxoff;
 	List   *result = NIL;
-	int		i;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
 	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundCmpArg	arg;
+	bool	is_equal,
+			scan_default = false;
+	int		null_partition_idx = partdesc->boundinfo->null_index;
 
-	for (i = 0; i < partdesc->nparts; i++)
-		result = lappend_int(result, i);
+	*null_partition_chosen = false;
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if partdesc->boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keynullness[i] == IS_NULL)
+		{
+			if (null_partition_idx >= 0)
+			{
+				*null_partition_chosen = true;
+				result = list_make1_int(null_partition_idx);
+			}
+			else
+				result = NIL;
+
+			return result;
+		}
+	}
+
+	/*
+	 * If query provides no quals, don't forget to scan the default partition.
+	 */
+	if (n_minkeys == 0 && n_maxkeys == 0)
+		scan_default = true;
+
+	if (n_minkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = minkeys;
+		arg.ndatums = n_minkeys;
+		minoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (min_inclusive)
+					minoff -= 1;
+				else
+					minoff += 1;
+				if (minoff < 0 ||
+					minoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+												 minoff, &arg);
+			} while (cmpval == 0);
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !min_inclusive)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Records returned by the query will be > bounds[minoff],
+				 * because min_scankey is >= bounds[minoff], that is, no
+				 * records of the partition at minoff will be returned.  Go
+				 * to the next bound.
+				 */
+				if (minoff < partdesc->boundinfo->ndatums - 1)
+					minoff += 1;
+
+				/*
+				 * Make sure to skip a gap.
+				 * Note: There are ndatums + 1 lots in the indexes array.
+				 */
+				if (partdesc->boundinfo->indexes[minoff] < 0 &&
+					partdesc->boundinfo->indexes[minoff + 1] >= 0)
+					minoff += 1;
+				break;
+		}
+	}
+	else
+		minoff = 0;
+
+	if (n_maxkeys > 0 && partdesc->nparts > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = maxkeys;
+		arg.ndatums = n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, partdesc->boundinfo,
+											&arg, &is_equal);
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (max_inclusive)
+					maxoff += 1;
+				else
+					maxoff -= 1;
+				if (maxoff < 0 ||
+					maxoff >= partdesc->boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, partdesc->boundinfo,
+											 maxoff, &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (max_inclusive)
+				maxoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !max_inclusive)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Because bounds[maxoff] <= max_scankey, we may need to
+				 * to consider the next partition as well, in addition to
+				 * the partition at maxoff and earlier.
+				 */
+				if (!is_equal || max_inclusive)
+					maxoff += 1;
+
+				/* Make sure to skip a gap. */
+				if (partdesc->boundinfo->indexes[maxoff] < 0 && maxoff >= 1)
+					maxoff -= 1;
+				break;
+		}
+	}
+	else
+		maxoff = partdesc->boundinfo->ndatums - 1;
+
+	*min_datum_index = minoff;
+	*max_datum_index = maxoff;
+
+	switch (partkey->strategy)
+	{
+		case PARTITION_STRATEGY_LIST:
+			for (i = minoff; i <= maxoff; i++)
+			{
+				int		partition_idx = partdesc->boundinfo->indexes[i];
+
+				/*
+				 * Multiple values may belong to the same partition, so make
+				 * sure we don't add the same partition index again.
+				 */
+				result = list_append_unique_int(result, partition_idx);
+			}
+
+			/* If no bounding keys exist, include the null partition too. */
+			if (null_partition_idx >= 0 &&
+				(keynullness[0] == -1 || keynullness[0] != IS_NOT_NULL))
+			{
+				*null_partition_chosen = true;
+				result = list_append_unique_int(result, null_partition_idx);
+			}
+
+			break;
+
+		case PARTITION_STRATEGY_RANGE:
+			for (i = minoff; i <= maxoff; i++)
+			{
+				int		partition_idx = partdesc->boundinfo->indexes[i];
+
+				/*
+				 * If a valid partition exists for this range, add its
+				 * index, if not, the default partition (if any) would be
+				 * covering that range, so request to include the same.
+				 */
+				if (partition_idx >= 0)
+					result = lappend_int(result, partition_idx);
+				else
+					scan_default = true;
+			}
+			break;
+	}
+
+	if (scan_default && partdesc->boundinfo->default_index >= 0)
+		result = lappend_int(result, partdesc->boundinfo->default_index);
 
 	return result;
 }
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 49578c0684..ef84dac7f2 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1160,7 +1160,9 @@ get_relation_constraints(PlannerInfo *root,
 	Index		varno = rel->relid;
 	Relation	relation;
 	TupleConstr *constr;
+#ifdef USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	List	   *pcqual;
+#endif
 
 	/*
 	 * We assume the relation has already been safely locked.
@@ -1246,6 +1248,7 @@ get_relation_constraints(PlannerInfo *root,
 		}
 	}
 
+#ifdef USE_PARTITION_CONSTRAINT_FOR_PRUNING
 	/* Append partition predicates, if any */
 	pcqual = RelationGetPartitionQual(relation);
 	if (pcqual)
@@ -1263,6 +1266,7 @@ get_relation_constraints(PlannerInfo *root,
 
 		result = list_concat(result, pcqual);
 	}
+#endif
 
 	heap_close(relation, NoLock);
 
-- 
2.11.0

From 5c0a9b4e102f4f6b44cd56a812f3d54855b45c6e Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 4/4] Add more tests for the new partitioning-related planning
 code

---
 src/test/regress/expected/partition.out | 465 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      |  82 ++++++
 4 files changed, 549 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..5242f4aa3d
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,465 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_null
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(3 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+(7 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+(15 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+(5 rows)
+
+explain (costs off) select * from rlp where a > 10;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+(11 rows)
+
+explain (costs off) select * from rlp where a < 15;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+(5 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+(11 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(7 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where a > 30;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+(15 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(5 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(9 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+(15 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+(5 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+        QUERY PLAN         
+---------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+(7 rows)
+
+-- XXX - '<>' clauses cannot be handled yet
+explain (costs off) select * from lp where a <> 'a';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'a'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'a'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'a'::bpchar)
+(7 rows)
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p3
+         Filter: ((a > 1) AND (a = 10))
+   ->  Seq Scan on mc3p4
+         Filter: ((a > 1) AND (a = 10))
+(11 rows)
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlp3 where b = 'ab' or b = 'ef';
+                               QUERY PLAN                               
+------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((b)::text = 'ab'::text) OR ((b)::text = 'ef'::text))
+(7 rows)
+
+drop table lp, rlp, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 2fd3f2b1b1..2eb81fcf41 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 76b0de30a7..6611662149 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..4b562a655b
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,82 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* empty */
+explain (costs off) select * from rlp where a <= 31;
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+
+-- XXX - '<>' clauses cannot be handled yet
+explain (costs off) select * from lp where a <> 'a';
+
+-- XXX - redundant clause elimination does not happen yet
+explain (costs off) select * from mc3p where a = 10 and a > 1;
+
+-- XXX - the OR clauses don't contribute to partition-pruning yet
+explain (costs off) select * from rlp3 where b = 'ab' or b = 'ef';
+
+drop table lp, rlp, mc3p;
-- 
2.11.0

From a484fbf69a1debe97bbc3ef724ad858275a44688 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1034 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1031 insertions(+), 3 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index f8da91d0fe..abccb77393 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid repeated recomputation in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartitionScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartitionScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartitionScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								int rt_index, List *clauses);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static Datum partkey_datum_from_expr(const Expr *expr);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartitionScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1439,15 +1554,928 @@ get_partitions_from_clauses(Relation relation, int rt_index,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	PartitionSet   *partset;
+
+	partset = get_partitions_from_clauses_guts(relation, rt_index,
+											   partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_using_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, int rt_index,
+								 List *clauses)
+{
+	PartitionSet *partset;
+	PartitionScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset = partset_new(true, false);
+			List   *partconstr = RelationGetPartitionQual(relation);
+
+			/*
+			 * If this orarg refutes the table's partition constraint (if the
+			 * the table is a partition at all), don't go looking for its
+			 * partitions, that is, leave the partition set we're building
+			 * for this OR clause untouched.
+			 */
+			if (partconstr)
+			{
+				partconstr = (List *) expression_planner((Expr *) partconstr);
+				partconstr = (List *) canonicalize_qual((Expr *) partconstr);
+				Assert(rt_index > 0);
+				if (rt_index != 1)
+					ChangeVarNodes((Node *) partconstr, 1, rt_index, 0);
+
+				/*
+				 * NB: if the clause may contain Param, replace them with
+				 * equivalent Vars before proceeding, because predtest.c does
+				 * not know about Params.
+				 */
+				if (predicate_refuted_by(partconstr,
+										 list_make1(orarg), false))
+					continue;
+			}
+
+			arg_partset = get_partitions_from_clauses_guts(relation, 0,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+
+static Datum
+partkey_datum_from_expr(const Expr *expr)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			return ((Const *) expr)->constvalue;
+
+		default:
+			elog(ERROR, "invalid expression for partition key");
+	}
+
+	Assert(false);	/* should never get here! */
+	return 0;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(a) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that informatin in the output argument
+ *		*keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	memset(keynullness, 0, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc = palloc0(sizeof(PartClause));
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			keynullness[i] = -1;
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Does leftop match with this partition key column? */
+				if ((IsA(leftop, Var) && partattno != 0 &&
+					((Var *) leftop)->varattno == partattno) ||
+					equal(leftop, partexpr))
+				{
+					pc->op = opclause;
+					pc->constarg = rightop;
+					keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * We would've accepted this saop only if its operator's
+				 * negator was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					*or_clauses = lappend(*or_clauses, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					keynullness[i] = nulltest->nulltesttype;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is addmissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate PartClauseValSet */
+	for (i = 0; i < n_eqkeys; i++)
+		keys->eqkeys[i] = partkey_datum_from_expr(eqkey_exprs[i]);
+	keys->n_eqkeys = n_eqkeys;
+
+	for (i = 0; i < n_minkeys; i++)
+		keys->minkeys[i] = partkey_datum_from_expr(minkey_exprs[i]);
+	keys->n_minkeys = n_minkeys;
+	keys->min_incl = min_incl;
+
+	for (i = 0; i < n_maxkeys; i++)
+		keys->maxkeys[i] = partkey_datum_from_expr(maxkey_exprs[i]);
+	keys->n_maxkeys = n_maxkeys;
+	keys->max_incl = max_incl;
+
+	for (i = 0; i < partkey->partnatts; i++)
+		keys->keynullness[i] = keynullness[i];
+
+	return n_eqkeys + n_minkeys + n_maxkeys + n_keynullness;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	leftarg_const = partkey_datum_from_expr(leftarg->constarg);
+	rightarg_const = partkey_datum_from_expr(rightarg->constarg);
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions hat will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartitionScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 0248e01994a97a9845d60af4cc7bad7571f2370b Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index abccb77393..857cd5f707 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3475,12 +3510,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3492,6 +3530,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3522,12 +3561,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3724,12 +3764,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3751,11 +3791,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3763,17 +3803,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3784,12 +3842,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3803,20 +3862,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3829,8 +3887,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 7815b8538f85c257448709613d91d1be297e317a Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 365 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  64 ++----
 3 files changed, 385 insertions(+), 48 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 857cd5f707..b7cca33f1f 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2507,7 +2507,370 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal,
+			include_default = false;
+
+	/* Quick exit, if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * partset->empty will be set to true if we find out below that that's
+	 * the case.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the two can hold nulls at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition, or there could be both.  We already checked
+	 * above if we should return the null-accepting partition, but being
+	 * here means the query doesn't want nulls.  All remaining data must
+	 * be in the default partition if there is one.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		return partset;
+	}
+
+	/*
+	 * keys->eqkeys identifies a unique partition only if values for all
+	 * partition keys are provided.  If only a prefix of all the partition
+	 * keys is provided, then they identify a sequence of partitions each of
+	 * whose upper bound allows the values in that prefix.
+	 */
+	eqoff = -1;
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, key must exactly match datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * The bound at offset eqoff is <= eqkeys given how
+					 * partition_bound_bsearch works, so the partition we're
+					 * looking for is the one whose upper bound is at offset
+					 * eqoff + 1.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/* No minkeys and maxkeys to look at in this case. */
+		minoff = maxoff = -1;
+		goto collect_partitions;
+	}
+
+	/*
+	 * Using minkeys and maxkeys, identify the offsets of qualifying minimum
+	 * and maximum bounds.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (keys->min_incl)
+					minoff -= 1;
+				else
+					minoff += 1;
+				if (minoff < 0 || minoff >= boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, boundinfo, minoff,
+											 &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (!keys->min_incl)
+				minoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found a bound matching the query specified value, but this
+				 * may be a range query and we may have been asked us to
+				 * exclude the value itself.  So, go to the next bound value.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Rows returned by the query will be > the bound value at
+				 * minoff, because query's minkey is known to be >= that bound
+				 * value given how partition_bound_bsearch works.  IOW, no
+				 * rows of the partition whose upper bound is the value at
+				 * minoff will be returned, so go to the next bound value.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+		}
+	}
+
+	/*
+	 * Skip a gap, which might exist in the case of range partitioning, but
+	 * also instruct the later code to consider default partition (if one
+	 * exists) and could potentially contain data that satisfies the keys,
+	 * which it only could if there are still still finite keys left on
+	 * that side.
+	 */
+	if (minoff >= 0 && boundinfo->indexes[minoff] < 0)
+	{
+		if (keys->n_minkeys > 0 &&
+			boundinfo->kind[minoff][keys->n_minkeys - 1] ==
+			PARTITION_RANGE_DATUM_VALUE)
+			include_default = true;
+		minoff += 1;
+	}
+
+	maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (keys->max_incl)
+					maxoff += 1;
+				else
+					maxoff -= 1;
+				if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, boundinfo, maxoff,
+											 &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (keys->max_incl)
+				maxoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Because the bound value at maxoff is known to be <= query's
+				 * maxkey, we may need to consider the next partition as well.
+				 * That would be in addition to the partition whose upper
+				 * bound is the value at maxoff and earlier partitions.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Skip a gap, which might exist in the case of range partitioning, but
+	 * also instruct the later code to consider default partition (if one
+	 * exists) and could potentially contain data that satisfies the keys,
+	 * which it only could if there are still still finite keys left on
+	 * that side.
+	 */
+	if (maxoff >= 0 && boundinfo->indexes[maxoff] < 0)
+	{
+		if (keys->n_maxkeys > 0 &&
+			boundinfo->kind[maxoff][keys->n_maxkeys-1] ==
+			PARTITION_RANGE_DATUM_VALUE)
+			include_default = true;
+		maxoff -= 1;
+	}
+
+collect_partitions:
+	/*
+	 * eqoff set to a valid bound offset means a unique partition has been
+	 * identified.  Otherwise, generate a sequence of partitions corresponding
+	 * to bounds at offsets in range given by minoff and maxoff.
+	 */
+	if (eqoff >= 0)
+	{
+		if (boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->indexes[eqoff]);
+		else
+			include_default = true;
+	}
+	else if (minoff >= 0 && maxoff >= 0)
+	{
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* Add out-of-line partitions to the *other_parts set. */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				break;
+		}
+	}
+
+	/*
+	 * Check if we need to add the null-accepting partition to the set, if
+	 * it's not been already included by virtue of it being in the above
+	 * range of partitions.
+	 *
+	 * There are couple of cases when it will need to be scanned:
+	 *
+	 *	1. If there are no quals, null partition trivially needs to be
+	 *	scanned.
+	 *
+	 *	2. If it's not already contained in *other_parts and the query does
+	 *	not prevent nulls in the result.  If such a partition also accepts
+	 *	non-null datums, it would already be in *other_parts if one of those
+	 *	datums also qualify for the query.
+	 */
+	if (partition_bound_accepts_nulls(boundinfo) &&
+		!bms_is_member(boundinfo->null_index, partset->other_parts) &&
+		 keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0 &&
+		 (keys->keynullness[0] == -1 || keys->keynullness[0] != IS_NOT_NULL))
+	{
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->null_index);
+	}
+
+	/*
+	 * Check if we need to add the default partition to the set.  There are
+	 * couple of cases when it will need to be scanned:
+	 *
+	 *	1. If there are no quals, default partitions trivially needs to be
+	 *	scanned.
+	 *
+	 *	2. If there exist datums between those at offsets minoff and maxoff
+	 *	which don't have a partition that accepts them, then default partition
+	 *	would have caught them.  With range partitioning, simply check if
+	 *	there is a -1 in boundinfo->indexes, which indicates unassigned
+	 *	portion of range.  With list partitioning, if minoff != maxoff, it
+	 *	means there might be datums in that range that don't have a
+	 *	non-default partition assigned, whereas minoff == maxoff means only
+	 *	the partition containing that datum needs to be scanned.
+	 *	containing that datum needs to be scanned.
+	 */
+	if (boundinfo->default_index >= 0)
+	{
+		if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+			include_default = true;
+		else
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					include_default = (minoff != maxoff);
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					for (i = minoff; i <= maxoff; i++)
+					{
+						if (boundinfo->indexes[i] < 0)
+						{
+							include_default = true;
+							break;
+						}
+					}
+					break;
+			}
+		}
+
+		if (include_default)
+			partset->other_parts = bms_add_member(partset->other_parts,
+										  boundinfo->default_index);
+	}
+
+	if (partset->min_part_idx < 0 && partset->max_part_idx < 0 &&
+		partset->other_parts == NULL)
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 61c4596bc7..69a7819171 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -198,16 +198,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-          QUERY PLAN           
--------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -453,15 +451,13 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3nullxy
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(9 rows)
+(7 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -475,16 +471,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -502,9 +496,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -514,9 +506,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -530,9 +520,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -571,16 +559,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -592,9 +578,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -621,8 +605,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -630,9 +614,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -672,9 +654,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -712,9 +692,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -726,8 +704,6 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From 6410811627a92e3544385dd226f637d1399a4ae3 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 733 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 105 +++++
 4 files changed, 840 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..61c4596bc7
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,733 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default
+         Filter: ((a)::numeric = '1'::numeric)
+(17 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 10)
+(7 rows)
+
+explain (costs off) select * from rlp where a > 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 10)
+(13 rows)
+
+explain (costs off) select * from rlp where a < 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a < 15)
+(7 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 15)
+(13 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(7 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 30)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 31)
+(17 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(13 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a >= 15))
+(13 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(9 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+drop table lp, coll_pruning, rlp, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 53d4f49197..834f057d79 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index ed1df5ae24..f74615b0a2 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..d716fee69f
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,105 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From 5ba062828dfc83dff2a3521fce7532b2afc2c7b9 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning, although
    as of this commit this contains *all* appinfos.

 5. Code to handle the possibility that a partition RelOptInfo may
    not have the basic information set (set_append_rel_size() does
    that normally, but for partitioned tables, it will only do it
    for the *live* partitions, but partitionwise-join code would
    look at *all* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, do not call
get_partitions_from_clauses() right away.  Instead, store the clauses
(somewhere, such as in the Append plan node) until such a time as
when all the "constant" values in them will be available.  As of this
commit, we only pick up clauses from the baserestrictinfo list, so
it's safe to assume that each of the matched clause will provide the
constant value needed for pruning.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  24 ++
 src/backend/optimizer/path/allpaths.c | 694 ++++++++++++++++++++++++++++------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 +++++
 src/include/catalog/partition.h       |   6 +
 src/include/nodes/relation.h          |  31 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 779 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 07fdf66c38..f8da91d0fe 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,30 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, int rt_index,
+							List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 5535b63803..536ef22c58 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,15 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
+static BoolExpr *process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop);
 
 
 /*
@@ -834,6 +845,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +868,488 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, rel->relid, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's recorded in the
+ * PartClauseValSet as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and and that the latter is compatible with
+				 * partitioning.  If it is, we turn this into a OR BoolExpr:
+				 * (key < val OR key > val), if the partitioning method
+				 * supports such notion of inequlity.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+
+					if (partscheme->strategy == PARTITION_STRATEGY_RANGE ||
+						partscheme->strategy == PARTITION_STRATEGY_LIST)
+					{
+						BoolExpr *ne_or;
+
+						ne_or = process_partition_ne_op(rel, negator,
+														partopfamily,
+														partcoll,
+														(Expr *) leftop,
+														(Expr *) rightop);
+						result = lappend(result, ne_or);
+					}
+					continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * Check if saop_op is compatible with partitioning.  If so and
+				 * if this saop is of type 'key op ANY (...)', convert this into
+				 * a OR BoolExpr.
+				 */
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle its negator is indeed a part of the partitioning
+				 * operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+					negated = true;
+				}
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					result = lappend(result, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+static BoolExpr *
+process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop)
+{
+	Expr   *ltexpr,
+		   *gtexpr;
+	Oid		ltop,
+			gtop;
+	int		strategy;
+	Oid		lefttype,
+			righttype;
+
+	get_op_opfamily_properties(negator, partopfamily, false,
+							   &strategy, &lefttype, &righttype);
+	if (strategy != BTEqualStrategyNumber)
+		elog(LOG, "unexpected negator of '<>' operator");
+	ltop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTLessStrategyNumber);
+	gtop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTGreaterStrategyNumber);
+	ltexpr = make_opclause(ltop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+	gtexpr = make_opclause(gtop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+
+	return (BoolExpr *) makeBoolExpr(OR_EXPR, list_make2(ltexpr, gtexpr), -1);
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1364,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1378,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1415,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1428,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1438,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
-		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1608,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1714,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1807,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpcted rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1857,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index ecdd7280eb..d9bbf20acb 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6160,14 +6160,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..e74a87035e 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -101,6 +101,7 @@ extern int get_partition_for_tuple(PartitionDispatch *pd,
 						EState *estate,
 						PartitionDispatchData **failed_at,
 						TupleTableSlot **failed_slot);
+
 extern Oid	get_default_oid_from_partdesc(PartitionDesc partdesc);
 extern Oid	get_default_partition_oid(Oid parentId);
 extern void update_default_partition_oid(Oid parentId, Oid defaultPartId);
@@ -108,4 +109,9 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, int rt_index,
+								 List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..e47f6e5cd3 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -575,6 +581,8 @@ typedef enum RelOptKind
 	((rel)->reloptkind == RELOPT_OTHER_MEMBER_REL || \
 	 (rel)->reloptkind == RELOPT_OTHER_JOINREL)
 
+typedef struct AppendRelInfo AppendRelInfo;
+
 typedef struct RelOptInfo
 {
 	NodeTag		type;
@@ -657,10 +665,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 17fedc446f25a397d86d8e65ad9e6478f6252cd4 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 733 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 105 +++++
 4 files changed, 840 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..61c4596bc7
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,733 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default
+         Filter: ((a)::numeric = '1'::numeric)
+(17 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 10)
+(7 rows)
+
+explain (costs off) select * from rlp where a > 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 10)
+(13 rows)
+
+explain (costs off) select * from rlp where a < 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a < 15)
+(7 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 15)
+(13 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(7 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 30)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 31)
+(17 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(13 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a >= 15))
+(13 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(9 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+drop table lp, coll_pruning, rlp, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..d716fee69f
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,105 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From 6b8c2a466faae16578bd1727481576a6b727a905 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning, although
    as of this commit this contains *all* appinfos.

 5. Code to handle the possibility that a partition RelOptInfo may
    not have the basic information set (set_append_rel_size() does
    that normally, but for partitioned tables, it will only do it
    for the *live* partitions, but partitionwise-join code would
    look at *all* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, do not call
get_partitions_from_clauses() right away.  Instead, store the clauses
(somewhere, such as in the Append plan node) until such a time as
when all the "constant" values in them will be available.  As of this
commit, we only pick up clauses from the baserestrictinfo list, so
it's safe to assume that each of the matched clause will provide the
constant value needed for pruning.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  24 ++
 src/backend/optimizer/path/allpaths.c | 694 ++++++++++++++++++++++++++++------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 +++++
 src/include/catalog/partition.h       |   5 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 776 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 07fdf66c38..f8da91d0fe 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,30 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, int rt_index,
+							List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..862309263d 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,15 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
+static BoolExpr *process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop);
 
 
 /*
@@ -834,6 +845,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +868,488 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, rel->relid, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's recorded in the
+ * PartClauseValSet as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and and that the latter is compatible with
+				 * partitioning.  If it is, we turn this into a OR BoolExpr:
+				 * (key < val OR key > val), if the partitioning method
+				 * supports such notion of inequlity.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+
+					if (partscheme->strategy == PARTITION_STRATEGY_RANGE ||
+						partscheme->strategy == PARTITION_STRATEGY_LIST)
+					{
+						BoolExpr *ne_or;
+
+						ne_or = process_partition_ne_op(rel, negator,
+														partopfamily,
+														partcoll,
+														(Expr *) leftop,
+														(Expr *) rightop);
+						result = lappend(result, ne_or);
+					}
+					continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * Check if saop_op is compatible with partitioning.  If so and
+				 * if this saop is of type 'key op ANY (...)', convert this into
+				 * a OR BoolExpr.
+				 */
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle its negator is indeed a part of the partitioning
+				 * operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+					negated = true;
+				}
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					result = lappend(result, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+static BoolExpr *
+process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop)
+{
+	Expr   *ltexpr,
+		   *gtexpr;
+	Oid		ltop,
+			gtop;
+	int		strategy;
+	Oid		lefttype,
+			righttype;
+
+	get_op_opfamily_properties(negator, partopfamily, false,
+							   &strategy, &lefttype, &righttype);
+	if (strategy != BTEqualStrategyNumber)
+		elog(LOG, "unexpected negator of '<>' operator");
+	ltop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTLessStrategyNumber);
+	gtop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTGreaterStrategyNumber);
+	ltexpr = make_opclause(ltop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+	gtexpr = make_opclause(gtop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+
+	return (BoolExpr *) makeBoolExpr(OR_EXPR, list_make2(ltexpr, gtexpr), -1);
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1364,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1378,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1415,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1428,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1438,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
-		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1608,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1714,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1807,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1857,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index ecdd7280eb..d9bbf20acb 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6160,14 +6160,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..5f55550952 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,9 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, int rt_index,
+								 List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 44db5602dde3508c0037e3b026c3477387b90f07 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1034 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1031 insertions(+), 3 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index f8da91d0fe..abccb77393 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid repeated recomputation in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartitionScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartitionScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartitionScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								int rt_index, List *clauses);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static Datum partkey_datum_from_expr(const Expr *expr);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartitionScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1439,15 +1554,928 @@ get_partitions_from_clauses(Relation relation, int rt_index,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	PartitionSet   *partset;
+
+	partset = get_partitions_from_clauses_guts(relation, rt_index,
+											   partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_using_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, int rt_index,
+								 List *clauses)
+{
+	PartitionSet *partset;
+	PartitionScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset = partset_new(true, false);
+			List   *partconstr = RelationGetPartitionQual(relation);
+
+			/*
+			 * If this orarg refutes the table's partition constraint (if the
+			 * the table is a partition at all), don't go looking for its
+			 * partitions, that is, leave the partition set we're building
+			 * for this OR clause untouched.
+			 */
+			if (partconstr)
+			{
+				partconstr = (List *) expression_planner((Expr *) partconstr);
+				partconstr = (List *) canonicalize_qual((Expr *) partconstr);
+				Assert(rt_index > 0);
+				if (rt_index != 1)
+					ChangeVarNodes((Node *) partconstr, 1, rt_index, 0);
+
+				/*
+				 * NB: if the clause may contain Param, replace them with
+				 * equivalent Vars before proceeding, because predtest.c does
+				 * not know about Params.
+				 */
+				if (predicate_refuted_by(partconstr,
+										 list_make1(orarg), false))
+					continue;
+			}
+
+			arg_partset = get_partitions_from_clauses_guts(relation, 0,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+
+static Datum
+partkey_datum_from_expr(const Expr *expr)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			return ((Const *) expr)->constvalue;
+
+		default:
+			elog(ERROR, "invalid expression for partition key");
+	}
+
+	Assert(false);	/* should never get here! */
+	return 0;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(a) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that informatin in the output argument
+ *		*keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	memset(keynullness, 0, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc = palloc0(sizeof(PartClause));
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			keynullness[i] = -1;
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Does leftop match with this partition key column? */
+				if ((IsA(leftop, Var) && partattno != 0 &&
+					((Var *) leftop)->varattno == partattno) ||
+					equal(leftop, partexpr))
+				{
+					pc->op = opclause;
+					pc->constarg = rightop;
+					keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * We would've accepted this saop only if its operator's
+				 * negator was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					*or_clauses = lappend(*or_clauses, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					keynullness[i] = nulltest->nulltesttype;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is addmissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate PartClauseValSet */
+	for (i = 0; i < n_eqkeys; i++)
+		keys->eqkeys[i] = partkey_datum_from_expr(eqkey_exprs[i]);
+	keys->n_eqkeys = n_eqkeys;
+
+	for (i = 0; i < n_minkeys; i++)
+		keys->minkeys[i] = partkey_datum_from_expr(minkey_exprs[i]);
+	keys->n_minkeys = n_minkeys;
+	keys->min_incl = min_incl;
+
+	for (i = 0; i < n_maxkeys; i++)
+		keys->maxkeys[i] = partkey_datum_from_expr(maxkey_exprs[i]);
+	keys->n_maxkeys = n_maxkeys;
+	keys->max_incl = max_incl;
+
+	for (i = 0; i < partkey->partnatts; i++)
+		keys->keynullness[i] = keynullness[i];
+
+	return n_eqkeys + n_minkeys + n_maxkeys + n_keynullness;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	leftarg_const = partkey_datum_from_expr(leftarg->constarg);
+	rightarg_const = partkey_datum_from_expr(rightarg->constarg);
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions hat will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartitionScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 7ffaa0cbc1e63aa2b18350ee78f0b56bc94e4263 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index abccb77393..857cd5f707 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3475,12 +3510,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3492,6 +3530,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3522,12 +3561,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3724,12 +3764,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3751,11 +3791,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3763,17 +3803,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3784,12 +3842,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3803,20 +3862,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3829,8 +3887,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From c9a097462bc99b2450bbb9886fadbbf47555468f Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 365 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  64 ++----
 3 files changed, 385 insertions(+), 48 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 857cd5f707..b7cca33f1f 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2507,7 +2507,370 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal,
+			include_default = false;
+
+	/* Quick exit, if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * partset->empty will be set to true if we find out below that that's
+	 * the case.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the two can hold nulls at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition, or there could be both.  We already checked
+	 * above if we should return the null-accepting partition, but being
+	 * here means the query doesn't want nulls.  All remaining data must
+	 * be in the default partition if there is one.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		return partset;
+	}
+
+	/*
+	 * keys->eqkeys identifies a unique partition only if values for all
+	 * partition keys are provided.  If only a prefix of all the partition
+	 * keys is provided, then they identify a sequence of partitions each of
+	 * whose upper bound allows the values in that prefix.
+	 */
+	eqoff = -1;
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, key must exactly match datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * The bound at offset eqoff is <= eqkeys given how
+					 * partition_bound_bsearch works, so the partition we're
+					 * looking for is the one whose upper bound is at offset
+					 * eqoff + 1.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/* No minkeys and maxkeys to look at in this case. */
+		minoff = maxoff = -1;
+		goto collect_partitions;
+	}
+
+	/*
+	 * Using minkeys and maxkeys, identify the offsets of qualifying minimum
+	 * and maximum bounds.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (keys->min_incl)
+					minoff -= 1;
+				else
+					minoff += 1;
+				if (minoff < 0 || minoff >= boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, boundinfo, minoff,
+											 &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (!keys->min_incl)
+				minoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found a bound matching the query specified value, but this
+				 * may be a range query and we may have been asked us to
+				 * exclude the value itself.  So, go to the next bound value.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Rows returned by the query will be > the bound value at
+				 * minoff, because query's minkey is known to be >= that bound
+				 * value given how partition_bound_bsearch works.  IOW, no
+				 * rows of the partition whose upper bound is the value at
+				 * minoff will be returned, so go to the next bound value.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+		}
+	}
+
+	/*
+	 * Skip a gap, which might exist in the case of range partitioning, but
+	 * also instruct the later code to consider default partition (if one
+	 * exists) and could potentially contain data that satisfies the keys,
+	 * which it only could if there are still still finite keys left on
+	 * that side.
+	 */
+	if (minoff >= 0 && boundinfo->indexes[minoff] < 0)
+	{
+		if (keys->n_minkeys > 0 &&
+			boundinfo->kind[minoff][keys->n_minkeys - 1] ==
+			PARTITION_RANGE_DATUM_VALUE)
+			include_default = true;
+		minoff += 1;
+	}
+
+	maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+		if (is_equal && arg.ndatums < partkey->partnatts)
+		{
+			int32	cmpval;
+
+			is_equal = false;
+
+			do
+			{
+				if (keys->max_incl)
+					maxoff += 1;
+				else
+					maxoff -= 1;
+				if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+					break;
+				cmpval = partition_bound_cmp(partkey, boundinfo, maxoff,
+											 &arg);
+			} while (cmpval == 0);
+
+			/* Back up if went too far. */
+			if (keys->max_incl)
+				maxoff -= 1;
+		}
+
+		/* Interpret the result per partition strategy. */
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Found, but if the query may have asked us to exclude it.
+				 */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * Because the bound value at maxoff is known to be <= query's
+				 * maxkey, we may need to consider the next partition as well.
+				 * That would be in addition to the partition whose upper
+				 * bound is the value at maxoff and earlier partitions.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Skip a gap, which might exist in the case of range partitioning, but
+	 * also instruct the later code to consider default partition (if one
+	 * exists) and could potentially contain data that satisfies the keys,
+	 * which it only could if there are still still finite keys left on
+	 * that side.
+	 */
+	if (maxoff >= 0 && boundinfo->indexes[maxoff] < 0)
+	{
+		if (keys->n_maxkeys > 0 &&
+			boundinfo->kind[maxoff][keys->n_maxkeys-1] ==
+			PARTITION_RANGE_DATUM_VALUE)
+			include_default = true;
+		maxoff -= 1;
+	}
+
+collect_partitions:
+	/*
+	 * eqoff set to a valid bound offset means a unique partition has been
+	 * identified.  Otherwise, generate a sequence of partitions corresponding
+	 * to bounds at offsets in range given by minoff and maxoff.
+	 */
+	if (eqoff >= 0)
+	{
+		if (boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->indexes[eqoff]);
+		else
+			include_default = true;
+	}
+	else if (minoff >= 0 && maxoff >= 0)
+	{
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* Add out-of-line partitions to the *other_parts set. */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				break;
+		}
+	}
+
+	/*
+	 * Check if we need to add the null-accepting partition to the set, if
+	 * it's not been already included by virtue of it being in the above
+	 * range of partitions.
+	 *
+	 * There are couple of cases when it will need to be scanned:
+	 *
+	 *	1. If there are no quals, null partition trivially needs to be
+	 *	scanned.
+	 *
+	 *	2. If it's not already contained in *other_parts and the query does
+	 *	not prevent nulls in the result.  If such a partition also accepts
+	 *	non-null datums, it would already be in *other_parts if one of those
+	 *	datums also qualify for the query.
+	 */
+	if (partition_bound_accepts_nulls(boundinfo) &&
+		!bms_is_member(boundinfo->null_index, partset->other_parts) &&
+		 keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0 &&
+		 (keys->keynullness[0] == -1 || keys->keynullness[0] != IS_NOT_NULL))
+	{
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->null_index);
+	}
+
+	/*
+	 * Check if we need to add the default partition to the set.  There are
+	 * couple of cases when it will need to be scanned:
+	 *
+	 *	1. If there are no quals, default partitions trivially needs to be
+	 *	scanned.
+	 *
+	 *	2. If there exist datums between those at offsets minoff and maxoff
+	 *	which don't have a partition that accepts them, then default partition
+	 *	would have caught them.  With range partitioning, simply check if
+	 *	there is a -1 in boundinfo->indexes, which indicates unassigned
+	 *	portion of range.  With list partitioning, if minoff != maxoff, it
+	 *	means there might be datums in that range that don't have a
+	 *	non-default partition assigned, whereas minoff == maxoff means only
+	 *	the partition containing that datum needs to be scanned.
+	 *	containing that datum needs to be scanned.
+	 */
+	if (boundinfo->default_index >= 0)
+	{
+		if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+			include_default = true;
+		else
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					include_default = (minoff != maxoff);
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					for (i = minoff; i <= maxoff; i++)
+					{
+						if (boundinfo->indexes[i] < 0)
+						{
+							include_default = true;
+							break;
+						}
+					}
+					break;
+			}
+		}
+
+		if (include_default)
+			partset->other_parts = bms_add_member(partset->other_parts,
+										  boundinfo->default_index);
+	}
+
+	if (partset->min_part_idx < 0 && partset->max_part_idx < 0 &&
+		partset->other_parts == NULL)
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 61c4596bc7..69a7819171 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -198,16 +198,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-          QUERY PLAN           
--------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -453,15 +451,13 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3nullxy
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(9 rows)
+(7 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -475,16 +471,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -502,9 +496,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -514,9 +506,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -530,9 +520,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -571,16 +559,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -592,9 +578,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -621,8 +605,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -630,9 +614,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -672,9 +654,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -712,9 +692,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -726,8 +704,6 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From 823d3bff6e067172da3ba97b771e1bb32681f629 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 733 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 105 +++++
 4 files changed, 840 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..61c4596bc7
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,733 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default
+         Filter: ((a)::numeric = '1'::numeric)
+(17 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 10)
+(7 rows)
+
+explain (costs off) select * from rlp where a > 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 10)
+(13 rows)
+
+explain (costs off) select * from rlp where a < 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a < 15)
+(7 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 15)
+(13 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(7 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 30)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 31)
+(17 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(13 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a >= 15))
+(13 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(9 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+drop table lp, coll_pruning, rlp, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..d716fee69f
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,105 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From fa5e4f977d8db8ad54bb10fbb026d37bb329bd67 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning, although
    as of this commit this contains *all* appinfos.

 5. Code to handle the possibility that a partition RelOptInfo may
    not have the basic information set (set_append_rel_size() does
    that normally, but for partitioned tables, it will only do it
    for the *live* partitions, but partitionwise-join code would
    look at *all* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, do not call
get_partitions_from_clauses() right away.  Instead, store the clauses
(somewhere, such as in the Append plan node) until such a time as
when all the "constant" values in them will be available.  As of this
commit, we only pick up clauses from the baserestrictinfo list, so
it's safe to assume that each of the matched clause will provide the
constant value needed for pruning.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  24 ++
 src/backend/optimizer/path/allpaths.c | 694 ++++++++++++++++++++++++++++------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 +++++
 src/include/catalog/partition.h       |   5 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 776 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 07fdf66c38..f8da91d0fe 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,30 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, int rt_index,
+							List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..862309263d 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,15 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
+static BoolExpr *process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop);
 
 
 /*
@@ -834,6 +845,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +868,488 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, rel->relid, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's recorded in the
+ * PartClauseValSet as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and and that the latter is compatible with
+				 * partitioning.  If it is, we turn this into a OR BoolExpr:
+				 * (key < val OR key > val), if the partitioning method
+				 * supports such notion of inequlity.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+
+					if (partscheme->strategy == PARTITION_STRATEGY_RANGE ||
+						partscheme->strategy == PARTITION_STRATEGY_LIST)
+					{
+						BoolExpr *ne_or;
+
+						ne_or = process_partition_ne_op(rel, negator,
+														partopfamily,
+														partcoll,
+														(Expr *) leftop,
+														(Expr *) rightop);
+						result = lappend(result, ne_or);
+					}
+					continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * Check if saop_op is compatible with partitioning.  If so and
+				 * if this saop is of type 'key op ANY (...)', convert this into
+				 * a OR BoolExpr.
+				 */
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle its negator is indeed a part of the partitioning
+				 * operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+					negated = true;
+				}
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					result = lappend(result, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+static BoolExpr *
+process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop)
+{
+	Expr   *ltexpr,
+		   *gtexpr;
+	Oid		ltop,
+			gtop;
+	int		strategy;
+	Oid		lefttype,
+			righttype;
+
+	get_op_opfamily_properties(negator, partopfamily, false,
+							   &strategy, &lefttype, &righttype);
+	if (strategy != BTEqualStrategyNumber)
+		elog(LOG, "unexpected negator of '<>' operator");
+	ltop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTLessStrategyNumber);
+	gtop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTGreaterStrategyNumber);
+	ltexpr = make_opclause(ltop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+	gtexpr = make_opclause(gtop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+
+	return (BoolExpr *) makeBoolExpr(OR_EXPR, list_make2(ltexpr, gtexpr), -1);
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1364,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1378,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1415,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1428,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1438,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
-		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1608,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1714,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1807,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1857,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index ecdd7280eb..d9bbf20acb 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6160,14 +6160,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..5f55550952 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,9 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, int rt_index,
+								 List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 72ffddb1c510a256d862dd183fbbeed6a2d68958 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1034 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1031 insertions(+), 3 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index f8da91d0fe..362ebba75b 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid repeated recomputation in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartitionScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartitionScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartitionScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								int rt_index, List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static Datum partkey_datum_from_expr(const Expr *expr);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartitionScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1439,15 +1554,928 @@ get_partitions_from_clauses(Relation relation, int rt_index,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	PartitionSet   *partset;
+
+	partset = get_partitions_from_clauses_guts(relation, rt_index,
+											   partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_using_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, int rt_index,
+								 List *clauses)
+{
+	PartitionSet *partset;
+	PartitionScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset = partset_new(true, false);
+			List   *partconstr = RelationGetPartitionQual(relation);
+
+			/*
+			 * If this orarg refutes the table's partition constraint (if the
+			 * the table is a partition at all), don't go looking for its
+			 * partitions, that is, leave the partition set we're building
+			 * for this OR clause untouched.
+			 */
+			if (partconstr)
+			{
+				partconstr = (List *) expression_planner((Expr *) partconstr);
+				partconstr = (List *) canonicalize_qual((Expr *) partconstr);
+				Assert(rt_index > 0);
+				if (rt_index != 1)
+					ChangeVarNodes((Node *) partconstr, 1, rt_index, 0);
+
+				/*
+				 * NB: if the clause may contain Param, replace them with
+				 * equivalent Vars before proceeding, because predtest.c does
+				 * not know about Params.
+				 */
+				if (predicate_refuted_by(partconstr,
+										 list_make1(orarg), false))
+					continue;
+			}
+
+			arg_partset = get_partitions_from_clauses_guts(relation, 0,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(a) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that informatin in the output argument
+ *		*keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	memset(keynullness, 0, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc = palloc0(sizeof(PartClause));
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			keynullness[i] = -1;
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Does leftop match with this partition key column? */
+				if ((IsA(leftop, Var) && partattno != 0 &&
+					((Var *) leftop)->varattno == partattno) ||
+					equal(leftop, partexpr))
+				{
+					pc->op = opclause;
+					pc->constarg = rightop;
+					keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * We would've accepted this saop only if its operator's
+				 * negator was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					*or_clauses = lappend(*or_clauses, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					keynullness[i] = nulltest->nulltesttype;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is addmissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartitionScanKeyInfo));
+	for (i = 0; i < n_eqkeys; i++)
+		keys->eqkeys[i] = partkey_datum_from_expr(eqkey_exprs[i]);
+	keys->n_eqkeys = n_eqkeys;
+
+	for (i = 0; i < n_minkeys; i++)
+		keys->minkeys[i] = partkey_datum_from_expr(minkey_exprs[i]);
+	keys->n_minkeys = n_minkeys;
+	keys->min_incl = min_incl;
+
+	for (i = 0; i < n_maxkeys; i++)
+		keys->maxkeys[i] = partkey_datum_from_expr(maxkey_exprs[i]);
+	keys->n_maxkeys = n_maxkeys;
+	keys->max_incl = max_incl;
+
+	for (i = 0; i < partkey->partnatts; i++)
+		keys->keynullness[i] = keynullness[i];
+
+	return n_eqkeys + n_minkeys + n_maxkeys + n_keynullness;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static Datum
+partkey_datum_from_expr(const Expr *expr)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			return ((Const *) expr)->constvalue;
+
+		default:
+			elog(ERROR, "invalid expression for partition key");
+	}
+
+	Assert(false);	/* should never get here! */
+	return 0;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	leftarg_const = partkey_datum_from_expr(leftarg->constarg);
+	rightarg_const = partkey_datum_from_expr(rightarg->constarg);
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartitionScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From e7aa6826c055c3c02bd902bc9bee4ed3c31d644f Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 362ebba75b..73f4e7ab95 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3475,12 +3510,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3492,6 +3530,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3522,12 +3561,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3724,12 +3764,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3751,11 +3791,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3763,17 +3803,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3784,12 +3842,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3803,20 +3862,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3829,8 +3887,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 5760054b91f7f115b3690432d1ee72a3db7770f3 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 356 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  64 ++----
 3 files changed, 371 insertions(+), 53 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 73f4e7ab95..c5875dc064 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -178,7 +178,7 @@ typedef struct PartitionScanKeyInfo
 typedef struct PartitionSet
 {
 	/*
-	 * If either  empty or all_parts is true, values of the other fields are
+	 * If either  empty or all is true, values of the other fields are
 	 * invalid.
 	 */
 	bool	empty;				/* contains no partitions */
@@ -1121,7 +1121,7 @@ check_default_allows_bound(Relation parent, Relation default_rel,
 {
 	List	   *new_part_constraints;
 	List	   *def_part_constraints;
-	List	   *all_parts;
+	List	   *all;
 	ListCell   *lc;
 
 	new_part_constraints = (new_spec->strategy == PARTITION_STRATEGY_LIST)
@@ -1148,12 +1148,12 @@ check_default_allows_bound(Relation parent, Relation default_rel,
 	 * that do not satisfy the revised partition constraints.
 	 */
 	if (default_rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
-		all_parts = find_all_inheritors(RelationGetRelid(default_rel),
+		all = find_all_inheritors(RelationGetRelid(default_rel),
 										AccessExclusiveLock, NULL);
 	else
-		all_parts = list_make1_oid(RelationGetRelid(default_rel));
+		all = list_make1_oid(RelationGetRelid(default_rel));
 
-	foreach(lc, all_parts)
+	foreach(lc, all)
 	{
 		Oid			part_relid = lfirst_oid(lc);
 		Relation	part_rel;
@@ -2507,7 +2507,351 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff = -1,
+			minoff = -1,
+			maxoff = -1;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its outout.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provoide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	eqoff = -1;
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1,
+				 * then, would be the upper bound of the leftmost partition
+				 * that needs to be scanned.
+				 */
+				minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * itself is the upper bound of the rightmost partition that
+				 * needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	include_default = false;
+
+		/*
+		 * If the bound at minoff or maxoff looks like it's an upper bound of a
+		 * range of values unassigned to any partition, move to the adjacent
+		 * bound which instead must be the upper bound of the leftmost or
+		 * rightmost partition, respectively, that needs to be scanned.
+		 *
+		 * By doing that, we skip over a portion of values that do indeed
+		 * satisfy the query, but don't have a valid partition assigned.
+		 * Include the default partition in that case.  Although, if the
+		 * original bound in question is an infinite value, there would not
+		 * be any unassigned range, because the range is unbounded in that
+		 * direction by definition.
+		 */
+		if (boundinfo->indexes[minoff] < 0)
+		{
+			int		last_key;
+
+			Assert(partkey->strategy == PARTITION_STRATEGY_RANGE);
+			last_key = keys->n_minkeys > 0 ? keys->n_minkeys - 1
+										   : partkey->partnatts - 1;
+			if (boundinfo->kind[minoff][last_key] == PARTITION_RANGE_DATUM_VALUE)
+				include_default = true;
+			minoff += 1;
+		}
+
+		if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+		{
+			int		last_key;
+
+			Assert(partkey->strategy == PARTITION_STRATEGY_RANGE);
+			last_key = keys->n_maxkeys > 0 ? keys->n_maxkeys - 1
+										   : partkey->partnatts - 1;
+			maxoff -= 1;
+			if (boundinfo->kind[maxoff][last_key] == PARTITION_RANGE_DATUM_VALUE)
+				include_default = true;
+		}
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+				/*
+				 * If minoff != maxoff, there might be datums in that range
+				 * range that don't have a non-default partition assigned.
+				 */
+				include_default = (minoff != maxoff);
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						include_default = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if (include_default && partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 61c4596bc7..69a7819171 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -198,16 +198,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-          QUERY PLAN           
--------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -453,15 +451,13 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3nullxy
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(9 rows)
+(7 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -475,16 +471,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -502,9 +496,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -514,9 +506,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -530,9 +520,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -571,16 +559,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -592,9 +578,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -621,8 +605,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -630,9 +614,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -672,9 +654,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -712,9 +692,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -726,8 +704,6 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From b7b2db5d70865b7a1e4124e639aa36dfec59d984 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 770 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 108 +++++
 4 files changed, 880 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..c0365f0b52
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,770 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default
+         Filter: ((a)::numeric = '1'::numeric)
+(19 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 10)
+(7 rows)
+
+explain (costs off) select * from rlp where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 10)
+(15 rows)
+
+explain (costs off) select * from rlp where a < 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a < 15)
+(7 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 15)
+(15 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 30)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 31)
+(19 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(13 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a >= 15))
+(15 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+drop table lp, coll_pruning, rlp, mc3p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..093f676f26
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,108 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From 28dc86c17c8597a3015f86d74afad3d577d1d61b Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning, although
    as of this commit this contains *all* appinfos.

 5. Code to handle the possibility that a partition RelOptInfo may
    not have the basic information set (set_append_rel_size() does
    that normally, but for partitioned tables, it will only do it
    for the *live* partitions, but partitionwise-join code would
    look at *all* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, do not call
get_partitions_from_clauses() right away.  Instead, store the clauses
(somewhere, such as in the Append plan node) until such a time as
when all the "constant" values in them will be available.  As of this
commit, we only pick up clauses from the baserestrictinfo list, so
it's safe to assume that each of the matched clause will provide the
constant value needed for pruning.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  24 ++
 src/backend/optimizer/path/allpaths.c | 694 ++++++++++++++++++++++++++++------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 +++++
 src/include/catalog/partition.h       |   5 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 776 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 07fdf66c38..f8da91d0fe 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,30 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, int rt_index,
+							List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..862309263d 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,15 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
+static BoolExpr *process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop);
 
 
 /*
@@ -834,6 +845,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +868,488 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, rel->relid, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's recorded in the
+ * PartClauseValSet as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and and that the latter is compatible with
+				 * partitioning.  If it is, we turn this into a OR BoolExpr:
+				 * (key < val OR key > val), if the partitioning method
+				 * supports such notion of inequlity.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+
+					if (partscheme->strategy == PARTITION_STRATEGY_RANGE ||
+						partscheme->strategy == PARTITION_STRATEGY_LIST)
+					{
+						BoolExpr *ne_or;
+
+						ne_or = process_partition_ne_op(rel, negator,
+														partopfamily,
+														partcoll,
+														(Expr *) leftop,
+														(Expr *) rightop);
+						result = lappend(result, ne_or);
+					}
+					continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * Check if saop_op is compatible with partitioning.  If so and
+				 * if this saop is of type 'key op ANY (...)', convert this into
+				 * a OR BoolExpr.
+				 */
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle its negator is indeed a part of the partitioning
+				 * operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+					negated = true;
+				}
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					result = lappend(result, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+static BoolExpr *
+process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop)
+{
+	Expr   *ltexpr,
+		   *gtexpr;
+	Oid		ltop,
+			gtop;
+	int		strategy;
+	Oid		lefttype,
+			righttype;
+
+	get_op_opfamily_properties(negator, partopfamily, false,
+							   &strategy, &lefttype, &righttype);
+	if (strategy != BTEqualStrategyNumber)
+		elog(LOG, "unexpected negator of '<>' operator");
+	ltop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTLessStrategyNumber);
+	gtop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTGreaterStrategyNumber);
+	ltexpr = make_opclause(ltop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+	gtexpr = make_opclause(gtop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+
+	return (BoolExpr *) makeBoolExpr(OR_EXPR, list_make2(ltexpr, gtexpr), -1);
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1364,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1378,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1415,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1428,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1438,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
-		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1608,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1714,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1807,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1857,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index ecdd7280eb..d9bbf20acb 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6160,14 +6160,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..5f55550952 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,9 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, int rt_index,
+								 List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 575f8a4e84f2124bd32548d3d8d02dac85ae61a6 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1034 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1031 insertions(+), 3 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index f8da91d0fe..362ebba75b 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid repeated recomputation in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartitionScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartitionScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartitionScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								int rt_index, List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static Datum partkey_datum_from_expr(const Expr *expr);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartitionScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1439,15 +1554,928 @@ get_partitions_from_clauses(Relation relation, int rt_index,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	PartitionSet   *partset;
+
+	partset = get_partitions_from_clauses_guts(relation, rt_index,
+											   partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_using_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, int rt_index,
+								 List *clauses)
+{
+	PartitionSet *partset;
+	PartitionScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset = partset_new(true, false);
+			List   *partconstr = RelationGetPartitionQual(relation);
+
+			/*
+			 * If this orarg refutes the table's partition constraint (if the
+			 * the table is a partition at all), don't go looking for its
+			 * partitions, that is, leave the partition set we're building
+			 * for this OR clause untouched.
+			 */
+			if (partconstr)
+			{
+				partconstr = (List *) expression_planner((Expr *) partconstr);
+				partconstr = (List *) canonicalize_qual((Expr *) partconstr);
+				Assert(rt_index > 0);
+				if (rt_index != 1)
+					ChangeVarNodes((Node *) partconstr, 1, rt_index, 0);
+
+				/*
+				 * NB: if the clause may contain Param, replace them with
+				 * equivalent Vars before proceeding, because predtest.c does
+				 * not know about Params.
+				 */
+				if (predicate_refuted_by(partconstr,
+										 list_make1(orarg), false))
+					continue;
+			}
+
+			arg_partset = get_partitions_from_clauses_guts(relation, 0,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(a) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that informatin in the output argument
+ *		*keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	memset(keynullness, 0, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc = palloc0(sizeof(PartClause));
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			keynullness[i] = -1;
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Does leftop match with this partition key column? */
+				if ((IsA(leftop, Var) && partattno != 0 &&
+					((Var *) leftop)->varattno == partattno) ||
+					equal(leftop, partexpr))
+				{
+					pc->op = opclause;
+					pc->constarg = rightop;
+					keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * We would've accepted this saop only if its operator's
+				 * negator was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					*or_clauses = lappend(*or_clauses, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					keynullness[i] = nulltest->nulltesttype;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is addmissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartitionScanKeyInfo));
+	for (i = 0; i < n_eqkeys; i++)
+		keys->eqkeys[i] = partkey_datum_from_expr(eqkey_exprs[i]);
+	keys->n_eqkeys = n_eqkeys;
+
+	for (i = 0; i < n_minkeys; i++)
+		keys->minkeys[i] = partkey_datum_from_expr(minkey_exprs[i]);
+	keys->n_minkeys = n_minkeys;
+	keys->min_incl = min_incl;
+
+	for (i = 0; i < n_maxkeys; i++)
+		keys->maxkeys[i] = partkey_datum_from_expr(maxkey_exprs[i]);
+	keys->n_maxkeys = n_maxkeys;
+	keys->max_incl = max_incl;
+
+	for (i = 0; i < partkey->partnatts; i++)
+		keys->keynullness[i] = keynullness[i];
+
+	return n_eqkeys + n_minkeys + n_maxkeys + n_keynullness;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static Datum
+partkey_datum_from_expr(const Expr *expr)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			return ((Const *) expr)->constvalue;
+
+		default:
+			elog(ERROR, "invalid expression for partition key");
+	}
+
+	Assert(false);	/* should never get here! */
+	return 0;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	leftarg_const = partkey_datum_from_expr(leftarg->constarg);
+	rightarg_const = partkey_datum_from_expr(rightarg->constarg);
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartitionScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 22a1821c2223db8a6e6c3fa3afb3b7ce66a103af Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 362ebba75b..73f4e7ab95 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3475,12 +3510,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3492,6 +3530,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3522,12 +3561,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3724,12 +3764,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3751,11 +3791,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3763,17 +3803,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3784,12 +3842,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3803,20 +3862,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3829,8 +3887,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From b0f307b1c79f9313bb80c8185a16d5e2dba9424e Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 347 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  64 ++----
 3 files changed, 367 insertions(+), 48 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 73f4e7ab95..1e27f72d6e 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2507,7 +2507,352 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff = -1,
+			minoff = -1,
+			maxoff = -1;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its outout.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provoide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	eqoff = -1;
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * itself is the upper bound of the rightmost partition that
+				 * needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	include_default = false;
+
+		/*
+		 * If the bound at minoff or maxoff looks like it's an upper bound of a
+		 * range of values unassigned to any partition, move to the adjacent
+		 * bound which instead must be the upper bound of the leftmost or
+		 * rightmost partition, respectively, that needs to be scanned.
+		 *
+		 * By doing that, we skip over a portion of values that do indeed
+		 * satisfy the query, but don't have a valid partition assigned.
+		 * Include the default partition in that case.  Although, if the
+		 * original bound in question is an infinite value, there would not
+		 * be any unassigned range, because the range is unbounded in that
+		 * direction by definition.
+		 */
+		if (boundinfo->indexes[minoff] < 0)
+		{
+			int		last_key;
+
+			Assert(partkey->strategy == PARTITION_STRATEGY_RANGE);
+			last_key = keys->n_minkeys > 0 ? keys->n_minkeys - 1
+										   : partkey->partnatts - 1;
+			if (boundinfo->kind[minoff][last_key] == PARTITION_RANGE_DATUM_VALUE)
+				include_default = true;
+			minoff += 1;
+		}
+
+		if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+		{
+			int		last_key;
+
+			Assert(partkey->strategy == PARTITION_STRATEGY_RANGE);
+			last_key = keys->n_maxkeys > 0 ? keys->n_maxkeys - 1
+										   : partkey->partnatts - 1;
+			maxoff -= 1;
+			if (boundinfo->kind[maxoff][last_key] == PARTITION_RANGE_DATUM_VALUE)
+				include_default = true;
+		}
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+				/*
+				 * If minoff != maxoff, there might be datums in that range
+				 * range that don't have a non-default partition assigned.
+				 */
+				include_default = (minoff != maxoff);
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						include_default = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if (include_default && partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index c0365f0b52..5c64e60ebe 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -199,16 +199,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-          QUERY PLAN           
--------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -488,8 +486,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -498,7 +494,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -512,16 +508,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -539,9 +533,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -551,9 +543,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -567,9 +557,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -608,16 +596,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -629,9 +615,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -658,8 +642,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -667,9 +651,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -709,9 +691,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -749,9 +729,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -763,8 +741,6 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 drop table lp, coll_pruning, rlp, mc3p;
-- 
2.11.0

From 126db4e77ac7d6d684e1ec97430cbbdc5d142432 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 841 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 125 +++++
 4 files changed, 968 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..cda067da3a
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,841 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default
+         Filter: ((a)::numeric = '1'::numeric)
+(19 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 10)
+(7 rows)
+
+explain (costs off) select * from rlp where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 10)
+(15 rows)
+
+explain (costs off) select * from rlp where a < 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a < 15)
+(7 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 15)
+(15 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text < 'ab'::text) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+(5 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp5
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 30)
+(5 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 31)
+(19 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(13 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a >= 15))
+(15 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+explain select * from mc2p where a < 2;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p0  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p1  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p2  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+(9 rows)
+
+explain select * from mc2p where a = 2 and b < 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b < 1) AND (a = 2))
+(3 rows)
+
+explain select * from mc2p where a > 1;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p4  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p5  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+(9 rows)
+
+explain select * from mc2p where a = 1 and b > 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p2  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b > 1) AND (a = 1))
+(3 rows)
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..b71849959d
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,125 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30);
+create table rlp5 partition of rlp for values from (31) to (maxvalue);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+
+explain select * from mc2p where a < 2;
+explain select * from mc2p where a = 2 and b < 1;
+explain select * from mc2p where a > 1;
+explain select * from mc2p where a = 1 and b > 1;
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
-- 
2.11.0

From f478c8a815c9e354a586b48e2bcb86e2dc38672f Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning, although
    as of this commit this contains *all* appinfos.

 5. Code to handle the possibility that a partition RelOptInfo may
    not have the basic information set (set_append_rel_size() does
    that normally, but for partitioned tables, it will only do it
    for the *live* partitions, but partitionwise-join code would
    look at *all* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, do not call
get_partitions_from_clauses() right away.  Instead, store the clauses
(somewhere, such as in the Append plan node) until such a time as
when all the "constant" values in them will be available.  As of this
commit, we only pick up clauses from the baserestrictinfo list, so
it's safe to assume that each of the matched clause will provide the
constant value needed for pruning.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  24 ++
 src/backend/optimizer/path/allpaths.c | 694 ++++++++++++++++++++++++++++------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 +++++
 src/include/catalog/partition.h       |   5 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 776 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 07fdf66c38..f8da91d0fe 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,30 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, int rt_index,
+							List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..e4427b8a58 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,15 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
+static BoolExpr *process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop);
 
 
 /*
@@ -834,6 +845,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +868,488 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, rel->relid, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's recorded in the
+ * PartClauseValSet as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and that the latter is compatible with partitioning.
+				 * If it is, we turn this into a OR BoolExpr: (key < val OR
+				 * key > val), if the partitioning method supports such
+				 * notion of inequality.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+
+					if (partscheme->strategy == PARTITION_STRATEGY_RANGE ||
+						partscheme->strategy == PARTITION_STRATEGY_LIST)
+					{
+						BoolExpr *ne_or;
+
+						ne_or = process_partition_ne_op(rel, negator,
+														partopfamily,
+														partcoll,
+														(Expr *) leftop,
+														(Expr *) rightop);
+						result = lappend(result, ne_or);
+					}
+					continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * Check if saop_op is compatible with partitioning.  If so and
+				 * if this saop is of type 'key op ANY (...)', convert this into
+				 * a OR BoolExpr.
+				 */
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle its negator is indeed a part of the partitioning
+				 * operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+					negated = true;
+				}
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					result = lappend(result, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+static BoolExpr *
+process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop)
+{
+	Expr   *ltexpr,
+		   *gtexpr;
+	Oid		ltop,
+			gtop;
+	int		strategy;
+	Oid		lefttype,
+			righttype;
+
+	get_op_opfamily_properties(negator, partopfamily, false,
+							   &strategy, &lefttype, &righttype);
+	if (strategy != BTEqualStrategyNumber)
+		elog(LOG, "unexpected negator of '<>' operator");
+	ltop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTLessStrategyNumber);
+	gtop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTGreaterStrategyNumber);
+	ltexpr = make_opclause(ltop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+	gtexpr = make_opclause(gtop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+
+	return (BoolExpr *) makeBoolExpr(OR_EXPR, list_make2(ltexpr, gtexpr), -1);
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1364,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1378,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1415,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1428,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1438,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
-		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1608,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1714,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1807,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1857,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index d58635c887..24d800d8b7 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6182,14 +6182,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..5f55550952 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,9 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, int rt_index,
+								 List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 84b77f2380a1669b374d843b4954cb6e3f0880e5 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1036 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1032 insertions(+), 4 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index f8da91d0fe..8d508e549b 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid recomputing the same in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartitionScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartitionScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartitionScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								int rt_index, List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static Datum partkey_datum_from_expr(const Expr *expr);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartitionScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1422,7 +1537,7 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 }
 
 /*
- * get_partitions_using_clauses
+ * get_partitions_from_clauses
  *		Determine the set of partitions of relation that will satisfy all
  *		the clauses contained in partclauses
  *
@@ -1439,15 +1554,928 @@ get_partitions_from_clauses(Relation relation, int rt_index,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	PartitionSet   *partset;
+
+	partset = get_partitions_from_clauses_guts(relation, rt_index,
+											   partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_from_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, int rt_index,
+								 List *clauses)
+{
+	PartitionSet *partset;
+	PartitionScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset = partset_new(true, false);
+			List   *partconstr = RelationGetPartitionQual(relation);
+
+			/*
+			 * If this orarg refutes the table's partition constraint (if the
+			 * the table is a partition at all), don't go looking for its
+			 * partitions, that is, leave the partition set we're building
+			 * for this OR clause untouched.
+			 */
+			if (partconstr)
+			{
+				partconstr = (List *) expression_planner((Expr *) partconstr);
+				partconstr = (List *) canonicalize_qual((Expr *) partconstr);
+				Assert(rt_index > 0);
+				if (rt_index != 1)
+					ChangeVarNodes((Node *) partconstr, 1, rt_index, 0);
+
+				/*
+				 * NB: if the clause may contain Param, replace them with
+				 * equivalent Vars before proceeding, because predtest.c does
+				 * not know about Params.
+				 */
+				if (predicate_refuted_by(partconstr,
+										 list_make1(orarg), false))
+					continue;
+			}
+
+			arg_partset = get_partitions_from_clauses_guts(relation, 0,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(a) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that information in the output
+ *		argument keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartitionScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	memset(keynullness, 0, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc = palloc0(sizeof(PartClause));
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			keynullness[i] = -1;
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Does leftop match with this partition key column? */
+				if ((IsA(leftop, Var) && partattno != 0 &&
+					((Var *) leftop)->varattno == partattno) ||
+					equal(leftop, partexpr))
+				{
+					pc->op = opclause;
+					pc->constarg = rightop;
+					keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+					/* A strict operator implies NOT NULL argument. */
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * We would've accepted this saop only if its operator's
+				 * negator was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					*or_clauses = lappend(*or_clauses, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					keynullness[i] = nulltest->nulltesttype;
+					n_keynullness++;
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is admissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartitionScanKeyInfo));
+	for (i = 0; i < n_eqkeys; i++)
+		keys->eqkeys[i] = partkey_datum_from_expr(eqkey_exprs[i]);
+	keys->n_eqkeys = n_eqkeys;
+
+	for (i = 0; i < n_minkeys; i++)
+		keys->minkeys[i] = partkey_datum_from_expr(minkey_exprs[i]);
+	keys->n_minkeys = n_minkeys;
+	keys->min_incl = min_incl;
+
+	for (i = 0; i < n_maxkeys; i++)
+		keys->maxkeys[i] = partkey_datum_from_expr(maxkey_exprs[i]);
+	keys->n_maxkeys = n_maxkeys;
+	keys->max_incl = max_incl;
+
+	for (i = 0; i < partkey->partnatts; i++)
+		keys->keynullness[i] = keynullness[i];
+
+	return n_eqkeys + n_minkeys + n_maxkeys + n_keynullness;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static Datum
+partkey_datum_from_expr(const Expr *expr)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			return ((Const *) expr)->constvalue;
+
+		default:
+			elog(ERROR, "invalid expression for partition key");
+	}
+
+	Assert(false);	/* should never get here! */
+	return 0;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	leftarg_const = partkey_datum_from_expr(leftarg->constarg);
+	rightarg_const = partkey_datum_from_expr(rightarg->constarg);
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartitionScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 878357a06082c5ca156edb27212693d8d959d8f1 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 8d508e549b..1d6d1d042c 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3475,12 +3510,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3492,6 +3530,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3522,12 +3561,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3724,12 +3764,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3751,11 +3791,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3763,17 +3803,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3784,12 +3842,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3803,20 +3862,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3829,8 +3887,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 9ee1df62011c543cc558531593c300999237cb37 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 372 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  64 ++----
 3 files changed, 392 insertions(+), 48 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 1d6d1d042c..a4ce8db5e7 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2507,7 +2507,377 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartitionScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its output.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * at maxoff itself is the upper bound of the rightmost
+				 * partition that needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	/*
+	 * minoff/maxoff set to -1 means none of the datums in PartitionBoundInfo
+	 * satisfies minkeys/maxkeys.  If both are set to a valid datum offset,
+	 * that means there exists at least some datums (and hence partitions)
+	 * satisfying both minkeys and maxkeys.
+	 */
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	list_include_def = false,
+				range_include_def = false;
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+
+				/*
+				 * If the query doesn't specify either the lower or the upper
+				 * bound, consider including the default partition in the
+				 * result set, because the existing partitions may not cover
+				 * all of the values that such an unbounded range contains.
+				 *
+				 * Also, if minoff != maxoff, there might be datums in that
+				 * range that don't have a non-default partition assigned.
+				 */
+				if (keys->n_minkeys == 0 || keys->n_maxkeys == 0 ||
+					minoff != maxoff)
+					list_include_def = true;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If the bound at minoff or maxoff looks like it's an upper
+				 * bound of an unassigned range of values, move to the
+				 * adjacent bound which must be the upper bound of the
+				 * leftmost or rightmost partition, respectively, that needs
+				 * to be scanned.
+				 *
+				 * By doing that, we skip over a portion of values that do
+				 * indeed satisfy the query, but don't have a valid partition
+				 * assigned. The default partition would've been included to
+				 * cover those values.  Although, if the original bound in
+				 * question is an infinite value, there would not be any
+				 * unassigned range to speak of, because the range is unbounded
+				 * in that direction by definition, so no need to include the
+				 * default.
+				 */
+				if (boundinfo->indexes[minoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_minkeys > 0)
+						lastkey = keys->n_minkeys - 1;
+					if (boundinfo->kind[minoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					minoff += 1;
+				}
+
+				if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_maxkeys > 0)
+						lastkey = keys->n_maxkeys - 1;
+					if (boundinfo->kind[maxoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					maxoff -= 1;
+				}
+
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						range_include_def = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if ((list_include_def || range_include_def) &&
+			partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else if (partition_bound_has_default(boundinfo))
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->default_index);
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index cda067da3a..4c13a7ffa4 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -199,16 +199,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-          QUERY PLAN           
--------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -506,8 +504,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -516,7 +512,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -530,16 +526,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -557,9 +551,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -569,9 +561,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -585,9 +575,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -626,16 +614,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -647,9 +633,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -676,8 +660,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -685,9 +669,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -727,9 +709,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -767,9 +747,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -781,9 +759,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 -- a simpler multi-column keys case
 create table mc2p (a int, b int) partition by range (a, b);
-- 
2.11.0

From 575c88e4616b16d0a56bfc6f04bdf22c4976fa8b Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 918 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 135 +++++
 4 files changed, 1055 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..06e4b52632
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,918 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default
+         Filter: ((a)::numeric = '1'::numeric)
+(25 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 10)
+(7 rows)
+
+explain (costs off) select * from rlp where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_2
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 10)
+(21 rows)
+
+explain (costs off) select * from rlp where a < 15;
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a < 15)
+(7 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 15)
+(15 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(15 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text < 'ab'::text) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+(5 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 30)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default
+         Filter: (a > 30)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+          QUERY PLAN           
+-------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default
+         Filter: (a <= 31)
+(25 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(19 rows)
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 20) AND (a < 27))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 29;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a = 29)
+(3 rows)
+
+explain (costs off) select * from rlp where a >= 29;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_1
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default
+         Filter: (a >= 29)
+(9 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default
+         Filter: ((a > 1) AND (a >= 15))
+(21 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+explain select * from mc2p where a < 2;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p0  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p1  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p2  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+(9 rows)
+
+explain select * from mc2p where a = 2 and b < 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b < 1) AND (a = 2))
+(3 rows)
+
+explain select * from mc2p where a > 1;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p4  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p5  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+(9 rows)
+
+explain select * from mc2p where a = 1 and b > 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p2  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b > 1) AND (a = 1))
+(3 rows)
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..d79db585b8
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,135 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default;
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+explain (costs off) select * from rlp where a = 29;
+explain (costs off) select * from rlp where a >= 29;
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+
+explain select * from mc2p where a < 2;
+explain select * from mc2p where a = 2 and b < 1;
+explain select * from mc2p where a > 1;
+explain select * from mc2p where a = 1 and b > 1;
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
-- 
2.11.0

From d5b7ba8b57284cc2a4faab595f58fba7765e6b62 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning, although
    as of this commit this contains *all* appinfos.

 5. Code to handle the possibility that a partition RelOptInfo may
    not have the basic information set (set_append_rel_size() does
    that normally, but for partitioned tables, it will only do it
    for the *live* partitions, but partitionwise-join code would
    look at *all* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, do not call
get_partitions_from_clauses() right away.  Instead, store the clauses
(somewhere, such as in the Append plan node) until such a time as
when all the "constant" values in them will be available.  As of this
commit, we only pick up clauses from the baserestrictinfo list, so
it's safe to assume that each of the matched clause will provide the
constant value needed for pruning.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  24 ++
 src/backend/optimizer/path/allpaths.c | 694 ++++++++++++++++++++++++++++------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 +++++
 src/include/catalog/partition.h       |   5 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 776 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 07fdf66c38..f8da91d0fe 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,30 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, int rt_index,
+							List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..e4427b8a58 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,15 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
+static BoolExpr *process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop);
 
 
 /*
@@ -834,6 +845,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +868,488 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, rel->relid, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's recorded in the
+ * PartClauseValSet as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and that the latter is compatible with partitioning.
+				 * If it is, we turn this into a OR BoolExpr: (key < val OR
+				 * key > val), if the partitioning method supports such
+				 * notion of inequality.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+
+					if (partscheme->strategy == PARTITION_STRATEGY_RANGE ||
+						partscheme->strategy == PARTITION_STRATEGY_LIST)
+					{
+						BoolExpr *ne_or;
+
+						ne_or = process_partition_ne_op(rel, negator,
+														partopfamily,
+														partcoll,
+														(Expr *) leftop,
+														(Expr *) rightop);
+						result = lappend(result, ne_or);
+					}
+					continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * Check if saop_op is compatible with partitioning.  If so and
+				 * if this saop is of type 'key op ANY (...)', convert this into
+				 * a OR BoolExpr.
+				 */
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle its negator is indeed a part of the partitioning
+				 * operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+					negated = true;
+				}
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					result = lappend(result, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+static BoolExpr *
+process_partition_ne_op(RelOptInfo *rel,
+						Oid negator, Oid partopfamily, Oid partcoll,
+						Expr *leftop, Expr *rightop)
+{
+	Expr   *ltexpr,
+		   *gtexpr;
+	Oid		ltop,
+			gtop;
+	int		strategy;
+	Oid		lefttype,
+			righttype;
+
+	get_op_opfamily_properties(negator, partopfamily, false,
+							   &strategy, &lefttype, &righttype);
+	if (strategy != BTEqualStrategyNumber)
+		elog(LOG, "unexpected negator of '<>' operator");
+	ltop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTLessStrategyNumber);
+	gtop = get_opfamily_member(partopfamily, lefttype, righttype,
+							   BTGreaterStrategyNumber);
+	ltexpr = make_opclause(ltop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+	gtexpr = make_opclause(gtop, BOOLOID, false,
+						   (Expr *) leftop, (Expr *) rightop,
+						   InvalidOid, partcoll);
+
+	return (BoolExpr *) makeBoolExpr(OR_EXPR, list_make2(ltexpr, gtexpr), -1);
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1364,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1378,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1415,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1428,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1438,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
-		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1608,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1714,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1807,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1857,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index d58635c887..24d800d8b7 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6182,14 +6182,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..5f55550952 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,9 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, int rt_index,
+								 List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 2e0d01f82c7cc8e04c923e0ce23c22809b859f22 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1080 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1076 insertions(+), 4 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index f8da91d0fe..35e7d871ee 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid recomputing the same in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								int rt_index, List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static bool partkey_datum_from_expr(const Expr *expr, Datum *value);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1422,7 +1537,7 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 }
 
 /*
- * get_partitions_using_clauses
+ * get_partitions_from_clauses
  *		Determine the set of partitions of relation that will satisfy all
  *		the clauses contained in partclauses
  *
@@ -1439,15 +1554,972 @@ get_partitions_from_clauses(Relation relation, int rt_index,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	List		   *partconstr = RelationGetPartitionQual(relation);
+	PartitionSet   *partset;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+	partclauses = list_concat(partclauses, partconstr);
+	partset = get_partitions_from_clauses_guts(relation, rt_index,
+											   partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
+
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_from_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, int rt_index,
+								 List *clauses)
+{
+	PartitionSet *partset;
+	PartScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset = partset_new(true, false);
+			List   *partconstr = RelationGetPartitionQual(relation);
+
+			/*
+			 * If this orarg refutes the table's partition constraint (if the
+			 * the table is a partition at all), don't go looking for its
+			 * partitions, that is, leave the partition set we're building
+			 * for this OR clause untouched.
+			 */
+			if (partconstr)
+			{
+				partconstr = (List *) expression_planner((Expr *) partconstr);
+				partconstr = (List *) canonicalize_qual((Expr *) partconstr);
+				Assert(rt_index > 0);
+				if (rt_index != 1)
+					ChangeVarNodes((Node *) partconstr, 1, rt_index, 0);
+
+				/*
+				 * NB: if the clause may contain Param, replace them with
+				 * equivalent Vars before proceeding, because predtest.c does
+				 * not know about Params.
+				 */
+				if (predicate_refuted_by(partconstr,
+										 list_make1(orarg), false))
+					continue;
+			}
+
+			arg_partset = get_partitions_from_clauses_guts(relation, 0,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else if (b->all_parts)
+		return a;
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(b) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that information in the output
+ *		argument keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0,
+			n_total = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	/* -1 represents an invalid value of NullTestType. */
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc = palloc0(sizeof(PartClause));
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Does leftop match with this partition key column? */
+				if ((IsA(leftop, Var) && partattno != 0 &&
+					((Var *) leftop)->varattno == partattno) ||
+					equal(leftop, partexpr))
+				{
+					pc->op = opclause;
+					pc->constarg = rightop;
+					keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+					/* A strict operator implies NOT NULL argument. */
+					if (keynullness[i] == -1)
+					{
+						keynullness[i] = IS_NOT_NULL;
+						n_keynullness++;
+					}
+					only_bool_clauses = false;
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args);
+				Const  *arrconst = (Const *) lsecond(saop->args);
+				ArrayType *arrval = DatumGetArrayTypeP(arrconst->constvalue);
+				int16	elemlen;
+				bool	elembyval;
+				char	elemalign;
+				Datum  *elem_values;
+				bool   *elem_nulls;
+				int		num_elems;
+				List   *elem_exprs;
+				bool	negated = false;
+
+				/*
+				 * We would've accepted this saop only if its operator's
+				 * negator was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/* Build clauses for the individual values in the array. */
+				get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+									 &elemlen, &elembyval, &elemalign);
+				deconstruct_array(arrval,
+								  ARR_ELEMTYPE(arrval),
+								  elemlen, elembyval, elemalign,
+								  &elem_values, &elem_nulls,
+								  &num_elems);
+				elem_exprs = NIL;
+				for (i = 0; i < num_elems; i++)
+				{
+					Expr *elem_expr;
+
+					if (!elem_nulls[i])
+					{
+						Const  *rightop;
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						rightop = makeConst(ARR_ELEMTYPE(arrval),
+											-1, arrconst->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_expr = (Expr *) opexpr;
+					}
+					else
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+														  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_expr = (Expr *) nulltest;
+					}
+
+					elem_exprs = lappend(elem_exprs, elem_expr);
+				}
+
+				/* Build the OR clause and generate its PartClauseSetOr. */
+				if (saop->useOr)
+				{
+					BoolExpr *orexpr;
+
+					Assert(elem_exprs != NIL);
+					orexpr = (BoolExpr *) makeBoolExpr(OR_EXPR, elem_exprs,
+													   -1);
+					*or_clauses = lappend(*or_clauses, orexpr);
+				}
+				else
+					/*
+					 * To be ANDed with the clauses in the original list, just
+					 * like what we do for the arguments of Boolean AND clause
+					 * above.
+					 */
+					clauses = list_concat(clauses, elem_exprs);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					if (keynullness[i] == -1)
+					{
+						keynullness[i] = nulltest->nulltesttype;
+						n_keynullness++;
+					}
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is admissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartScanKeyInfo));
+	if (n_eqkeys + n_minkeys + n_maxkeys + n_keynullness > 0)
+	{
+		Datum	value;
+		int		n_datums_resolved;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_eqkeys; i++)
+		{
+			if (partkey_datum_from_expr(eqkey_exprs[i], &value))
+			{
+				keys->eqkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_eqkeys = n_datums_resolved;
+		n_total += keys->n_eqkeys;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_minkeys; i++)
+		{
+			if (partkey_datum_from_expr(minkey_exprs[i], &value))
+			{
+				keys->minkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_minkeys = n_datums_resolved;
+		n_total += keys->n_minkeys;
+		keys->min_incl = min_incl;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_maxkeys; i++)
+		{
+			if (partkey_datum_from_expr(maxkey_exprs[i], &value))
+			{
+				keys->maxkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_maxkeys = n_datums_resolved;
+		n_total += keys->n_maxkeys;
+		keys->max_incl = max_incl;
+
+		for (i = 0; i < partkey->partnatts; i++)
+			keys->keynullness[i] = keynullness[i];
+		n_total += n_keynullness;
+	}
+
+	return n_total;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static bool
+partkey_datum_from_expr(const Expr *expr, Datum *value)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			*value = ((Const *) expr)->constvalue;
+			return true;
+
+		default:
+			return false;
+	}
+
+	Assert(false);	/* don't ever get here */
+	return false;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	if (!partkey_datum_from_expr(leftarg->constarg, &leftarg_const))
+		return false;
+	if (!partkey_datum_from_expr(rightarg->constarg, &rightarg_const))
+		return false;
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From eb76d1a20bbc002c9c994a3177b0eebef83f025e Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 35e7d871ee..d9f12c5f42 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3519,12 +3554,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3536,6 +3574,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3566,12 +3605,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3768,12 +3808,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3795,11 +3835,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3807,17 +3847,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3828,12 +3886,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3847,20 +3906,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3873,8 +3931,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 7f87307b03a1ce507909bc8e086e480532d55869 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 376 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  72 ++----
 3 files changed, 398 insertions(+), 54 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index d9f12c5f42..770a2a1ac9 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2551,7 +2551,381 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its output.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+		/* 1 more index than datums in this case */
+		maxoff = boundinfo->ndatums;
+	else
+		maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * at maxoff itself is the upper bound of the rightmost
+				 * partition that needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	/*
+	 * minoff/maxoff set to -1 means none of the datums in PartitionBoundInfo
+	 * satisfies minkeys/maxkeys.  If both are set to a valid datum offset,
+	 * that means there exists at least some datums (and hence partitions)
+	 * satisfying both minkeys and maxkeys.
+	 */
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	list_include_def = false,
+				range_include_def = false;
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+
+				/*
+				 * If the query doesn't specify either the lower or the upper
+				 * bound, consider including the default partition in the
+				 * result set, because the existing partitions may not cover
+				 * all of the values that such an unbounded range contains.
+				 *
+				 * Also, if minoff != maxoff, there might be datums in that
+				 * range that don't have a non-default partition assigned.
+				 */
+				if (keys->n_minkeys == 0 || keys->n_maxkeys == 0 ||
+					minoff != maxoff)
+					list_include_def = true;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If the bound at minoff or maxoff looks like it's an upper
+				 * bound of an unassigned range of values, move to the
+				 * adjacent bound which must be the upper bound of the
+				 * leftmost or rightmost partition, respectively, that needs
+				 * to be scanned.
+				 *
+				 * By doing that, we skip over a portion of values that do
+				 * indeed satisfy the query, but don't have a valid partition
+				 * assigned. The default partition would've been included to
+				 * cover those values.  Although, if the original bound in
+				 * question is an infinite value, there would not be any
+				 * unassigned range to speak of, because the range is unbounded
+				 * in that direction by definition, so no need to include the
+				 * default.
+				 */
+				if (boundinfo->indexes[minoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_minkeys > 0)
+						lastkey = keys->n_minkeys - 1;
+					if (boundinfo->kind[minoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					minoff += 1;
+				}
+
+				if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_maxkeys > 0)
+						lastkey = keys->n_maxkeys - 1;
+					if (boundinfo->kind[maxoff - 1][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					maxoff -= 1;
+				}
+
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						range_include_def = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if ((list_include_def || range_include_def) &&
+			partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else if (partition_bound_has_default(boundinfo))
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->default_index);
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 06e4b52632..5ca3c45b6b 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -204,16 +204,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-          QUERY PLAN           
--------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -465,11 +463,9 @@ explain (costs off) select * from rlp where a <= 31;
          Filter: (a <= 31)
    ->  Seq Scan on rlp5_1
          Filter: (a <= 31)
-   ->  Seq Scan on rlp5_default
-         Filter: (a <= 31)
    ->  Seq Scan on rlp_default
          Filter: (a <= 31)
-(25 rows)
+(23 rows)
 
 explain (costs off) select * from rlp where a = 1 or a = 7;
               QUERY PLAN              
@@ -511,9 +507,7 @@ explain (costs off) select * from rlp where a > 20 and a < 27;
          Filter: ((a > 20) AND (a < 27))
    ->  Seq Scan on rlp4_2
          Filter: ((a > 20) AND (a < 27))
-   ->  Seq Scan on rlp4_default
-         Filter: ((a > 20) AND (a < 27))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 29;
            QUERY PLAN           
@@ -583,8 +577,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -593,7 +585,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -607,16 +599,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -634,9 +624,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -646,9 +634,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -662,9 +648,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -703,16 +687,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -724,9 +706,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -753,8 +733,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -762,9 +742,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -804,9 +782,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -844,9 +820,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -858,9 +832,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 -- a simpler multi-column keys case
 create table mc2p (a int, b int) partition by range (a, b);
-- 
2.11.0

From 83ca3621117c76375b8d52af2f70af0fe5350ce6 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 947 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 138 +++++
 4 files changed, 1087 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..44f8713319
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,947 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a)::numeric = '1'::numeric)
+(29 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 10)
+(9 rows)
+
+explain (costs off) select * from rlp where a > 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_2
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 10)
+(23 rows)
+
+explain (costs off) select * from rlp where a < 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a < 15)
+(9 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 15)
+(17 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(17 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text < 'ab'::text) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+(5 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 30)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 30)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+            QUERY PLAN            
+----------------------------------
+ Append
+   ->  Seq Scan on rlp_default_30
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 31)
+(29 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(23 rows)
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 20) AND (a < 27))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 29;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a = 29)
+(3 rows)
+
+explain (costs off) select * from rlp where a >= 29;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_1
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a >= 29)
+(11 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 1) AND (a >= 15))
+(23 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+explain select * from mc2p where a < 2;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p0  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p1  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p2  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+(9 rows)
+
+explain select * from mc2p where a = 2 and b < 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b < 1) AND (a = 2))
+(3 rows)
+
+explain select * from mc2p where a > 1;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p4  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p5  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+(9 rows)
+
+explain select * from mc2p where a = 1 and b > 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p2  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b > 1) AND (a = 1))
+(3 rows)
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..5fd8b0cd14
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,138 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+explain (costs off) select * from rlp where a = 29;
+explain (costs off) select * from rlp where a >= 29;
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+
+explain select * from mc2p where a < 2;
+explain select * from mc2p where a = 2 and b < 1;
+explain select * from mc2p where a > 1;
+explain select * from mc2p where a = 1 and b > 1;
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
-- 
2.11.0

From a4c506ce64f0a78381ebcc317e1343a22f148323 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning (although
    as of this commit this contains *all* appinfos as mentioned
    below).

 5. Some code in try_partition_wise_join in to handle the
    possibility that a partition RelOptInfo may not have the basic
    information set (note that as noted in 0, set_append_rel_size
    now sets such information for only the *live* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, get_partitions_from_clauses
would returns without pruning any partitions.  In most cases, it's
obvious in the planner that a set of clauses identified as matching
the partition key don't contain the constant values right away, in
which case, there is no need to call get_partitions_from_clauses
right away.  Instead, it should be deferred to another piece of code
which can receive the above list of clauses and runs at a time when
the constant values become available.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  24 ++
 src/backend/optimizer/path/allpaths.c | 566 +++++++++++++++++++++++++++-------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 ++++++
 src/include/catalog/partition.h       |   5 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 648 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 66ec214e02..31c47d23e1 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,30 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, int rt_index,
+							List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..aca372a0d2 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,12 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
 
 
 /*
@@ -834,6 +842,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +865,363 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, rel->relid, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's added to the
+ * result as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way. */
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and that the latter is compatible with partitioning.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				/* Check if saop_op is compatible with partitioning. */
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle the same if its negator is indeed a part of the
+				 * partitioning operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) rightop))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1236,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1250,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1287,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1300,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1310,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
-		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1480,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1586,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1679,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1729,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index d58635c887..24d800d8b7 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6182,14 +6182,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..5f55550952 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,9 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, int rt_index,
+								 List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 4a90dd60f7e31af25e6240b83dc121ff15463552 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1151 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1147 insertions(+), 4 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 31c47d23e1..41c30f9327 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid recomputing the same in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								int rt_index, List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static bool partkey_datum_from_expr(const Expr *expr, Datum *value);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1422,7 +1537,7 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 }
 
 /*
- * get_partitions_using_clauses
+ * get_partitions_from_clauses
  *		Determine the set of partitions of relation that will satisfy all
  *		the clauses contained in partclauses
  *
@@ -1439,15 +1554,1043 @@ get_partitions_from_clauses(Relation relation, int rt_index,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	List		   *partconstr = RelationGetPartitionQual(relation);
+	PartitionSet   *partset;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+	partclauses = list_concat(partclauses, partconstr);
+	partset = get_partitions_from_clauses_guts(relation, rt_index,
+											   partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
+
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_from_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, int rt_index,
+								 List *clauses)
+{
+	PartitionSet *partset;
+	PartScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset = partset_new(true, false);
+			List   *partconstr = RelationGetPartitionQual(relation);
+
+			/*
+			 * If this orarg refutes the table's partition constraint (if the
+			 * the table is a partition at all), don't go looking for its
+			 * partitions, that is, leave the partition set we're building
+			 * for this OR clause untouched.
+			 */
+			if (partconstr)
+			{
+				partconstr = (List *) expression_planner((Expr *) partconstr);
+				partconstr = (List *) canonicalize_qual((Expr *) partconstr);
+				Assert(rt_index > 0);
+				if (rt_index != 1)
+					ChangeVarNodes((Node *) partconstr, 1, rt_index, 0);
+
+				/*
+				 * NB: if the clause may contain Param, replace them with
+				 * equivalent Vars before proceeding, because predtest.c does
+				 * not know about Params.
+				 */
+				if (predicate_refuted_by(partconstr,
+										 list_make1(orarg), false))
+					continue;
+			}
+
+			arg_partset = get_partitions_from_clauses_guts(relation, 0,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else if (b->all_parts)
+		return a;
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(b) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that information in the output
+ *		argument keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0,
+			n_total = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	/* -1 represents an invalid value of NullTestType. */
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i],
+					partcoll = partkey->partcollation[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc;
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Skip if leftop doesn't match this partition key column. */
+				if ((!IsA(leftop, Var) ||
+					 ((Var *) leftop)->varattno != partattno) &&
+					!equal(leftop, partexpr))
+					continue;
+
+				/*
+				 * Deal with <> operators that the planner allows if it finds
+				 * out that <>'s negator is indeed a valid partopfamily member.
+				 * Make an equivalent OR expression and add to the *or_clauses
+				 * list.  That is, we convert a <> opclause into
+				 * (leftop < rightop) OR (leftop > rightop).
+				 */
+				if (!op_in_opfamily(opclause->opno, partopfamily) &&
+					(partkey->strategy == PARTITION_STRATEGY_RANGE ||
+					 partkey->strategy == PARTITION_STRATEGY_LIST))
+				{
+					Expr   *ltexpr,
+						   *gtexpr;
+					Oid		negator,
+							ltop,
+							gtop;
+					int		strategy;
+					Oid		lefttype,
+							righttype;
+
+					negator = get_negator(opclause->opno);
+					get_op_opfamily_properties(negator, partopfamily, false,
+											   &strategy,
+											   &lefttype, &righttype);
+					if (strategy != BTEqualStrategyNumber)
+						elog(LOG, "unexpected negator of '<>' operator");
+					ltop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTLessStrategyNumber);
+					gtop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTGreaterStrategyNumber);
+					ltexpr = make_opclause(ltop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					gtexpr = make_opclause(gtop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					*or_clauses = lappend(*or_clauses,
+									makeBoolExpr(OR_EXPR,
+												 list_make2(ltexpr, gtexpr),
+												 -1));
+					continue;
+				}
+
+				pc = palloc0(sizeof(PartClause));
+				pc->op = opclause;
+				pc->constarg = rightop;
+				keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+				/* A strict operator implies NOT NULL argument. */
+				if (keynullness[i] == -1)
+				{
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+				}
+				only_bool_clauses = false;
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+				List   *elem_exprs,
+					   *elem_clauses;
+				ListCell *lc1;
+				bool	negated = false;
+
+				/*
+				 * Planner must have accepted this saop iff saop_op's negator
+				 * was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/*
+				 * First generate a list of Const nodes, one for each array
+				 * element.
+				 */
+				elem_exprs = NIL;
+				if (IsA(rightop, Const))
+				{
+					Const *arr = (Const *) lsecond(saop->args);
+					ArrayType *arrval = DatumGetArrayTypeP(arr->constvalue);
+					int16	elemlen;
+					bool	elembyval;
+					char	elemalign;
+					Datum  *elem_values;
+					bool   *elem_nulls;
+					int		num_elems;
+
+					get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+										 &elemlen, &elembyval, &elemalign);
+					deconstruct_array(arrval,
+									  ARR_ELEMTYPE(arrval),
+									  elemlen, elembyval, elemalign,
+									  &elem_values, &elem_nulls,
+									  &num_elems);
+					for (i = 0; i < num_elems; i++)
+					{
+						if (!elem_nulls[i])
+							elem_exprs = lappend(elem_exprs,
+											makeConst(ARR_ELEMTYPE(arrval),
+											-1, arr->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval));
+						else
+							elem_exprs = lappend(elem_exprs,
+											makeNullConst(ARR_ELEMTYPE(arrval),
+														  -1,
+														  arr->constcollid));
+					}
+				}
+				else
+				{
+					ArrayExpr *arrexpr = castNode(ArrayExpr, rightop);
+
+					elem_exprs = list_copy(arrexpr->elements);
+				}
+
+				/*
+				 * Now generate a list of clauses, one for each array element,
+				 * of the form: saop_leftop saop_op elem_expr
+				 */
+				elem_clauses = NIL;
+				foreach(lc1, elem_exprs)
+				{
+					Const  *rightop = castNode(Const, lfirst(lc1));
+					Expr   *elem_clause;
+
+					if (rightop->constisnull)
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+															  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_clause = (Expr *) nulltest;
+					}
+					else
+					{
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_clause = (Expr *) opexpr;
+					}
+
+					elem_clauses = lappend(elem_clauses, elem_clause);
+				}
+
+				/*
+				 * Build the OR clause if needed or add the clauses to the end
+				 * of the list that's being processed currently.
+				 */
+				if (saop->useOr)
+					*or_clauses = lappend(*or_clauses,
+										  makeBoolExpr(OR_EXPR, elem_clauses,
+													   -1));
+				else
+					clauses = list_concat(clauses, elem_clauses);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					if (keynullness[i] == -1)
+					{
+						keynullness[i] = nulltest->nulltesttype;
+						n_keynullness++;
+					}
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is admissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartScanKeyInfo));
+	if (n_eqkeys + n_minkeys + n_maxkeys + n_keynullness > 0)
+	{
+		Datum	value;
+		int		n_datums_resolved;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_eqkeys; i++)
+		{
+			if (partkey_datum_from_expr(eqkey_exprs[i], &value))
+			{
+				keys->eqkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_eqkeys = n_datums_resolved;
+		n_total += keys->n_eqkeys;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_minkeys; i++)
+		{
+			if (partkey_datum_from_expr(minkey_exprs[i], &value))
+			{
+				keys->minkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_minkeys = n_datums_resolved;
+		n_total += keys->n_minkeys;
+		keys->min_incl = min_incl;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_maxkeys; i++)
+		{
+			if (partkey_datum_from_expr(maxkey_exprs[i], &value))
+			{
+				keys->maxkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_maxkeys = n_datums_resolved;
+		n_total += keys->n_maxkeys;
+		keys->max_incl = max_incl;
+
+		for (i = 0; i < partkey->partnatts; i++)
+			keys->keynullness[i] = keynullness[i];
+		n_total += n_keynullness;
+	}
+
+	return n_total;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static bool
+partkey_datum_from_expr(const Expr *expr, Datum *value)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			*value = ((Const *) expr)->constvalue;
+			return true;
+
+		default:
+			return false;
+	}
+
+	Assert(false);	/* don't ever get here */
+	return false;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	if (!partkey_datum_from_expr(leftarg->constarg, &leftarg_const))
+		return false;
+	if (!partkey_datum_from_expr(rightarg->constarg, &rightarg_const))
+		return false;
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 6ec99b7dedbd348ba1969f602eb6737817fbff41 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 41c30f9327..caf52f4210 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3590,12 +3625,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3607,6 +3645,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3637,12 +3676,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3839,12 +3879,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3866,11 +3906,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3878,17 +3918,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3899,12 +3957,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3918,20 +3977,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3944,8 +4002,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From dec7931f2cca258a26d3d04afd3b89be333d5836 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 376 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  72 ++----
 3 files changed, 398 insertions(+), 54 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index caf52f4210..6b663bb11f 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2622,7 +2622,381 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its output.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+		/* 1 more index than datums in this case */
+		maxoff = boundinfo->ndatums;
+	else
+		maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * at maxoff itself is the upper bound of the rightmost
+				 * partition that needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	/*
+	 * minoff/maxoff set to -1 means none of the datums in PartitionBoundInfo
+	 * satisfies minkeys/maxkeys.  If both are set to a valid datum offset,
+	 * that means there exists at least some datums (and hence partitions)
+	 * satisfying both minkeys and maxkeys.
+	 */
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	list_include_def = false,
+				range_include_def = false;
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+
+				/*
+				 * If the query doesn't specify either the lower or the upper
+				 * bound, consider including the default partition in the
+				 * result set, because the existing partitions may not cover
+				 * all of the values that such an unbounded range contains.
+				 *
+				 * Also, if minoff != maxoff, there might be datums in that
+				 * range that don't have a non-default partition assigned.
+				 */
+				if (keys->n_minkeys == 0 || keys->n_maxkeys == 0 ||
+					minoff != maxoff)
+					list_include_def = true;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If the bound at minoff or maxoff looks like it's an upper
+				 * bound of an unassigned range of values, move to the
+				 * adjacent bound which must be the upper bound of the
+				 * leftmost or rightmost partition, respectively, that needs
+				 * to be scanned.
+				 *
+				 * By doing that, we skip over a portion of values that do
+				 * indeed satisfy the query, but don't have a valid partition
+				 * assigned. The default partition would've been included to
+				 * cover those values.  Although, if the original bound in
+				 * question is an infinite value, there would not be any
+				 * unassigned range to speak of, because the range is unbounded
+				 * in that direction by definition, so no need to include the
+				 * default.
+				 */
+				if (boundinfo->indexes[minoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_minkeys > 0)
+						lastkey = keys->n_minkeys - 1;
+					if (boundinfo->kind[minoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					minoff += 1;
+				}
+
+				if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_maxkeys > 0)
+						lastkey = keys->n_maxkeys - 1;
+					if (boundinfo->kind[maxoff - 1][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					maxoff -= 1;
+				}
+
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						range_include_def = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if ((list_include_def || range_include_def) &&
+			partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else if (partition_bound_has_default(boundinfo))
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->default_index);
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 44f8713319..21267dad98 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -207,16 +207,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-              QUERY PLAN               
----------------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -482,15 +480,13 @@ explain (costs off) select * from rlp where a <= 31;
          Filter: (a <= 31)
    ->  Seq Scan on rlp5_1
          Filter: (a <= 31)
-   ->  Seq Scan on rlp5_default
-         Filter: (a <= 31)
    ->  Seq Scan on rlp_default_10
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_30
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_default
          Filter: (a <= 31)
-(29 rows)
+(27 rows)
 
 explain (costs off) select * from rlp where a = 1 or a = 7;
               QUERY PLAN              
@@ -536,9 +532,7 @@ explain (costs off) select * from rlp where a > 20 and a < 27;
          Filter: ((a > 20) AND (a < 27))
    ->  Seq Scan on rlp4_2
          Filter: ((a > 20) AND (a < 27))
-   ->  Seq Scan on rlp4_default
-         Filter: ((a > 20) AND (a < 27))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 29;
            QUERY PLAN           
@@ -612,8 +606,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -622,7 +614,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -636,16 +628,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -663,9 +653,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -675,9 +663,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -691,9 +677,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -732,16 +716,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -753,9 +735,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -782,8 +762,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -791,9 +771,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -833,9 +811,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -873,9 +849,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -887,9 +861,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 -- a simpler multi-column keys case
 create table mc2p (a int, b int) partition by range (a, b);
-- 
2.11.0

From 83ca3621117c76375b8d52af2f70af0fe5350ce6 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/5] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 947 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 138 +++++
 4 files changed, 1087 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..44f8713319
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,947 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a)::numeric = '1'::numeric)
+(29 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 10)
+(9 rows)
+
+explain (costs off) select * from rlp where a > 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_2
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 10)
+(23 rows)
+
+explain (costs off) select * from rlp where a < 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a < 15)
+(9 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 15)
+(17 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(17 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text < 'ab'::text) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+(5 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_default
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a > 30;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 30)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 30)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+            QUERY PLAN            
+----------------------------------
+ Append
+   ->  Seq Scan on rlp_default_30
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 31)
+(29 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(23 rows)
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 20) AND (a < 27))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 29;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a = 29)
+(3 rows)
+
+explain (costs off) select * from rlp where a >= 29;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_1
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a >= 29)
+(11 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 1) AND (a >= 15))
+(23 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+explain select * from mc2p where a < 2;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p0  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p1  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p2  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+(9 rows)
+
+explain select * from mc2p where a = 2 and b < 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b < 1) AND (a = 2))
+(3 rows)
+
+explain select * from mc2p where a > 1;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p4  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p5  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+(9 rows)
+
+explain select * from mc2p where a = 1 and b > 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p2  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b > 1) AND (a = 1))
+(3 rows)
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..5fd8b0cd14
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,138 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;		/* while we're on nulls */
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+explain (costs off) select * from rlp where a = 29;
+explain (costs off) select * from rlp where a >= 29;
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+
+explain select * from mc2p where a < 2;
+explain select * from mc2p where a = 2 and b < 1;
+explain select * from mc2p where a > 1;
+explain select * from mc2p where a = 1 and b > 1;
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
-- 
2.11.0

From 521ddc6eaabf11b3cbbe7b0c8b02ddcf5bce60f2 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/5] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning (although
    as of this commit this contains *all* appinfos as mentioned
    below).

 5. Some code in try_partition_wise_join in to handle the
    possibility that a partition RelOptInfo may not have the basic
    information set (note that as noted in 0, set_append_rel_size
    now sets such information for only the *live* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, get_partitions_from_clauses
would returns without pruning any partitions.  In most cases, it's
obvious in the planner that a set of clauses identified as matching
the partition key don't contain the constant values right away, in
which case, there is no need to call get_partitions_from_clauses
right away.  Instead, it should be deferred to another piece of code
which can receive the above list of clauses and runs at a time when
the constant values become available.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  23 ++
 src/backend/optimizer/path/allpaths.c | 566 +++++++++++++++++++++++++++-------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 ++++++
 src/include/catalog/partition.h       |   4 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 646 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 66ec214e02..0ed5fbea48 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,29 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..50f869448d 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,12 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
 
 
 /*
@@ -834,6 +842,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +865,363 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's added to the
+ * result as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way. */
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and that the latter is compatible with partitioning.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				/* Check if saop_op is compatible with partitioning. */
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle the same if its negator is indeed a part of the
+				 * partitioning operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) rightop))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1236,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1250,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1287,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1300,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1310,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
-		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1480,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1586,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1679,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1729,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index d58635c887..24d800d8b7 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6182,14 +6182,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..7da99a9f41 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,8 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 240f3da72b90be9927472b88080dfd6ce67fec04 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/5] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1124 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1120 insertions(+), 4 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 0ed5fbea48..a845288127 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid recomputing the same in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static bool partkey_datum_from_expr(const Expr *expr, Datum *value);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1422,7 +1537,7 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 }
 
 /*
- * get_partitions_using_clauses
+ * get_partitions_from_clauses
  *		Determine the set of partitions of relation that will satisfy all
  *		the clauses contained in partclauses
  *
@@ -1438,15 +1553,1016 @@ get_partitions_from_clauses(Relation relation, List *partclauses,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	List		   *partconstr = RelationGetPartitionQual(relation);
+	PartitionSet   *partset;
+
+	partclauses = list_concat(partclauses, partconstr);
+	partset = get_partitions_from_clauses_guts(relation, partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_from_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, List *clauses)
+{
+	PartitionSet *partset;
+	PartScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset;
+
+			arg_partset = get_partitions_from_clauses_guts(relation,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else if (b->all_parts)
+		return a;
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(b) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that information in the output
+ *		argument keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0,
+			n_total = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	/* -1 represents an invalid value of NullTestType. */
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i],
+					partcoll = partkey->partcollation[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc;
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Skip if leftop doesn't match this partition key column. */
+				if ((!IsA(leftop, Var) ||
+					 ((Var *) leftop)->varattno != partattno) &&
+					!equal(leftop, partexpr))
+					continue;
+
+				/*
+				 * Deal with <> operators that the planner allows if it finds
+				 * out that <>'s negator is indeed a valid partopfamily member.
+				 * Make an equivalent OR expression and add to the *or_clauses
+				 * list.  That is, we convert a <> opclause into
+				 * (leftop < rightop) OR (leftop > rightop).
+				 */
+				if (!op_in_opfamily(opclause->opno, partopfamily) &&
+					(partkey->strategy == PARTITION_STRATEGY_RANGE ||
+					 partkey->strategy == PARTITION_STRATEGY_LIST))
+				{
+					Expr   *ltexpr,
+						   *gtexpr;
+					Oid		negator,
+							ltop,
+							gtop;
+					int		strategy;
+					Oid		lefttype,
+							righttype;
+
+					negator = get_negator(opclause->opno);
+					get_op_opfamily_properties(negator, partopfamily, false,
+											   &strategy,
+											   &lefttype, &righttype);
+					if (strategy != BTEqualStrategyNumber)
+						elog(LOG, "unexpected negator of '<>' operator");
+					ltop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTLessStrategyNumber);
+					gtop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTGreaterStrategyNumber);
+					ltexpr = make_opclause(ltop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					gtexpr = make_opclause(gtop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					*or_clauses = lappend(*or_clauses,
+									makeBoolExpr(OR_EXPR,
+												 list_make2(ltexpr, gtexpr),
+												 -1));
+					continue;
+				}
+
+				pc = palloc0(sizeof(PartClause));
+				pc->op = opclause;
+				pc->constarg = rightop;
+				keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+				/* A strict operator implies NOT NULL argument. */
+				if (keynullness[i] == -1)
+				{
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+				}
+				only_bool_clauses = false;
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+				List   *elem_exprs,
+					   *elem_clauses;
+				ListCell *lc1;
+				bool	negated = false;
+
+				/*
+				 * Planner must have accepted this saop iff saop_op's negator
+				 * was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/*
+				 * First generate a list of Const nodes, one for each array
+				 * element.
+				 */
+				elem_exprs = NIL;
+				if (IsA(rightop, Const))
+				{
+					Const *arr = (Const *) lsecond(saop->args);
+					ArrayType *arrval = DatumGetArrayTypeP(arr->constvalue);
+					int16	elemlen;
+					bool	elembyval;
+					char	elemalign;
+					Datum  *elem_values;
+					bool   *elem_nulls;
+					int		num_elems;
+
+					get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+										 &elemlen, &elembyval, &elemalign);
+					deconstruct_array(arrval,
+									  ARR_ELEMTYPE(arrval),
+									  elemlen, elembyval, elemalign,
+									  &elem_values, &elem_nulls,
+									  &num_elems);
+					for (i = 0; i < num_elems; i++)
+					{
+						if (!elem_nulls[i])
+							elem_exprs = lappend(elem_exprs,
+											makeConst(ARR_ELEMTYPE(arrval),
+											-1, arr->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval));
+						else
+							elem_exprs = lappend(elem_exprs,
+											makeNullConst(ARR_ELEMTYPE(arrval),
+														  -1,
+														  arr->constcollid));
+					}
+				}
+				else
+				{
+					ArrayExpr *arrexpr = castNode(ArrayExpr, rightop);
+
+					elem_exprs = list_copy(arrexpr->elements);
+				}
+
+				/*
+				 * Now generate a list of clauses, one for each array element,
+				 * of the form: saop_leftop saop_op elem_expr
+				 */
+				elem_clauses = NIL;
+				foreach(lc1, elem_exprs)
+				{
+					Const  *rightop = castNode(Const, lfirst(lc1));
+					Expr   *elem_clause;
+
+					if (rightop->constisnull)
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+															  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_clause = (Expr *) nulltest;
+					}
+					else
+					{
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_clause = (Expr *) opexpr;
+					}
+
+					elem_clauses = lappend(elem_clauses, elem_clause);
+				}
+
+				/*
+				 * Build the OR clause if needed or add the clauses to the end
+				 * of the list that's being processed currently.
+				 */
+				if (saop->useOr)
+					*or_clauses = lappend(*or_clauses,
+										  makeBoolExpr(OR_EXPR, elem_clauses,
+													   -1));
+				else
+					clauses = list_concat(clauses, elem_clauses);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					if (keynullness[i] == -1)
+					{
+						keynullness[i] = nulltest->nulltesttype;
+						n_keynullness++;
+					}
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is admissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartScanKeyInfo));
+	if (n_eqkeys + n_minkeys + n_maxkeys + n_keynullness > 0)
+	{
+		Datum	value;
+		int		n_datums_resolved;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_eqkeys; i++)
+		{
+			if (partkey_datum_from_expr(eqkey_exprs[i], &value))
+			{
+				keys->eqkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_eqkeys = n_datums_resolved;
+		n_total += keys->n_eqkeys;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_minkeys; i++)
+		{
+			if (partkey_datum_from_expr(minkey_exprs[i], &value))
+			{
+				keys->minkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_minkeys = n_datums_resolved;
+		n_total += keys->n_minkeys;
+		keys->min_incl = min_incl;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_maxkeys; i++)
+		{
+			if (partkey_datum_from_expr(maxkey_exprs[i], &value))
+			{
+				keys->maxkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_maxkeys = n_datums_resolved;
+		n_total += keys->n_maxkeys;
+		keys->max_incl = max_incl;
+
+		for (i = 0; i < partkey->partnatts; i++)
+			keys->keynullness[i] = keynullness[i];
+		n_total += n_keynullness;
+	}
+
+	return n_total;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static bool
+partkey_datum_from_expr(const Expr *expr, Datum *value)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			*value = ((Const *) expr)->constvalue;
+			return true;
+
+		default:
+			return false;
+	}
+
+	Assert(false);	/* don't ever get here */
+	return false;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	if (!partkey_datum_from_expr(leftarg->constarg, &leftarg_const))
+		return false;
+	if (!partkey_datum_from_expr(rightarg->constarg, &rightarg_const))
+		return false;
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 3ece043a9644b6b10c86a074d996b9bfa78f9970 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/5] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index a845288127..0ab076b98c 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3562,12 +3597,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3579,6 +3617,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3609,12 +3648,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3811,12 +3851,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3838,11 +3878,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3850,17 +3890,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3871,12 +3929,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3890,20 +3949,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3916,8 +3974,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 4c5a0e7e5efc023c68fbfeeaf325fb59f341f8b1 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 5/5] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 376 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  72 ++----
 3 files changed, 398 insertions(+), 54 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 0ab076b98c..5cc304cca0 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2594,7 +2594,381 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its output.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+		/* 1 more index than datums in this case */
+		maxoff = boundinfo->ndatums;
+	else
+		maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * at maxoff itself is the upper bound of the rightmost
+				 * partition that needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	/*
+	 * minoff/maxoff set to -1 means none of the datums in PartitionBoundInfo
+	 * satisfies minkeys/maxkeys.  If both are set to a valid datum offset,
+	 * that means there exists at least some datums (and hence partitions)
+	 * satisfying both minkeys and maxkeys.
+	 */
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	list_include_def = false,
+				range_include_def = false;
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+
+				/*
+				 * If the query doesn't specify either the lower or the upper
+				 * bound, consider including the default partition in the
+				 * result set, because the existing partitions may not cover
+				 * all of the values that such an unbounded range contains.
+				 *
+				 * Also, if minoff != maxoff, there might be datums in that
+				 * range that don't have a non-default partition assigned.
+				 */
+				if (keys->n_minkeys == 0 || keys->n_maxkeys == 0 ||
+					minoff != maxoff)
+					list_include_def = true;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If the bound at minoff or maxoff looks like it's an upper
+				 * bound of an unassigned range of values, move to the
+				 * adjacent bound which must be the upper bound of the
+				 * leftmost or rightmost partition, respectively, that needs
+				 * to be scanned.
+				 *
+				 * By doing that, we skip over a portion of values that do
+				 * indeed satisfy the query, but don't have a valid partition
+				 * assigned. The default partition would've been included to
+				 * cover those values.  Although, if the original bound in
+				 * question is an infinite value, there would not be any
+				 * unassigned range to speak of, because the range is unbounded
+				 * in that direction by definition, so no need to include the
+				 * default.
+				 */
+				if (boundinfo->indexes[minoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_minkeys > 0)
+						lastkey = keys->n_minkeys - 1;
+					if (boundinfo->kind[minoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					minoff += 1;
+				}
+
+				if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_maxkeys > 0)
+						lastkey = keys->n_maxkeys - 1;
+					if (boundinfo->kind[maxoff - 1][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					maxoff -= 1;
+				}
+
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						range_include_def = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if ((list_include_def || range_include_def) &&
+			partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else if (partition_bound_has_default(boundinfo))
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->default_index);
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 44f8713319..21267dad98 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -207,16 +207,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-              QUERY PLAN               
----------------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -482,15 +480,13 @@ explain (costs off) select * from rlp where a <= 31;
          Filter: (a <= 31)
    ->  Seq Scan on rlp5_1
          Filter: (a <= 31)
-   ->  Seq Scan on rlp5_default
-         Filter: (a <= 31)
    ->  Seq Scan on rlp_default_10
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_30
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_default
          Filter: (a <= 31)
-(29 rows)
+(27 rows)
 
 explain (costs off) select * from rlp where a = 1 or a = 7;
               QUERY PLAN              
@@ -536,9 +532,7 @@ explain (costs off) select * from rlp where a > 20 and a < 27;
          Filter: ((a > 20) AND (a < 27))
    ->  Seq Scan on rlp4_2
          Filter: ((a > 20) AND (a < 27))
-   ->  Seq Scan on rlp4_default
-         Filter: ((a > 20) AND (a < 27))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 29;
            QUERY PLAN           
@@ -612,8 +606,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -622,7 +614,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -636,16 +628,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -663,9 +653,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -675,9 +663,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -691,9 +677,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -732,16 +716,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -753,9 +735,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -782,8 +762,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -791,9 +771,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -833,9 +811,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -873,9 +849,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -887,9 +861,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 -- a simpler multi-column keys case
 create table mc2p (a int, b int) partition by range (a, b);
-- 
2.11.0

From 15b02441e0194f21af99878605846045449c7067 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/6] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 986 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 140 +++++
 4 files changed, 1128 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..6c669ffdfc
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,986 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp_default_null partition of rlp_default for values in (null);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_null
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a)::numeric = '1'::numeric)
+(31 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 10)
+(9 rows)
+
+explain (costs off) select * from rlp where a > 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_2
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 10)
+(23 rows)
+
+explain (costs off) select * from rlp where a < 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a < 15)
+(9 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 15)
+(17 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(17 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text < 'ab'::text) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+(5 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a is not null;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp2
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_2
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp5_1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp5_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a IS NOT NULL)
+(29 rows)
+
+explain (costs off) select * from rlp where a > 30;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 30)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 30)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+            QUERY PLAN            
+----------------------------------
+ Append
+   ->  Seq Scan on rlp_default_30
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 31)
+(29 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_null
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(25 rows)
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 20) AND (a < 27))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 29;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a = 29)
+(3 rows)
+
+explain (costs off) select * from rlp where a >= 29;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_1
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a >= 29)
+(11 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 1) AND (a >= 15))
+(23 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+explain select * from mc2p where a < 2;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p0  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p1  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p2  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+(9 rows)
+
+explain select * from mc2p where a = 2 and b < 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b < 1) AND (a = 2))
+(3 rows)
+
+explain select * from mc2p where a > 1;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p4  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p5  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+(9 rows)
+
+explain select * from mc2p where a = 1 and b > 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p2  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b > 1) AND (a = 1))
+(3 rows)
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..75e8a58f36
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,140 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp_default_null partition of rlp_default for values in (null);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;
+explain (costs off) select * from rlp where a is not null;
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+explain (costs off) select * from rlp where a = 29;
+explain (costs off) select * from rlp where a >= 29;
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+
+explain select * from mc2p where a < 2;
+explain select * from mc2p where a = 2 and b < 1;
+explain select * from mc2p where a > 1;
+explain select * from mc2p where a = 1 and b > 1;
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
-- 
2.11.0

From 5c8a0c598eae43654c51fbebd3329b0d502fac4b Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/6] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning (although
    as of this commit this contains *all* appinfos as mentioned
    below).

 5. Some code in try_partition_wise_join in to handle the
    possibility that a partition RelOptInfo may not have the basic
    information set (note that as noted in 0, set_append_rel_size
    now sets such information for only the *live* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, get_partitions_from_clauses
would returns without pruning any partitions.  In most cases, it's
obvious in the planner that a set of clauses identified as matching
the partition key don't contain the constant values right away, in
which case, there is no need to call get_partitions_from_clauses
right away.  Instead, it should be deferred to another piece of code
which can receive the above list of clauses and runs at a time when
the constant values become available.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  23 ++
 src/backend/optimizer/path/allpaths.c | 568 +++++++++++++++++++++++++++-------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  |  90 ++++++
 src/include/catalog/partition.h       |   4 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 648 insertions(+), 118 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 5daa8a1c19..7b0e022865 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,29 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index 4e565b3c00..9e12e2b00e 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,12 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse);
 
 
 /*
@@ -834,6 +842,17 @@ set_foreign_size(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->rows = clamp_row_est(rel->rows);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
 /*
  * set_foreign_pathlist
  *		Build access paths for a foreign table RTE
@@ -846,6 +865,365 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_append_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &constfalse);
+
+	/*
+	 * Since the clauses in rel->baserestrictinfo should all contain Const
+	 * operands, it should be possible to prune partitions right away.
+	 */
+	if (partclauses != NIL && !constfalse)
+	{
+		get_partitions_from_clauses(parent, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's added to the
+ * result as well.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				(IsA(clause, Const) &&
+				 ((((Const *) clause)->constisnull) ||
+				  !DatumGetBool(((Const *) clause)->constvalue))))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way. */
+		if (IsA(clause, BoolExpr))
+		{
+			bool	constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &constfalse1) != NIL)
+				result = lappend(result, clause);
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and that the latter is compatible with partitioning.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				/* Check if saop_op is compatible with partitioning. */
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle the same if its negator is indeed a part of the
+				 * partitioning operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) rightop))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+					result = lappend(result, nulltest);
+			}
+		}
+	}
+
+	return result;
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1238,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1252,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1289,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1302,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,73 +1312,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
-		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1152,6 +1482,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1247,14 +1588,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1325,43 +1681,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1378,17 +1731,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index d58635c887..24d800d8b7 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6182,14 +6182,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..8e290e19b0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,82 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..7da99a9f41 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,8 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From bb6c8c3634ccb1b82fd2031b992af636eac09bd5 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/6] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1126 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1122 insertions(+), 4 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 7b0e022865..d3a039cd78 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid recomputing the same in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static bool partkey_datum_from_expr(const Expr *expr, Datum *value);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1422,7 +1537,7 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 }
 
 /*
- * get_partitions_using_clauses
+ * get_partitions_from_clauses
  *		Determine the set of partitions of relation that will satisfy all
  *		the clauses contained in partclauses
  *
@@ -1438,15 +1553,1018 @@ get_partitions_from_clauses(Relation relation, List *partclauses,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	List		   *partconstr = RelationGetPartitionQual(relation);
+	PartitionSet   *partset;
+
+	if (partconstr)
+		partconstr = (List *) expression_planner((Expr *) partconstr);
+	partclauses = list_concat(partclauses, partconstr);
+	partset = get_partitions_from_clauses_guts(relation, partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_from_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, List *clauses)
+{
+	PartitionSet *partset;
+	PartScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset;
+
+			arg_partset = get_partitions_from_clauses_guts(relation,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else if (b->all_parts)
+		return a;
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(b) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that information in the output
+ *		argument keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 in the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0,
+			n_total = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	/* -1 represents an invalid value of NullTestType. */
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType *));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i],
+					partcoll = partkey->partcollation[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc;
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Skip if leftop doesn't match this partition key column. */
+				if ((!IsA(leftop, Var) ||
+					 ((Var *) leftop)->varattno != partattno) &&
+					!equal(leftop, partexpr))
+					continue;
+
+				/*
+				 * Deal with <> operators that the planner allows if it finds
+				 * out that <>'s negator is indeed a valid partopfamily member.
+				 * Make an equivalent OR expression and add to the *or_clauses
+				 * list.  That is, we convert a <> opclause into
+				 * (leftop < rightop) OR (leftop > rightop).
+				 */
+				if (!op_in_opfamily(opclause->opno, partopfamily) &&
+					(partkey->strategy == PARTITION_STRATEGY_RANGE ||
+					 partkey->strategy == PARTITION_STRATEGY_LIST))
+				{
+					Expr   *ltexpr,
+						   *gtexpr;
+					Oid		negator,
+							ltop,
+							gtop;
+					int		strategy;
+					Oid		lefttype,
+							righttype;
+
+					negator = get_negator(opclause->opno);
+					get_op_opfamily_properties(negator, partopfamily, false,
+											   &strategy,
+											   &lefttype, &righttype);
+					if (strategy != BTEqualStrategyNumber)
+						elog(LOG, "unexpected negator of '<>' operator");
+					ltop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTLessStrategyNumber);
+					gtop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTGreaterStrategyNumber);
+					ltexpr = make_opclause(ltop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					gtexpr = make_opclause(gtop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					*or_clauses = lappend(*or_clauses,
+									makeBoolExpr(OR_EXPR,
+												 list_make2(ltexpr, gtexpr),
+												 -1));
+					continue;
+				}
+
+				pc = palloc0(sizeof(PartClause));
+				pc->op = opclause;
+				pc->constarg = rightop;
+				keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+				/* A strict operator implies NOT NULL argument. */
+				if (keynullness[i] == -1)
+				{
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+				}
+				only_bool_clauses = false;
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+				List   *elem_exprs,
+					   *elem_clauses;
+				ListCell *lc1;
+				bool	negated = false;
+
+				/*
+				 * Planner must have accepted this saop iff saop_op's negator
+				 * was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/*
+				 * First generate a list of Const nodes, one for each array
+				 * element.
+				 */
+				elem_exprs = NIL;
+				if (IsA(rightop, Const))
+				{
+					Const *arr = (Const *) lsecond(saop->args);
+					ArrayType *arrval = DatumGetArrayTypeP(arr->constvalue);
+					int16	elemlen;
+					bool	elembyval;
+					char	elemalign;
+					Datum  *elem_values;
+					bool   *elem_nulls;
+					int		num_elems;
+
+					get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+										 &elemlen, &elembyval, &elemalign);
+					deconstruct_array(arrval,
+									  ARR_ELEMTYPE(arrval),
+									  elemlen, elembyval, elemalign,
+									  &elem_values, &elem_nulls,
+									  &num_elems);
+					for (i = 0; i < num_elems; i++)
+					{
+						if (!elem_nulls[i])
+							elem_exprs = lappend(elem_exprs,
+											makeConst(ARR_ELEMTYPE(arrval),
+											-1, arr->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval));
+						else
+							elem_exprs = lappend(elem_exprs,
+											makeNullConst(ARR_ELEMTYPE(arrval),
+														  -1,
+														  arr->constcollid));
+					}
+				}
+				else
+				{
+					ArrayExpr *arrexpr = castNode(ArrayExpr, rightop);
+
+					elem_exprs = list_copy(arrexpr->elements);
+				}
+
+				/*
+				 * Now generate a list of clauses, one for each array element,
+				 * of the form: saop_leftop saop_op elem_expr
+				 */
+				elem_clauses = NIL;
+				foreach(lc1, elem_exprs)
+				{
+					Const  *rightop = castNode(Const, lfirst(lc1));
+					Expr   *elem_clause;
+
+					if (rightop->constisnull)
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+															  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_clause = (Expr *) nulltest;
+					}
+					else
+					{
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_clause = (Expr *) opexpr;
+					}
+
+					elem_clauses = lappend(elem_clauses, elem_clause);
+				}
+
+				/*
+				 * Build the OR clause if needed or add the clauses to the end
+				 * of the list that's being processed currently.
+				 */
+				if (saop->useOr)
+					*or_clauses = lappend(*or_clauses,
+										  makeBoolExpr(OR_EXPR, elem_clauses,
+													   -1));
+				else
+					clauses = list_concat(clauses, elem_clauses);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					if (keynullness[i] == -1)
+					{
+						keynullness[i] = nulltest->nulltesttype;
+						n_keynullness++;
+					}
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is admissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartScanKeyInfo));
+	if (n_eqkeys + n_minkeys + n_maxkeys + n_keynullness > 0)
+	{
+		Datum	value;
+		int		n_datums_resolved;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_eqkeys; i++)
+		{
+			if (partkey_datum_from_expr(eqkey_exprs[i], &value))
+			{
+				keys->eqkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_eqkeys = n_datums_resolved;
+		n_total += keys->n_eqkeys;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_minkeys; i++)
+		{
+			if (partkey_datum_from_expr(minkey_exprs[i], &value))
+			{
+				keys->minkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_minkeys = n_datums_resolved;
+		n_total += keys->n_minkeys;
+		keys->min_incl = min_incl;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_maxkeys; i++)
+		{
+			if (partkey_datum_from_expr(maxkey_exprs[i], &value))
+			{
+				keys->maxkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_maxkeys = n_datums_resolved;
+		n_total += keys->n_maxkeys;
+		keys->max_incl = max_incl;
+
+		for (i = 0; i < partkey->partnatts; i++)
+			keys->keynullness[i] = keynullness[i];
+		n_total += n_keynullness;
+	}
+
+	return n_total;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static bool
+partkey_datum_from_expr(const Expr *expr, Datum *value)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			*value = ((Const *) expr)->constvalue;
+			return true;
+
+		default:
+			return false;
+	}
+
+	Assert(false);	/* don't ever get here */
+	return false;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	if (!partkey_datum_from_expr(leftarg->constarg, &leftarg_const))
+		return false;
+	if (!partkey_datum_from_expr(rightarg->constarg, &rightarg_const))
+		return false;
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 90205c145922b0b0faf6f63452e9c0dfb4f63b94 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/6] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index d3a039cd78..ca55d65c4b 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3564,12 +3599,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3581,6 +3619,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3611,12 +3650,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3813,12 +3853,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3840,11 +3880,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3852,17 +3892,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3873,12 +3931,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3892,20 +3951,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3918,8 +3976,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From d6b496920133088a33d4a5bf417b220bd9d271a1 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 31 Oct 2017 16:26:55 +0900
Subject: [PATCH 5/6] Tweak default range partition's constraint a little

When using as a predicate, it's useful for it explicitly say that
the default range partition might contain nulls, because non-default
range partitions can't.
---
 src/backend/catalog/partition.c      | 29 +++++++++++++++++++++++------
 src/test/regress/expected/update.out |  2 +-
 2 files changed, 24 insertions(+), 7 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index ca55d65c4b..e9683efcaa 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -3116,12 +3116,29 @@ get_qual_for_range(Relation parent, PartitionBoundSpec *spec,
 
 		if (or_expr_args != NIL)
 		{
-			/* OR all the non-default partition constraints; then negate it */
-			result = lappend(result,
-							 list_length(or_expr_args) > 1
-							 ? makeBoolExpr(OR_EXPR, or_expr_args, -1)
-							 : linitial(or_expr_args));
-			result = list_make1(makeBoolExpr(NOT_EXPR, result, -1));
+			Expr   *other_parts_constr;
+
+			/*
+			 * Combine the constraints obtained for non-default partitions
+			 * using OR.  As requested, each of the OR's args doesn't include
+			 * the NOT NULL test for partition keys (which is to avoid its
+			 * useless repetition).  Add the same now.
+			 */
+			other_parts_constr =
+						makeBoolExpr(AND_EXPR,
+							lappend(get_range_nulltest(key),
+									list_length(or_expr_args) > 1
+										? makeBoolExpr(OR_EXPR, or_expr_args,
+													   -1)
+										: linitial(or_expr_args)),
+									-1);
+
+			/*
+			 * Finally, the default partition contains everything *NOT*
+			 * contained in the non-default partitions.
+			 */
+			result = list_make1(makeBoolExpr(NOT_EXPR,
+										list_make1(other_parts_constr), -1));
 		}
 
 		return result;
diff --git a/src/test/regress/expected/update.out b/src/test/regress/expected/update.out
index cef70b1a1e..40217bdf9c 100644
--- a/src/test/regress/expected/update.out
+++ b/src/test/regress/expected/update.out
@@ -227,7 +227,7 @@ create table part_def partition of range_parted default;
  a      | text    |           |          |         | extended |              | 
  b      | integer |           |          |         | plain    |              | 
 Partition of: range_parted DEFAULT
-Partition constraint: (NOT (((a = 'a'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'a'::text) AND (b >= 10) AND (b < 20)) OR ((a = 'b'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'b'::text) AND (b >= 10) AND (b < 20))))
+Partition constraint: (NOT ((a IS NOT NULL) AND (b IS NOT NULL) AND (((a = 'a'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'a'::text) AND (b >= 10) AND (b < 20)) OR ((a = 'b'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'b'::text) AND (b >= 10) AND (b < 20)))))
 
 insert into range_parted values ('c', 9);
 -- ok
-- 
2.11.0

From f73f51400c6d02193d0c9a64508c90aa3a53137f Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 6/6] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 379 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  72 ++----
 3 files changed, 401 insertions(+), 54 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index e9683efcaa..032e9bffc4 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2596,7 +2596,384 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its output.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+
+				/*
+				 * minoff set to -1 means all datums are greater than
+				 * minkeys, which means all partitions satisfy minkeys.
+				 */
+				if (minoff == -1)
+					minoff = 0;
+
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+		/* 1 more index than datums in this case */
+		maxoff = boundinfo->ndatums;
+	else
+		maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * at maxoff itself is the upper bound of the rightmost
+				 * partition that needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	/*
+	 * minoff or maxoff set to -1 means none of the datums in
+	 * PartitionBoundInfo satisfies both minkeys and maxkeys.  If both are set
+	 * to a valid datum offset, that means there exists at least some
+	 * datums (and hence partitions) satisfying both minkeys and maxkeys.
+	 */
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	list_include_def = false,
+				range_include_def = false;
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+
+				/*
+				 * For range queries, always include the default list
+				 * partition.  Because list partitions divide the key space
+				 * in a discontinuous manner, not all values in the given
+				 * range will have a partition assigned.
+				 */
+				list_include_def = true;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If the bound at minoff or maxoff looks like it's an upper
+				 * bound of an unassigned range of values, move to the
+				 * adjacent bound which must be the upper bound of the
+				 * leftmost or rightmost partition, respectively, that needs
+				 * to be scanned.
+				 *
+				 * By doing that, we skip over a portion of values that do
+				 * indeed satisfy the query, but don't have a valid partition
+				 * assigned. The default partition would've been included to
+				 * cover those values.  Although, if the original bound in
+				 * question is an infinite value, there would not be any
+				 * unassigned range to speak of, because the range is unbounded
+				 * in that direction by definition, so no need to include the
+				 * default.
+				 */
+				if (boundinfo->indexes[minoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_minkeys > 0)
+						lastkey = keys->n_minkeys - 1;
+					if (boundinfo->kind[minoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					minoff += 1;
+				}
+
+				if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_maxkeys > 0)
+						lastkey = keys->n_maxkeys - 1;
+					if (boundinfo->kind[maxoff - 1][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					maxoff -= 1;
+				}
+
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						range_include_def = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if ((list_include_def || range_include_def) &&
+			partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else if (partition_bound_has_default(boundinfo))
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->default_index);
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 6c669ffdfc..35bbb5da97 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -208,16 +208,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-              QUERY PLAN               
----------------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -519,15 +517,13 @@ explain (costs off) select * from rlp where a <= 31;
          Filter: (a <= 31)
    ->  Seq Scan on rlp5_1
          Filter: (a <= 31)
-   ->  Seq Scan on rlp5_default
-         Filter: (a <= 31)
    ->  Seq Scan on rlp_default_10
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_30
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_default
          Filter: (a <= 31)
-(29 rows)
+(27 rows)
 
 explain (costs off) select * from rlp where a = 1 or a = 7;
               QUERY PLAN              
@@ -575,9 +571,7 @@ explain (costs off) select * from rlp where a > 20 and a < 27;
          Filter: ((a > 20) AND (a < 27))
    ->  Seq Scan on rlp4_2
          Filter: ((a > 20) AND (a < 27))
-   ->  Seq Scan on rlp4_default
-         Filter: ((a > 20) AND (a < 27))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 29;
            QUERY PLAN           
@@ -651,8 +645,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -661,7 +653,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -675,16 +667,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -702,9 +692,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -714,9 +702,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -730,9 +716,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -771,16 +755,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -792,9 +774,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -821,8 +801,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -830,9 +810,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -872,9 +850,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -912,9 +888,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -926,9 +900,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 -- a simpler multi-column keys case
 create table mc2p (a int, b int) partition by range (a, b);
-- 
2.11.0

From f32760f730048b35c1045f005a1b2c9898a759e6 Mon Sep 17 00:00:00 2001
From: David Rowley <dgrowley@gmail.com>
Date: Mon, 6 Nov 2017 14:39:33 +1300
Subject: [PATCH] Add bms_add_range to add members within the specified range

The same behavior could be obtained by looping and using bms_add_member,
however, using bms_add_range operates at the bitmapword level and should
be far faster when the range is large.

Author: David Rowley
---
 src/backend/nodes/bitmapset.c | 74 +++++++++++++++++++++++++++++++++++++++++++
 src/include/nodes/bitmapset.h |  1 +
 2 files changed, 75 insertions(+)

diff --git a/src/backend/nodes/bitmapset.c b/src/backend/nodes/bitmapset.c
index bf8545d..34d242b 100644
--- a/src/backend/nodes/bitmapset.c
+++ b/src/backend/nodes/bitmapset.c
@@ -785,6 +785,80 @@ bms_add_members(Bitmapset *a, const Bitmapset *b)
 }
 
 /*
+ * bms_add_range
+ *		Add members in the range of 'lower' to 'upper' to the set.
+ *
+ * Note this could also be done by calling bms_add_member in a loop, however,
+ * using this function will be faster when the range is large as we work with
+ * at the bitmapword level rather than at bit level.
+ */
+Bitmapset *
+bms_add_range(Bitmapset *a, int lower, int upper)
+{
+	int			lwordnum,
+				uwordnum,
+				wordnum;
+
+	if (lower < 0 || upper < 0)
+		elog(ERROR, "negative bitmapset member not allowed");
+	if (lower > upper)
+		elog(ERROR, "lower range must not be above upper range");
+	uwordnum = WORDNUM(upper);
+
+	if (a == NULL)
+	{
+		a = (Bitmapset *) palloc0(BITMAPSET_SIZE(uwordnum + 1));
+		a->nwords = uwordnum + 1;
+	}
+
+	/* ensure we have enough words to store the upper bit */
+	else if (uwordnum >= a->nwords)
+	{
+		int			oldnwords = a->nwords;
+		int			i;
+
+		a = (Bitmapset *) repalloc(a, BITMAPSET_SIZE(uwordnum + 1));
+		a->nwords = uwordnum + 1;
+		/* zero out the enlarged portion */
+		for (i = oldnwords; i < a->nwords; i++)
+			a->words[i] = 0;
+	}
+
+	wordnum = lwordnum = WORDNUM(lower);
+
+	/*
+	 * Starting at lower's wordnum, loop over each word up to upper's wordnum.
+	 * Along the way set all bits inclusively between lower and upper to 1. We
+	 * only need to handle the lwordnum and uwordnum specially so we don't set
+	 * any bits outside of the range.
+	 */
+	while (wordnum <= uwordnum)
+	{
+		bitmapword mask = (bitmapword) ~0;
+
+		/* If working on the lower word, zero out bits below 'lower'. */
+		if (wordnum == lwordnum)
+		{
+			int lbitnum = BITNUM(lower);
+			mask >>= lbitnum;
+			mask <<= lbitnum;
+		}
+
+		/* Likewise, if working on the upper word, zero bits above 'upper' */
+		if (wordnum == uwordnum)
+		{
+			int ushiftbits = BITS_PER_BITMAPWORD - (BITNUM(upper) + 1);
+			mask <<= ushiftbits;
+			mask >>= ushiftbits;
+		}
+
+		a->words[wordnum++] |= mask;
+	}
+
+	return a;
+}
+
+/*
  * bms_int_members - like bms_intersect, but left input is recycled
  */
 Bitmapset *
diff --git a/src/include/nodes/bitmapset.h b/src/include/nodes/bitmapset.h
index aa3fb25..3b62a97 100644
--- a/src/include/nodes/bitmapset.h
+++ b/src/include/nodes/bitmapset.h
@@ -90,6 +90,7 @@ extern bool bms_is_empty(const Bitmapset *a);
 extern Bitmapset *bms_add_member(Bitmapset *a, int x);
 extern Bitmapset *bms_del_member(Bitmapset *a, int x);
 extern Bitmapset *bms_add_members(Bitmapset *a, const Bitmapset *b);
+extern Bitmapset *bms_add_range(Bitmapset *a, int lower, int upper);
 extern Bitmapset *bms_int_members(Bitmapset *a, const Bitmapset *b);
 extern Bitmapset *bms_del_members(Bitmapset *a, const Bitmapset *b);
 extern Bitmapset *bms_join(Bitmapset *a, Bitmapset *b);
-- 
1.9.5.msysgit.1

From e13d6eda12d95dff7a11aed16841948c427f19af Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/6] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 986 ++++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |   2 +-
 src/test/regress/serial_schedule        |   1 +
 src/test/regress/sql/partition.sql      | 140 +++++
 4 files changed, 1128 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..6c669ffdfc
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,986 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp_default_null partition of rlp_default for values in (null);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_null
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a)::numeric = '1'::numeric)
+(31 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 10)
+(9 rows)
+
+explain (costs off) select * from rlp where a > 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_2
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 10)
+(23 rows)
+
+explain (costs off) select * from rlp where a < 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a < 15)
+(9 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 15)
+(17 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(17 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text < 'ab'::text) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+(5 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a is not null;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp2
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_2
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp5_1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp5_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a IS NOT NULL)
+(29 rows)
+
+explain (costs off) select * from rlp where a > 30;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 30)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 30)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+            QUERY PLAN            
+----------------------------------
+ Append
+   ->  Seq Scan on rlp_default_30
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 31)
+(29 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_null
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(25 rows)
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 20) AND (a < 27))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 29;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a = 29)
+(3 rows)
+
+explain (costs off) select * from rlp where a >= 29;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_1
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a >= 29)
+(11 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 1) AND (a >= 15))
+(23 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+explain select * from mc2p where a < 2;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p0  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p1  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p2  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a < 2)
+(9 rows)
+
+explain select * from mc2p where a = 2 and b < 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b < 1) AND (a = 2))
+(3 rows)
+
+explain select * from mc2p where a > 1;
+                             QUERY PLAN                              
+---------------------------------------------------------------------
+ Append  (cost=0.00..153.00 rows=3012 width=8)
+   ->  Seq Scan on mc2p3  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p4  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p5  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p_default  (cost=0.00..38.25 rows=753 width=8)
+         Filter: (a > 1)
+(9 rows)
+
+explain select * from mc2p where a = 1 and b > 1;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append  (cost=0.00..43.90 rows=4 width=8)
+   ->  Seq Scan on mc2p2  (cost=0.00..43.90 rows=4 width=8)
+         Filter: ((b > 1) AND (a = 1))
+(3 rows)
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..75e8a58f36
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,140 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp_default_null partition of rlp_default for values in (null);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;
+explain (costs off) select * from rlp where a is not null;
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+explain (costs off) select * from rlp where a = 29;
+explain (costs off) select * from rlp where a >= 29;
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+
+explain select * from mc2p where a < 2;
+explain select * from mc2p where a = 2 and b < 1;
+explain select * from mc2p where a > 1;
+explain select * from mc2p where a = 1 and b > 1;
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p;
-- 
2.11.0

From 88ec0ff42aa9359d89ffb738c854dedc2db0da74 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 2/6] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning (although
    as of this commit this contains *all* appinfos as mentioned
    below).

 5. Some code in try_partition_wise_join in to handle the
    possibility that a partition RelOptInfo may not have the basic
    information set (note that as noted in 0, set_append_rel_size
    now sets such information for only the *live* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, get_partitions_from_clauses
would returns without pruning any partitions.  In most cases, it's
obvious in the planner that a set of clauses identified as matching
the partition key don't contain the constant values right away, in
which case, there is no need to call get_partitions_from_clauses
right away.  Instead, it should be deferred to another piece of code
which can receive the above list of clauses and runs at a time when
the constant values become available.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  23 ++
 src/backend/optimizer/path/allpaths.c | 603 +++++++++++++++++++++++++++-------
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  | 101 ++++++
 src/include/catalog/partition.h       |   4 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 9 files changed, 682 insertions(+), 130 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 5daa8a1c19..7b0e022865 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,29 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	*min_part_idx and *max_part_idx constitutes a range of contiguous
+ *	indexes of partitions satisfying the query, while *other_parts
+ *	contains indexes of partitions that satisfy the query but are
+ *	not included in the aforementioned range
+ */
+void
+get_partitions_from_clauses(Relation relation, List *partclauses,
+							int *min_part_idx, int *max_part_idx,
+							Bitmapset **other_parts)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+
+	*min_part_idx = 0;
+	*max_part_idx = partdesc->nparts - 1;
+	*other_parts = NULL;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index a6efb4e1d3..452ecb9b03 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,11 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +137,13 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *contains_const,
+						 bool *constfalse);
 
 
 /*
@@ -845,6 +854,398 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 	rel->fdwroutine->GetForeignPaths(root, rel, rte->relid);
 }
 
+static int
+intcmp(const void *va, const void *vb)
+{
+	int	a = *((const int *) va);
+	int	b = *((const int *) vb);
+
+	if (a == b)
+		return 0;
+	return (a > b) ? 1 : -1;
+}
+
+/*
+ * get_append_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i,
+			num_parts = 0,
+			min_part_idx = -1,
+			max_part_idx = -1,
+		   *all_indexes = NULL;
+	Bitmapset *other_parts = NULL;
+	bool	contains_const,
+			constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &contains_const,
+										   &constfalse);
+
+	/*
+	 * If the matched clauses contains at least some constant operands, use
+	 * the same to prune partitions right away.
+	 */
+	if (partclauses != NIL && contains_const && !constfalse)
+	{
+		get_partitions_from_clauses(parent, partclauses,
+								   &min_part_idx, &max_part_idx,
+								   &other_parts);
+		/* Get *all* indexes in one place and sort. */
+		if (min_part_idx >= 0 && max_part_idx >= 0)
+			num_parts += (max_part_idx - min_part_idx + 1);
+		if (!bms_is_empty(other_parts))
+			num_parts += bms_num_members(other_parts);
+
+		if (num_parts > 0)
+		{
+			int		j;
+
+			all_indexes = (int *) palloc(num_parts * sizeof(int));
+			j = 0;
+			if (min_part_idx >= 0 && max_part_idx >= 0)
+			{
+				for (i = min_part_idx; i <= max_part_idx; i++)
+					all_indexes[j++] = i;
+			}
+			if (!bms_is_empty(other_parts))
+				while ((i = bms_first_member(other_parts)) >= 0)
+					all_indexes[j++] = i;
+			if (j > 1)
+				qsort((void *) all_indexes, j, sizeof(int), intcmp);
+		}
+	}
+	else if (!constfalse)
+	{
+		/* No clauses to prune paritions, so scan all partitions. */
+		num_parts = partdesc->nparts;
+		all_indexes = (int *) palloc(num_parts * sizeof(int));
+		for (i = 0; i < partdesc->nparts; i++)
+			all_indexes[i] = i;
+	}
+
+	/* Fetch the partition appinfos. */
+	for (i = 0; i < num_parts; i++)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[all_indexes[i]];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[all_indexes[i]] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+	if (all_indexes)
+		pfree(all_indexes);
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's added to the
+ * result as well.
+ *
+ * If clauses contains at least one constant operand or a Nullness test,
+ * *contains_const is set so that the caller can pass the clauses to the
+ * partitioning module right away.
+ *
+ * If the list contains a pseudo-constant RestrictInfo with constant false
+ * value, *constfalse is set.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *contains_const,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*contains_const = false;
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				(IsA(clause, Const) &&
+				 ((((Const *) clause)->constisnull) ||
+				  !DatumGetBool(((Const *) clause)->constvalue))))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way. */
+		if (IsA(clause, BoolExpr))
+		{
+			bool	contains_const1,
+					constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &contains_const1,
+										 &constfalse1) != NIL)
+			{
+				result = lappend(result, clause);
+				*contains_const = contains_const1;
+			}
+			else if (and_clause((Node *) clause))
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and that the latter is compatible with partitioning.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+
+				*contains_const = IsA(constexpr, Const);
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				/* Check if saop_op is compatible with partitioning. */
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle the same if its negator is indeed a part of the
+				 * partitioning operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) rightop))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+				{
+					result = lappend(result, nulltest);
+					/* Nullness test can be used right away. */
+					*contains_const = true;
+				}
+			}
+		}
+	}
+
+	return result;
+}
+
 /*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
@@ -860,6 +1261,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1275,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1312,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1325,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,85 +1335,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					/*
-					 * Ignore any column dropped from the parent.
-					 * Corresponding Var won't have any translation. It won't
-					 * have attr_needed information, since it can not be
-					 * referenced in the query.
-					 */
-					if (var == NULL)
-					{
-						Assert(attr_needed == NULL);
-						continue;
-					}
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
-		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
-		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
 		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1164,6 +1505,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1259,14 +1611,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1337,43 +1704,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1390,17 +1754,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index d58635c887..24d800d8b7 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6182,14 +6182,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..b06696b7f0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,93 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				/*
+				 * Ignore any column dropped from the parent.  Corresponding
+				 * Var won't have any translation. It won't have attr_needed
+				 * information, since it can not be referenced in the query.
+				 */
+				if (var == NULL)
+				{
+					Assert(attr_needed == NULL);
+					continue;
+				}
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..7da99a9f41 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,8 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+void get_partitions_from_clauses(Relation relation, List *partclauses,
+								 int *min_part_idx, int *max_part_idx,
+								 Bitmapset **other_parts);
 #endif							/* PARTITION_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From b4420e9b7e07a794cdea87cf4e5a9f1702bd86dc Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 3/6] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 1126 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 1122 insertions(+), 4 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 7b0e022865..e2c4e1fbcb 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -37,6 +37,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +113,100 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid recomputing the same in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartScanKeyInfo;
+
+ /* A data structure to represent a partition set. */
+typedef struct PartitionSet
+{
+	/*
+	 * If either  empty or all_parts is true, values of the other fields are
+	 * invalid.
+	 */
+	bool	empty;				/* contains no partitions */
+	bool	all_parts;			/* contains all partitions */
+
+	/*
+	 * In the case of range partitioning, min_part_index contains the index of
+	 * the lowest partition contained in the set and max_datum_index that of
+	 * the highest partition (all partitions between these two indexes
+	 * inclusive are part of the set.)  Since other types of partitioning do
+	 * not impose order on the data contained in successive partitions, these
+	 * fields are not set in that case.
+	 */
+	bool	use_range;
+	int		min_part_idx;
+	int		max_part_idx;
+
+	/*
+	 * other_parts contains the indexes of partitions that are not covered by
+	 * the range defined by min/max indexes.  For example, in the case of
+	 * range partitoning, it will include default partition index (if any).
+	 * Also, this is the only way to return list partitions, because list
+	 * partitions do not have the same ordering property as range partitions,
+	 * so it's pointless to use the min/max range method.
+	 */
+	Bitmapset *other_parts;
+} PartitionSet;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +246,25 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static PartitionSet *get_partitions_from_clauses_guts(Relation relation,
+								List *clauses);
+static PartitionSet *partset_copy(const PartitionSet *in);
+static PartitionSet *partset_intersect(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_union(PartitionSet *a, const PartitionSet *b);
+static PartitionSet *partset_new(bool empty, bool all_parts);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static bool partkey_datum_from_expr(const Expr *expr, Datum *value);
+static PartitionSet *get_partitions_for_keys(Relation rel,
+						PartScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1422,7 +1537,7 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 }
 
 /*
- * get_partitions_using_clauses
+ * get_partitions_from_clauses
  *		Determine the set of partitions of relation that will satisfy all
  *		the clauses contained in partclauses
  *
@@ -1438,15 +1553,1018 @@ get_partitions_from_clauses(Relation relation, List *partclauses,
 							Bitmapset **other_parts)
 {
 	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	List		   *partconstr = RelationGetPartitionQual(relation);
+	PartitionSet   *partset;
+
+	if (partconstr)
+		partconstr = (List *) expression_planner((Expr *) partconstr);
+	partclauses = list_concat(partclauses, partconstr);
+	partset = get_partitions_from_clauses_guts(relation, partclauses);
+	if (partset->empty)
+	{
+		*min_part_idx = *max_part_idx = -1;
+		*other_parts = NULL;
+	}
+	else if (partset->all_parts)
+	{
+		*min_part_idx = 0;
+		*max_part_idx = partdesc->nparts - 1;
+		*other_parts = NULL;
+	}
+	else
+	{
+		if (partset->use_range)
+		{
+			*min_part_idx = partset->min_part_idx;
+			*max_part_idx = partset->max_part_idx;
+		}
+		else
+			*min_part_idx = *max_part_idx = -1;
 
-	*min_part_idx = 0;
-	*max_part_idx = partdesc->nparts - 1;
-	*other_parts = NULL;
+		*other_parts = partset->other_parts;
+	}
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_from_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list (return value describes the set of such partitions)
+ *
+ * rt_index is the table's range table position needed to set varno of Vars
+ * contained in the table's partition constraint that is used in certain
+ * cases.
+ */
+static PartitionSet *
+get_partitions_from_clauses_guts(Relation relation, List *clauses)
+{
+	PartitionSet *partset;
+	PartScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	if (constfalse)
+		/* None of the partitions will satisfy the clauses. */
+		partset = partset_new(true, false);
+	else if (nkeys > 0)
+		/*
+		 * Only look up in the partition decriptor if the query provides
+		 * constraints on the keys at all.
+		 */
+		partset = get_partitions_for_keys(relation, &keys);
+	else
+		/* No constraints on the keys, so, return *all* partitions. */
+		partset = partset_new(false, true);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		PartitionSet *or_partset = partset_new(true, false);
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			PartitionSet  *arg_partset;
+
+			arg_partset = get_partitions_from_clauses_guts(relation,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = partset_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		partset = partset_intersect(partset, or_partset);
+	}
+
+	return partset;
+}
+
+/* Partition set manipulation functions. */
+
+static PartitionSet *
+partset_new(bool empty, bool all_parts)
+{
+	PartitionSet   *result = palloc0(sizeof(PartitionSet));
+
+	result->empty = empty;
+	result->all_parts = all_parts;
+	/*
+	 * Remains true until we explicitly turn it off in partset_union in a
+	 * certain case.
+	 */
+	result->use_range = true;
+	result->min_part_idx = result->max_part_idx = -1;
+	result->other_parts = NULL;
+
+	return result;
+}
+
+static PartitionSet *
+partset_copy(const PartitionSet *in)
+{
+	PartitionSet *result;
+
+	if (in == NULL)
+		return NULL;
+
+	result = partset_new(in->empty, in->all_parts);
+	result->min_part_idx = in->min_part_idx;
+	result->max_part_idx = in->max_part_idx;
+	result->other_parts = in->other_parts;	/* not bms_copy. */
+
+	return result;
+}
+
+/*
+ * Macros to manipulate the range of partitions specified in a given
+ * PartitionSet (s) using its min_part_idx and max_part_idx fields, which are
+ * both inclusive ends of the range.
+ */
+
+#define partset_range_empty(s)\
+		((s)->min_part_idx < 0 && (s)->max_part_idx < 0)
+
+#define partset_range_overlap(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->min_part_idx >= (s2)->min_part_idx &&\
+		   (s1)->min_part_idx <= (s2)->max_part_idx) ||\
+		  ((s2)->min_part_idx >= (s1)->min_part_idx &&\
+		   (s2)->min_part_idx <= (s1)->max_part_idx)))
+
+#define partset_range_adjacent(s1, s2)\
+		(!partset_range_empty((s1)) && !partset_range_empty((s2)) &&\
+		 (((s1)->max_part_idx == (s2)->min_part_idx) || \
+		  ((s2)->max_part_idx == (s1)->min_part_idx)))
+
+/* The result after intersection is stuffed back into 'a'. */
+static PartitionSet *
+partset_intersect(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->all_parts || b->empty)
+		a = partset_copy(b);
+	else if (b->all_parts)
+		return a;
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+
+		/*
+		 * If one or both sets' range is empty, or if they don't overlap,
+		 * then the result's range is empty.
+		 */
+		if (partset_range_empty(a) ||
+			partset_range_empty(b) ||
+			!partset_range_overlap(a, b))
+		{
+			a->min_part_idx = a->max_part_idx = -1;
+		}
+		else
+		{
+			a->min_part_idx = Max(a->min_part_idx, b->min_part_idx);
+			a->max_part_idx = Min(a->max_part_idx, b->max_part_idx);
+		}
+
+		a->other_parts = bms_intersect(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/* The result after union is stuffed back into 'a'. */
+static PartitionSet *
+partset_union(PartitionSet *a, const PartitionSet *b)
+{
+	Assert(a != NULL && b != NULL);
+
+	if (a->empty || b->all_parts)
+		a = partset_copy(b);
+	else
+	{
+		/*
+		 * Partition set is specified by min_part_idx, max_part_idx and/or
+		 * other_parts, so make the result set using those fields.
+		 */
+		int		i;
+
+		/*
+		 * Combine b's range into a's only if we're still using the range
+		 * representation.
+		 */
+		if (a->use_range)
+		{
+			if(!partset_range_empty(a) && !partset_range_empty(b))
+			{
+				/*
+				 * Unify into one range using range union only if it makes
+				 * sense, that is only if they are adjacent to or overlap with
+				 * each other.  If not, unify them by adding indexes within
+				 * both ranges to the other_parts bitmap and mark the set as
+				 * no longer using the range representation, because, the
+				 * indexes in this no longer have the property of being
+				 * contiguous.
+				 */
+				if (partset_range_overlap(a, b) ||
+					partset_range_adjacent(a, b))
+				{
+					a->min_part_idx = Min(a->min_part_idx, b->min_part_idx);
+					a->max_part_idx = Max(a->max_part_idx, b->max_part_idx);
+				}
+				else
+				{
+					for (i = a->min_part_idx; i <= a->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+					for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+						a->other_parts = bms_add_member(a->other_parts, i);
+
+					/* The set is no longer to be represented as range. */
+					a->use_range = false;
+					a->min_part_idx = a->max_part_idx = -1;
+				}
+			}
+			else if (partset_range_empty(a))
+			{
+				a->min_part_idx = b->min_part_idx;
+				a->max_part_idx = b->max_part_idx;
+			}
+		}
+		else
+		{
+			if (!partset_range_empty(b))
+			{
+				for (i = b->min_part_idx; i <= b->max_part_idx; i++)
+					a->other_parts = bms_add_member(a->other_parts, i);
+			}
+		}
+
+		a->other_parts = bms_union(a->other_parts, b->other_parts);
+	}
+
+	return a;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that information in the output
+ *		argument keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 in the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_bool_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0,
+			n_total = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	/* -1 represents an invalid value of NullTestType. */
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+				*or_clauses = lappend(*or_clauses, clause);
+			else
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+			continue;
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i],
+					partcoll = partkey->partcollation[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc;
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Skip if leftop doesn't match this partition key column. */
+				if ((!IsA(leftop, Var) ||
+					 ((Var *) leftop)->varattno != partattno) &&
+					!equal(leftop, partexpr))
+					continue;
+
+				/*
+				 * Deal with <> operators that the planner allows if it finds
+				 * out that <>'s negator is indeed a valid partopfamily member.
+				 * Make an equivalent OR expression and add to the *or_clauses
+				 * list.  That is, we convert a <> opclause into
+				 * (leftop < rightop) OR (leftop > rightop).
+				 */
+				if (!op_in_opfamily(opclause->opno, partopfamily) &&
+					(partkey->strategy == PARTITION_STRATEGY_RANGE ||
+					 partkey->strategy == PARTITION_STRATEGY_LIST))
+				{
+					Expr   *ltexpr,
+						   *gtexpr;
+					Oid		negator,
+							ltop,
+							gtop;
+					int		strategy;
+					Oid		lefttype,
+							righttype;
+
+					negator = get_negator(opclause->opno);
+					get_op_opfamily_properties(negator, partopfamily, false,
+											   &strategy,
+											   &lefttype, &righttype);
+					if (strategy != BTEqualStrategyNumber)
+						elog(LOG, "unexpected negator of '<>' operator");
+					ltop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTLessStrategyNumber);
+					gtop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTGreaterStrategyNumber);
+					ltexpr = make_opclause(ltop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					gtexpr = make_opclause(gtop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					*or_clauses = lappend(*or_clauses,
+									makeBoolExpr(OR_EXPR,
+												 list_make2(ltexpr, gtexpr),
+												 -1));
+					continue;
+				}
+
+				pc = palloc0(sizeof(PartClause));
+				pc->op = opclause;
+				pc->constarg = rightop;
+				keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+				/* A strict operator implies NOT NULL argument. */
+				if (keynullness[i] == -1)
+				{
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+				}
+				only_bool_clauses = false;
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+				List   *elem_exprs,
+					   *elem_clauses;
+				ListCell *lc1;
+				bool	negated = false;
+
+				/*
+				 * Planner must have accepted this saop iff saop_op's negator
+				 * was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/*
+				 * First generate a list of Const nodes, one for each array
+				 * element.
+				 */
+				elem_exprs = NIL;
+				if (IsA(rightop, Const))
+				{
+					Const *arr = (Const *) lsecond(saop->args);
+					ArrayType *arrval = DatumGetArrayTypeP(arr->constvalue);
+					int16	elemlen;
+					bool	elembyval;
+					char	elemalign;
+					Datum  *elem_values;
+					bool   *elem_nulls;
+					int		num_elems;
+
+					get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+										 &elemlen, &elembyval, &elemalign);
+					deconstruct_array(arrval,
+									  ARR_ELEMTYPE(arrval),
+									  elemlen, elembyval, elemalign,
+									  &elem_values, &elem_nulls,
+									  &num_elems);
+					for (i = 0; i < num_elems; i++)
+					{
+						if (!elem_nulls[i])
+							elem_exprs = lappend(elem_exprs,
+											makeConst(ARR_ELEMTYPE(arrval),
+											-1, arr->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval));
+						else
+							elem_exprs = lappend(elem_exprs,
+											makeNullConst(ARR_ELEMTYPE(arrval),
+														  -1,
+														  arr->constcollid));
+					}
+				}
+				else
+				{
+					ArrayExpr *arrexpr = castNode(ArrayExpr, rightop);
+
+					elem_exprs = list_copy(arrexpr->elements);
+				}
+
+				/*
+				 * Now generate a list of clauses, one for each array element,
+				 * of the form: saop_leftop saop_op elem_expr
+				 */
+				elem_clauses = NIL;
+				foreach(lc1, elem_exprs)
+				{
+					Const  *rightop = castNode(Const, lfirst(lc1));
+					Expr   *elem_clause;
+
+					if (rightop->constisnull)
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+															  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_clause = (Expr *) nulltest;
+					}
+					else
+					{
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_clause = (Expr *) opexpr;
+					}
+
+					elem_clauses = lappend(elem_clauses, elem_clause);
+				}
+
+				/*
+				 * Build the OR clause if needed or add the clauses to the end
+				 * of the list that's being processed currently.
+				 */
+				if (saop->useOr)
+					*or_clauses = lappend(*or_clauses,
+										  makeBoolExpr(OR_EXPR, elem_clauses,
+													   -1));
+				else
+					clauses = list_concat(clauses, elem_clauses);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					if (keynullness[i] == -1)
+					{
+						keynullness[i] = nulltest->nulltesttype;
+						n_keynullness++;
+					}
+					only_bool_clauses = false;
+				}
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_bool_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is admissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartScanKeyInfo));
+	if (n_eqkeys + n_minkeys + n_maxkeys + n_keynullness > 0)
+	{
+		Datum	value;
+		int		n_datums_resolved;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_eqkeys; i++)
+		{
+			if (partkey_datum_from_expr(eqkey_exprs[i], &value))
+			{
+				keys->eqkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_eqkeys = n_datums_resolved;
+		n_total += keys->n_eqkeys;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_minkeys; i++)
+		{
+			if (partkey_datum_from_expr(minkey_exprs[i], &value))
+			{
+				keys->minkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_minkeys = n_datums_resolved;
+		n_total += keys->n_minkeys;
+		keys->min_incl = min_incl;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_maxkeys; i++)
+		{
+			if (partkey_datum_from_expr(maxkey_exprs[i], &value))
+			{
+				keys->maxkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_maxkeys = n_datums_resolved;
+		n_total += keys->n_maxkeys;
+		keys->max_incl = max_incl;
+
+		for (i = 0; i < partkey->partnatts; i++)
+			keys->keynullness[i] = keynullness[i];
+		n_total += n_keynullness;
+	}
+
+	return n_total;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static bool
+partkey_datum_from_expr(const Expr *expr, Datum *value)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			*value = ((Const *) expr)->constvalue;
+			return true;
+
+		default:
+			return false;
+	}
+
+	Assert(false);	/* don't ever get here */
+	return false;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	Assert(IsA(leftarg->constarg, Const) &&
+		   IsA(rightarg->constarg, Const));
+	if (!partkey_datum_from_expr(leftarg->constarg, &leftarg_const))
+		return false;
+	if (!partkey_datum_from_expr(rightarg->constarg, &rightarg_const))
+		return false;
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static PartitionSet *
+get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
+{
+	return partset_new(false, true);
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
-- 
2.11.0

From 701e973f885534273d1509a5e14c23e2061dd198 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 4/6] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index e2c4e1fbcb..3ebedec335 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -207,6 +207,30 @@ typedef struct PartitionSet
 	Bitmapset *other_parts;
 } PartitionSet;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -235,14 +259,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -938,10 +963,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -992,6 +1023,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -1013,8 +1045,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -1028,9 +1063,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3564,12 +3599,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3581,6 +3619,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3611,12 +3650,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3813,12 +3853,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3840,11 +3880,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3852,17 +3892,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3873,12 +3931,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3892,20 +3951,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3918,8 +3976,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From b5d331a9483ec61784b3bf22a3ec0aa9afc4dedd Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 31 Oct 2017 16:26:55 +0900
Subject: [PATCH 5/6] Tweak default range partition's constraint a little

When using as a predicate, it's useful for it explicitly say that
the default range partition might contain nulls, because non-default
range partitions can't.
---
 src/backend/catalog/partition.c      | 29 +++++++++++++++++++++++------
 src/test/regress/expected/update.out |  2 +-
 2 files changed, 24 insertions(+), 7 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 3ebedec335..54fed3b5bc 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -3116,12 +3116,29 @@ get_qual_for_range(Relation parent, PartitionBoundSpec *spec,
 
 		if (or_expr_args != NIL)
 		{
-			/* OR all the non-default partition constraints; then negate it */
-			result = lappend(result,
-							 list_length(or_expr_args) > 1
-							 ? makeBoolExpr(OR_EXPR, or_expr_args, -1)
-							 : linitial(or_expr_args));
-			result = list_make1(makeBoolExpr(NOT_EXPR, result, -1));
+			Expr   *other_parts_constr;
+
+			/*
+			 * Combine the constraints obtained for non-default partitions
+			 * using OR.  As requested, each of the OR's args doesn't include
+			 * the NOT NULL test for partition keys (which is to avoid its
+			 * useless repetition).  Add the same now.
+			 */
+			other_parts_constr =
+						makeBoolExpr(AND_EXPR,
+							lappend(get_range_nulltest(key),
+									list_length(or_expr_args) > 1
+										? makeBoolExpr(OR_EXPR, or_expr_args,
+													   -1)
+										: linitial(or_expr_args)),
+									-1);
+
+			/*
+			 * Finally, the default partition contains everything *NOT*
+			 * contained in the non-default partitions.
+			 */
+			result = list_make1(makeBoolExpr(NOT_EXPR,
+										list_make1(other_parts_constr), -1));
 		}
 
 		return result;
diff --git a/src/test/regress/expected/update.out b/src/test/regress/expected/update.out
index cef70b1a1e..40217bdf9c 100644
--- a/src/test/regress/expected/update.out
+++ b/src/test/regress/expected/update.out
@@ -227,7 +227,7 @@ create table part_def partition of range_parted default;
  a      | text    |           |          |         | extended |              | 
  b      | integer |           |          |         | plain    |              | 
 Partition of: range_parted DEFAULT
-Partition constraint: (NOT (((a = 'a'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'a'::text) AND (b >= 10) AND (b < 20)) OR ((a = 'b'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'b'::text) AND (b >= 10) AND (b < 20))))
+Partition constraint: (NOT ((a IS NOT NULL) AND (b IS NOT NULL) AND (((a = 'a'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'a'::text) AND (b >= 10) AND (b < 20)) OR ((a = 'b'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'b'::text) AND (b >= 10) AND (b < 20)))))
 
 insert into range_parted values ('c', 9);
 -- ok
-- 
2.11.0

From 3eee5ab229e4704499fe80471e7baa4aed10dc9a Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 6/6] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 379 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out |  72 ++----
 3 files changed, 401 insertions(+), 54 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 54fed3b5bc..152ae60a6b 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2596,7 +2596,384 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static PartitionSet *
 get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
 {
-	return partset_new(false, true);
+	PartitionSet   *partset;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
+	PartitionBoundCmpArg	arg;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return partset_new(true, false);
+
+	/*
+	 * Initialize the set as one that's neither empty nor contains all
+	 * partitions.  The code below will set min_part_idx and max_part_idx
+	 * and/or other_parts as found out by comparing keys to the partition
+	 * bounds, as well as considering special partitions like null-accepting
+	 * and default partitions.  If it turns out that no partitions need to
+	 * be scanned, partset->empty will be set to true.
+	 */
+	partset = partset_new(false, false);
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				partset->other_parts = bms_make_singleton(other_idx);
+
+			return partset;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its output.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+		return partset;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		partset->all_parts = true;
+		return partset;
+	}
+
+	/* Valid keys->eqkeys must provide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
+
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			partset->other_parts =
+							bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			partset->other_parts =
+							bms_make_singleton(boundinfo->default_index);
+		else
+			partset->empty = true;
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return partset;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+
+				/*
+				 * minoff set to -1 means all datums are greater than
+				 * minkeys, which means all partitions satisfy minkeys.
+				 */
+				if (minoff == -1)
+					minoff = 0;
+
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+		/* 1 more index than datums in this case */
+		maxoff = boundinfo->ndatums;
+	else
+		maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * at maxoff itself is the upper bound of the rightmost
+				 * partition that needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	/*
+	 * minoff or maxoff set to -1 means none of the datums in
+	 * PartitionBoundInfo satisfies both minkeys and maxkeys.  If both are set
+	 * to a valid datum offset, that means there exists at least some
+	 * datums (and hence partitions) satisfying both minkeys and maxkeys.
+	 */
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	list_include_def = false,
+				range_include_def = false;
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					partset->other_parts =
+								bms_add_member(partset->other_parts,
+											   boundinfo->indexes[i]);
+
+				/*
+				 * For range queries, always include the default list
+				 * partition.  Because list partitions divide the key space
+				 * in a discontinuous manner, not all values in the given
+				 * range will have a partition assigned.
+				 */
+				list_include_def = true;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If the bound at minoff or maxoff looks like it's an upper
+				 * bound of an unassigned range of values, move to the
+				 * adjacent bound which must be the upper bound of the
+				 * leftmost or rightmost partition, respectively, that needs
+				 * to be scanned.
+				 *
+				 * By doing that, we skip over a portion of values that do
+				 * indeed satisfy the query, but don't have a valid partition
+				 * assigned. The default partition would've been included to
+				 * cover those values.  Although, if the original bound in
+				 * question is an infinite value, there would not be any
+				 * unassigned range to speak of, because the range is unbounded
+				 * in that direction by definition, so no need to include the
+				 * default.
+				 */
+				if (boundinfo->indexes[minoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_minkeys > 0)
+						lastkey = keys->n_minkeys - 1;
+					if (boundinfo->kind[minoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					minoff += 1;
+				}
+
+				if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_maxkeys > 0)
+						lastkey = keys->n_maxkeys - 1;
+					if (boundinfo->kind[maxoff - 1][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					maxoff -= 1;
+				}
+
+				partset->min_part_idx = boundinfo->indexes[minoff];
+				partset->max_part_idx = boundinfo->indexes[maxoff];
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						range_include_def = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if ((list_include_def || range_include_def) &&
+			partition_bound_has_default(boundinfo))
+			partset->other_parts = bms_add_member(partset->other_parts,
+												  boundinfo->default_index);
+	}
+	else if (partition_bound_has_default(boundinfo))
+		partset->other_parts = bms_add_member(partset->other_parts,
+											  boundinfo->default_index);
+	else
+		partset->empty = true;
+
+	return partset;
 }
 
 /*
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 6c669ffdfc..35bbb5da97 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -208,16 +208,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-              QUERY PLAN               
----------------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -519,15 +517,13 @@ explain (costs off) select * from rlp where a <= 31;
          Filter: (a <= 31)
    ->  Seq Scan on rlp5_1
          Filter: (a <= 31)
-   ->  Seq Scan on rlp5_default
-         Filter: (a <= 31)
    ->  Seq Scan on rlp_default_10
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_30
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_default
          Filter: (a <= 31)
-(29 rows)
+(27 rows)
 
 explain (costs off) select * from rlp where a = 1 or a = 7;
               QUERY PLAN              
@@ -575,9 +571,7 @@ explain (costs off) select * from rlp where a > 20 and a < 27;
          Filter: ((a > 20) AND (a < 27))
    ->  Seq Scan on rlp4_2
          Filter: ((a > 20) AND (a < 27))
-   ->  Seq Scan on rlp4_default
-         Filter: ((a > 20) AND (a < 27))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 29;
            QUERY PLAN           
@@ -651,8 +645,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -661,7 +653,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -675,16 +667,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -702,9 +692,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -714,9 +702,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -730,9 +716,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -771,16 +755,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -792,9 +774,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -821,8 +801,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -830,9 +810,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -872,9 +850,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -912,9 +888,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -926,9 +900,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 -- a simpler multi-column keys case
 create table mc2p (a int, b int) partition by range (a, b);
-- 
2.11.0

From cba48504d99e8149b6598dbb011f8e1c5a46738a Mon Sep 17 00:00:00 2001
From: David Rowley <dgrowley@gmail.com>
Date: Mon, 6 Nov 2017 14:39:33 +1300
Subject: [PATCH 2/7] Add bms_add_range to add members within the specified
 range

The same behavior could be obtained by looping and using bms_add_member,
however, using bms_add_range operates at the bitmapword level and should
be far faster when the range is large.

Author: David Rowley
---
 src/backend/nodes/bitmapset.c | 74 +++++++++++++++++++++++++++++++++++++++++++
 src/include/nodes/bitmapset.h |  1 +
 2 files changed, 75 insertions(+)

diff --git a/src/backend/nodes/bitmapset.c b/src/backend/nodes/bitmapset.c
index bf8545d437..34d242b357 100644
--- a/src/backend/nodes/bitmapset.c
+++ b/src/backend/nodes/bitmapset.c
@@ -785,6 +785,80 @@ bms_add_members(Bitmapset *a, const Bitmapset *b)
 }
 
 /*
+ * bms_add_range
+ *		Add members in the range of 'lower' to 'upper' to the set.
+ *
+ * Note this could also be done by calling bms_add_member in a loop, however,
+ * using this function will be faster when the range is large as we work with
+ * at the bitmapword level rather than at bit level.
+ */
+Bitmapset *
+bms_add_range(Bitmapset *a, int lower, int upper)
+{
+	int			lwordnum,
+				uwordnum,
+				wordnum;
+
+	if (lower < 0 || upper < 0)
+		elog(ERROR, "negative bitmapset member not allowed");
+	if (lower > upper)
+		elog(ERROR, "lower range must not be above upper range");
+	uwordnum = WORDNUM(upper);
+
+	if (a == NULL)
+	{
+		a = (Bitmapset *) palloc0(BITMAPSET_SIZE(uwordnum + 1));
+		a->nwords = uwordnum + 1;
+	}
+
+	/* ensure we have enough words to store the upper bit */
+	else if (uwordnum >= a->nwords)
+	{
+		int			oldnwords = a->nwords;
+		int			i;
+
+		a = (Bitmapset *) repalloc(a, BITMAPSET_SIZE(uwordnum + 1));
+		a->nwords = uwordnum + 1;
+		/* zero out the enlarged portion */
+		for (i = oldnwords; i < a->nwords; i++)
+			a->words[i] = 0;
+	}
+
+	wordnum = lwordnum = WORDNUM(lower);
+
+	/*
+	 * Starting at lower's wordnum, loop over each word up to upper's wordnum.
+	 * Along the way set all bits inclusively between lower and upper to 1. We
+	 * only need to handle the lwordnum and uwordnum specially so we don't set
+	 * any bits outside of the range.
+	 */
+	while (wordnum <= uwordnum)
+	{
+		bitmapword mask = (bitmapword) ~0;
+
+		/* If working on the lower word, zero out bits below 'lower'. */
+		if (wordnum == lwordnum)
+		{
+			int lbitnum = BITNUM(lower);
+			mask >>= lbitnum;
+			mask <<= lbitnum;
+		}
+
+		/* Likewise, if working on the upper word, zero bits above 'upper' */
+		if (wordnum == uwordnum)
+		{
+			int ushiftbits = BITS_PER_BITMAPWORD - (BITNUM(upper) + 1);
+			mask <<= ushiftbits;
+			mask >>= ushiftbits;
+		}
+
+		a->words[wordnum++] |= mask;
+	}
+
+	return a;
+}
+
+/*
  * bms_int_members - like bms_intersect, but left input is recycled
  */
 Bitmapset *
diff --git a/src/include/nodes/bitmapset.h b/src/include/nodes/bitmapset.h
index aa3fb253c2..3b62a97775 100644
--- a/src/include/nodes/bitmapset.h
+++ b/src/include/nodes/bitmapset.h
@@ -90,6 +90,7 @@ extern bool bms_is_empty(const Bitmapset *a);
 extern Bitmapset *bms_add_member(Bitmapset *a, int x);
 extern Bitmapset *bms_del_member(Bitmapset *a, int x);
 extern Bitmapset *bms_add_members(Bitmapset *a, const Bitmapset *b);
+extern Bitmapset *bms_add_range(Bitmapset *a, int lower, int upper);
 extern Bitmapset *bms_int_members(Bitmapset *a, const Bitmapset *b);
 extern Bitmapset *bms_del_members(Bitmapset *a, const Bitmapset *b);
 extern Bitmapset *bms_join(Bitmapset *a, Bitmapset *b);
-- 
2.11.0

From 902c005fb3c064da67550fea6d29a8bb21a8fb28 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 13 Sep 2017 18:24:55 +0900
Subject: [PATCH 3/7] Planner-side changes for partition-pruning

This adds all the necessary planner code and representations viz.

 0. Code to teach set_append_rel_size/pathlist to look at only
    the *live* partitions of partitioned tables.

 1. Add a field partcollation to PartitionScheme, which will be
    needed to verify that a operator clause's input collation
    indeed matches what is used for partitioning, to be able
    to use the clause for partition-pruning (using parttypcoll
    won't be correct, because that's not what's used by
    partitioning)

 2. Code to match the clauses to the table's partition key and
    generate a list of such matching clauses.

 3. Add a field to RelOptInfo to store an array of pointers of
    AppendRelInfo of *all* partitions (stored in the same order as
    their RelOptInfos in part_rels)

 4. Add a field to RelOptInfo to store a list of AppendRelInfos
    of *live* partitions that survived partition-pruning (although
    as of this commit this contains *all* appinfos as mentioned
    below).

 5. Some code in try_partition_wise_join in to handle the
    possibility that a partition RelOptInfo may not have the basic
    information set (note that as noted in 0, set_append_rel_size
    now sets such information for only the *live* partitions)

If the clauses identified in 2 above does not contain values
necessary to perform partition pruning, get_partitions_from_clauses
would returns without pruning any partitions.  In most cases, it's
obvious in the planner that a set of clauses identified as matching
the partition key don't contain the constant values right away, in
which case, there is no need to call get_partitions_from_clauses
right away.  Instead, it should be deferred to another piece of code
which can receive the above list of clauses and runs at a time when
the constant values become available.

In addition, a stub function get_partitions_from_clauses is added in
partition.c, which currently simply returns all partitions from the
partition descriptor.

Authors: Amit Langote, Dilip Kumar
---
 src/backend/catalog/partition.c       |  18 ++
 src/backend/optimizer/path/allpaths.c | 587 +++++++++++++++++++++++++++-------
 src/backend/optimizer/path/indxpath.c |   3 -
 src/backend/optimizer/path/joinrels.c |  24 ++
 src/backend/optimizer/plan/planner.c  |  20 +-
 src/backend/optimizer/util/plancat.c  |   4 +
 src/backend/optimizer/util/relnode.c  | 101 ++++++
 src/include/catalog/partition.h       |   2 +
 src/include/catalog/pg_opfamily.h     |   3 +
 src/include/nodes/relation.h          |  29 +-
 src/include/optimizer/pathnode.h      |   4 +
 11 files changed, 662 insertions(+), 133 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 5daa8a1c19..5e601dd0a4 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -1421,6 +1421,24 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 	}
 }
 
+/*
+ * get_partitions_using_clauses
+ *		Determine the set of partitions of relation that will satisfy all
+ *		the clauses contained in partclauses
+ *
+ * Outputs:
+ *	A Bitmapset containing indexes of all selected partitions.
+ */
+Bitmapset *
+get_partitions_from_clauses(Relation relation, List *partclauses)
+{
+	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
+	Bitmapset	   *result = NULL;
+
+	result = bms_add_range(result, 0, partdesc->nparts - 1);
+	return result;
+}
+
 /* Module-local functions */
 
 /*
diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c
index a6efb4e1d3..77b13ad397 100644
--- a/src/backend/optimizer/path/allpaths.c
+++ b/src/backend/optimizer/path/allpaths.c
@@ -20,9 +20,12 @@
 
 #include "access/sysattr.h"
 #include "access/tsmapi.h"
+#include "catalog/partition.h"
 #include "catalog/pg_class.h"
 #include "catalog/pg_operator.h"
+#include "catalog/pg_opfamily.h"
 #include "catalog/pg_proc.h"
+#include "catalog/pg_type.h"
 #include "foreign/fdwapi.h"
 #include "miscadmin.h"
 #include "nodes/makefuncs.h"
@@ -135,6 +138,13 @@ static void recurse_push_qual(Node *setOp, Query *topquery,
 static void remove_unused_subquery_outputs(Query *subquery, RelOptInfo *rel);
 static void add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 						List *live_childrels);
+static List *get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte);
+static List *match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *contains_const,
+						 bool *constfalse);
 
 
 /*
@@ -846,6 +856,381 @@ set_foreign_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
 }
 
 /*
+ * get_append_rel_partitions
+ *		Return the list of partitions of rel that pass the clauses mentioned
+ *		rel->baserestrictinfo
+ *
+ * Returned list contains the AppendRelInfos of chosen partitions.
+ */
+static List *
+get_append_rel_partitions(PlannerInfo *root,
+						  RelOptInfo *rel,
+						  RangeTblEntry *rte)
+{
+	Relation	parent = heap_open(rte->relid, NoLock);
+	PartitionDesc partdesc = RelationGetPartitionDesc(parent);
+	List   *partclauses;
+	List   *result = NIL;
+	int		i;
+	Bitmapset *partindexes = NULL;
+	bool	contains_const,
+			constfalse;
+
+	/*
+	 * Get the clauses that match the partition key, including information
+	 * about any nullness tests against partition keys.  Set keynullness to
+	 * a invalid value of NullTestType, which 0 is not.
+	 */
+	partclauses = match_clauses_to_partkey(rel,
+										   list_copy(rel->baserestrictinfo),
+										   &contains_const,
+										   &constfalse);
+
+	/*
+	 * If the matched clauses contains at least some constant operands, use
+	 * the same to prune partitions right away.
+	 */
+	if (partclauses != NIL && contains_const && !constfalse)
+		partindexes = get_partitions_from_clauses(parent, partclauses);
+	else if (!constfalse)
+		/* No clauses to prune paritions, so scan all partitions. */
+		partindexes = bms_add_range(partindexes, 0, partdesc->nparts - 1);
+
+	/* Fetch the partition appinfos. */
+	while ((i = bms_first_member(partindexes)) >= 0)
+	{
+		AppendRelInfo *appinfo = rel->part_appinfos[i];
+#ifdef USE_ASSERT_CHECKING
+		RangeTblEntry *rte = planner_rt_fetch(appinfo->child_relid, root);
+
+		/*
+		 * Must be the intended child's RTE here, because appinfos are ordered
+		 * the same way as partitions in the partition descriptor.
+		 */
+		Assert(partdesc->oids[i] == rte->relid);
+#endif
+		result = lappend(result, appinfo);
+	}
+
+	/* Remember for future users such as set_append_rel_pathlist(). */
+	rel->live_part_appinfos = result;
+
+	heap_close(parent, NoLock);
+
+	return result;
+}
+
+#define PartCollMatchesExprColl(partcoll, exprcoll) \
+	((partcoll) == InvalidOid || (partcoll) == (exprcoll))
+
+/*
+ * match_clauses_to_partkey
+ *		Match clauses with rel's partition key
+ *
+ * For an individual clause to match with a partition key column, the clause:
+ *
+ *	1.	must be in the form (partkey op const) or (const op partkey);
+ *	2.	must contain an operator which is in the same operator family as the
+ *		partitioning operator for the partition key column
+ *	3.	its input collation must match the partitioning collation
+ *
+ * The "const" mentioned in 1 means any expression that doesn't involve a
+ * volatile function or a Var of this relation.  We allow Vars belonging to
+ * other relations (for example, if the clause is a join clause), but they
+ * are treated as parameters whose values are not known now, so cannot be
+ * used for partition pruning right within the planner.  It's the
+ * responsibility of higher code levels to manage restriction and join
+ * clauses appropriately.
+ *
+ * If a NullTest against a partition key is encountered, it's added to the
+ * result as well.
+ *
+ * If clauses contains at least one constant operand or a Nullness test,
+ * *contains_const is set so that the caller can pass the clauses to the
+ * partitioning module right away.
+ *
+ * If the list contains a pseudo-constant RestrictInfo with constant false
+ * value, *constfalse is set.
+ */
+static List *
+match_clauses_to_partkey(RelOptInfo *rel,
+						 List *clauses,
+						 bool *contains_const,
+						 bool *constfalse)
+{
+	PartitionScheme	partscheme = rel->part_scheme;
+	List	   *result = NIL;
+	ListCell   *lc;
+
+	*contains_const = false;
+	*constfalse = false;
+
+	Assert (partscheme != NULL);
+
+	foreach(lc, clauses)
+	{
+		Expr   *clause;
+		int		i;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				(IsA(clause, Const) &&
+				 ((((Const *) clause)->constisnull) ||
+				  !DatumGetBool(((Const *) clause)->constvalue))))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way. */
+		if (IsA(clause, BoolExpr))
+		{
+			bool	contains_const1,
+					constfalse1;
+
+			/*
+			 * If the OR's args contain clauses that match, add the clause
+			 * to the result.
+			 */
+			if (or_clause((Node *) clause) &&
+				match_clauses_to_partkey(rel,
+									 list_copy(((BoolExpr *) clause)->args),
+										 &contains_const1,
+										 &constfalse1) != NIL)
+			{
+				result = lappend(result, clause);
+				*contains_const = contains_const1;
+				continue;
+			}
+			else if (and_clause((Node *) clause))
+			{
+				/*
+				 * These clauses are ANDed with the clauses in the
+				 * original list, so queue them after the latter.  Note
+				 * that it also means that a queued clause will be added to
+				 * the result if it happens to match.
+				 */
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+				continue;
+			}
+			/* Fall-through for a NOT clause, which is handled below. */
+		}
+
+		for (i = 0; i < partscheme->partnatts; i++)
+		{
+			Node   *partkey = linitial(rel->partexprs[i]);
+			Oid		partopfamily = partscheme->partopfamily[i],
+					partcoll = partscheme->partcollation[i];
+
+			/*
+			 * Check if the operator is compatible with partitioning and if
+			 * so, add it to the list of opclauses matched with this partition
+			 * key.
+			 */
+			if (is_opclause(clause))
+			{
+				Expr   *constexpr,
+					   *leftop,
+					   *rightop;
+				Relids	constrelids,
+						left_relids,
+						right_relids;
+				Oid		expr_op,
+						expr_coll;
+
+				leftop = (Expr *) get_leftop(clause);
+				rightop = (Expr *) get_rightop(clause);
+				expr_op = ((OpExpr *) clause)->opno;
+				expr_coll = ((OpExpr *) clause)->inputcollid;
+				left_relids = pull_varnos((Node *) leftop);
+				right_relids = pull_varnos((Node *) rightop);
+
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				if (IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				if (equal(leftop, partkey))
+				{
+					constexpr = rightop;
+					constrelids = right_relids;
+				}
+				else if (equal(rightop, partkey))
+				{
+					constexpr = leftop;
+					constrelids = left_relids;
+					expr_op = get_commutator(expr_op);
+					/*
+					 * If no commutator exists, cannot flip the qual's args,
+					 * so give up.
+					 */
+					if (!OidIsValid(expr_op))
+						continue;
+				}
+				else
+					/* Neither argument matches the partition key. */
+					continue;
+
+				/*
+				 * Only allow strict operators to think sanely about the
+				 * behavior with null arguments.
+				 */
+				if (!op_strict(expr_op))
+					continue;
+
+				/*
+				 * Check if the operator is in the partition operator family.
+				 * It the operator happens to be '<>', which is never listed
+				 * as part of the operator family, check if its negator
+				 * exists and that the latter is compatible with partitioning.
+				 */
+				if (!op_in_opfamily(expr_op, partopfamily))
+				{
+					Oid		negator = get_negator(expr_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/*
+				 * Useless if what we're thinking of as a constant is actually
+				 * a Var coming from this relation.
+				 */
+				if (bms_is_member(rel->relid, constrelids))
+					continue;
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) constexpr))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, expr_coll))
+					continue;
+
+				/*
+				 * Everything seems to be fine, so add it to the list of
+				 * clauses we will use for pruning.  Flip the left and right
+				 * args if we have to, because the code that extracts the
+				 * constant value to use for partition-pruning expects to find
+				 * it as the rightop of the clause.
+				 */
+				if (constexpr == rightop)
+					result = lappend(result, clause);
+				else
+				{
+					OpExpr   *commuted;
+
+					commuted = (OpExpr *) copyObject(clause);
+					commuted->opno = expr_op;
+					commuted->opfuncid = get_opcode(expr_op);
+					commuted->args = list_make2(rightop, leftop);
+					result = lappend(result, commuted);
+				}
+
+				*contains_const = IsA(constexpr, Const);
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+
+				if (IsA(leftop, RelabelType))
+					leftop = (Node *) ((RelabelType *) leftop)->arg;
+				if (!equal(leftop, partkey))
+					continue;
+
+				/* Check if saop_op is compatible with partitioning. */
+				if (!op_strict(saop_op))
+					continue;
+
+				/*
+				 * In case of NOT IN (..), we get a '<>', which while not
+				 * listed as part of any operator family, we are able to
+				 * handle the same if its negator is indeed a part of the
+				 * partitioning operator family.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+				{
+					Oid		negator = get_negator(saop_op);
+
+					if (!OidIsValid(negator) ||
+						!op_in_opfamily(negator, partopfamily))
+						continue;
+				}
+
+				/* Useless if the "constant" can change its value. */
+				if (contain_volatile_functions((Node *) rightop))
+					continue;
+
+				/*
+				 * Also, useless, if the clause's collation is different from
+				 * the partitioning collation.
+				 */
+				if (!PartCollMatchesExprColl(partcoll, saop_coll))
+					continue;
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest   *nulltest = (NullTest *) clause;
+				Node	   *arg = (Node *) nulltest->arg;
+
+				if (equal(arg, partkey))
+				{
+					result = lappend(result, nulltest);
+					/* Nullness test can be used right away. */
+					*contains_const = true;
+				}
+			}
+			/*
+			 * Boolean conditions have a special shape, which accept if the
+			 * partitioning opfamily accepts Boolean conditions.
+			 */
+			else if (IsBooleanOpfamily(partopfamily) &&
+					 (IsA(clause, BooleanTest) ||
+					  IsA(clause, Var) || not_clause((Node *) clause)))
+			{
+				if (IsA(clause, BooleanTest))
+				{
+					BooleanTest *btest = (BooleanTest *) clause;
+
+					if (btest->booltesttype != IS_UNKNOWN &&
+						btest->booltesttype != IS_NOT_UNKNOWN)
+						if (equal((Node *) btest->arg, partkey))
+							result = lappend(result, clause);
+				}
+				else if (IsA(clause, Var) && equal((Node *) clause, partkey))
+					result = lappend(result, clause);
+				else
+				{
+					Node   *arg = (Node *) get_notclausearg((Expr *) clause);
+
+					if (equal(arg, partkey))
+						result = lappend(result, clause);
+				}
+				*contains_const = true;
+			}
+		}
+	}
+
+	return result;
+}
+
+/*
  * set_append_rel_size
  *	  Set size estimates for a simple "append relation"
  *
@@ -860,6 +1245,7 @@ static void
 set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 					Index rti, RangeTblEntry *rte)
 {
+	List	   *rel_appinfos = NIL;
 	int			parentRTindex = rti;
 	bool		has_live_children;
 	double		parent_rows;
@@ -873,6 +1259,23 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 
 	Assert(IS_SIMPLE_REL(rel));
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+	{
+		rel_appinfos = get_append_rel_partitions(root, rel, rte);
+		rel->live_partitioned_rels = list_make1_int(rti);
+	}
+
 	/*
 	 * Initialize to compute size estimates for whole append relation.
 	 *
@@ -893,7 +1296,7 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 	nattrs = rel->max_attr - rel->min_attr + 1;
 	parent_attrsizes = (double *) palloc0(nattrs * sizeof(double));
 
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -906,10 +1309,6 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *childvars;
 		ListCell   *lc;
 
-		/* append_rel_list contains all append rels; ignore others */
-		if (appinfo->parent_relid != parentRTindex)
-			continue;
-
 		childRTindex = appinfo->child_relid;
 		childRTE = root->simple_rte_array[childRTindex];
 
@@ -920,85 +1319,11 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		childrel = find_base_rel(root, childRTindex);
 		Assert(childrel->reloptkind == RELOPT_OTHER_MEMBER_REL);
 
-		if (rel->part_scheme)
-		{
-			AttrNumber	attno;
-
-			/*
-			 * We need attr_needed data for building targetlist of a join
-			 * relation representing join between matching partitions for
-			 * partition-wise join. A given attribute of a child will be
-			 * needed in the same highest joinrel where the corresponding
-			 * attribute of parent is needed. Hence it suffices to use the
-			 * same Relids set for parent and child.
-			 */
-			for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
-			{
-				int			index = attno - rel->min_attr;
-				Relids		attr_needed = rel->attr_needed[index];
-
-				/* System attributes do not need translation. */
-				if (attno <= 0)
-				{
-					Assert(rel->min_attr == childrel->min_attr);
-					childrel->attr_needed[index] = attr_needed;
-				}
-				else
-				{
-					Var		   *var = list_nth_node(Var,
-													appinfo->translated_vars,
-													attno - 1);
-					int			child_index;
-
-					/*
-					 * Ignore any column dropped from the parent.
-					 * Corresponding Var won't have any translation. It won't
-					 * have attr_needed information, since it can not be
-					 * referenced in the query.
-					 */
-					if (var == NULL)
-					{
-						Assert(attr_needed == NULL);
-						continue;
-					}
-
-					child_index = var->varattno - childrel->min_attr;
-					childrel->attr_needed[child_index] = attr_needed;
-				}
-			}
-		}
-
 		/*
-		 * Copy/Modify targetlist. Even if this child is deemed empty, we need
-		 * its targetlist in case it falls on nullable side in a child-join
-		 * because of partition-wise join.
-		 *
-		 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
-		 * expressions, which otherwise would not occur in a rel's targetlist.
-		 * Code that might be looking at an appendrel child must cope with
-		 * such.  (Normally, a rel's targetlist would only include Vars and
-		 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
-		 * fields of childrel->reltarget; not clear if that would be useful.
+		 * Initialize some properties of child rel from the parent rel, such
+		 * target list, equivalence class members, etc.
 		 */
-		childrel->reltarget->exprs = (List *)
-			adjust_appendrel_attrs(root,
-								   (Node *) rel->reltarget->exprs,
-								   1, &appinfo);
-
-		/*
-		 * We have to make child entries in the EquivalenceClass data
-		 * structures as well.  This is needed either if the parent
-		 * participates in some eclass joins (because we will want to consider
-		 * inner-indexscan joins on the individual children) or if the parent
-		 * has useful pathkeys (because we should try to build MergeAppend
-		 * paths that produce those sort orderings). Even if this child is
-		 * deemed dummy, it may fall on nullable side in a child-join, which
-		 * in turn may participate in a MergeAppend, where we will need the
-		 * EquivalenceClass data structures.
-		 */
-		if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
-			add_child_rel_equivalences(root, appinfo, rel, childrel);
-		childrel->has_eclass_joins = rel->has_eclass_joins;
+		set_basic_child_rel_properties(root, rel, childrel, appinfo);
 
 		/*
 		 * We have to copy the parent's quals to the child, with appropriate
@@ -1164,6 +1489,17 @@ set_append_rel_size(PlannerInfo *root, RelOptInfo *rel,
 		has_live_children = true;
 
 		/*
+		 * If childrel is itself partitioned, add it and its partitioned
+		 * children to the list being propagated up to the root rel.
+		 */
+		if (childrel->part_scheme && rel->part_scheme)
+		{
+			rel->live_partitioned_rels =
+					list_concat(rel->live_partitioned_rels,
+								list_copy(childrel->live_partitioned_rels));
+		}
+
+		/*
 		 * If any live child is not parallel-safe, treat the whole appendrel
 		 * as not parallel-safe.  In future we might be able to generate plans
 		 * in which some children are farmed out to workers while others are
@@ -1259,14 +1595,29 @@ set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
 						Index rti, RangeTblEntry *rte)
 {
 	int			parentRTindex = rti;
-	List	   *live_childrels = NIL;
+	List	   *rel_appinfos = NIL,
+			   *live_childrels = NIL;
 	ListCell   *l;
 
+	if (rte->relkind != RELKIND_PARTITIONED_TABLE)
+	{
+		foreach (l, root->append_rel_list)
+		{
+			AppendRelInfo   *appinfo = lfirst(l);
+
+			/* append_rel_list contains all append rels; ignore others */
+			if (appinfo->parent_relid == parentRTindex)
+				rel_appinfos = lappend(rel_appinfos, appinfo);
+		}
+	}
+	else
+		rel_appinfos = rel->live_part_appinfos;
+
 	/*
 	 * Generate access paths for each member relation, and remember the
 	 * non-dummy children.
 	 */
-	foreach(l, root->append_rel_list)
+	foreach(l, rel_appinfos)
 	{
 		AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
 		int			childRTindex;
@@ -1337,43 +1688,40 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 	ListCell   *l;
 	List	   *partitioned_rels = NIL;
 	RangeTblEntry *rte;
-	bool		build_partitioned_rels = false;
 
+	/*
+	 * AppendPath we are about to generate must record the RT indexes of
+	 * partitioned tables that are direct or indirect children of this Append
+	 * rel.  For partitioned tables, we collect its live partitioned children
+	 * from rel->painfo.  However, it will contain only its immediate children,
+	 * so collect live partitioned children from all children that are
+	 * themselves partitioned and concatenate to our list before finally
+	 * passing the list to create_append_path() and/or
+	 * generate_mergeappend_paths().
+	 *
+	 * If this is a sub-query RTE, its RelOptInfo doesn't itself contain the
+	 * list of live partitioned children, so we must assemble the same in the
+	 * loop below from the children that are known to correspond to
+	 * partitioned rels.  (This assumes that we don't need to look through
+	 * multiple levels of subquery RTEs; if we ever do, we could consider
+	 * stuffing the list we generate here into sub-query RTE's RelOptInfo, just
+	 * like we do for partitioned rels, which would be used when populating our
+	 * parent rel with paths.  For the present, that appears to be
+	 * unnecessary.)
+	 */
 	if (IS_SIMPLE_REL(rel))
 	{
-		/*
-		 * A root partition will already have a PartitionedChildRelInfo, and a
-		 * non-root partitioned table doesn't need one, because its Append
-		 * paths will get flattened into the parent anyway.  For a subquery
-		 * RTE, no PartitionedChildRelInfo exists; we collect all
-		 * partitioned_rels associated with any child.  (This assumes that we
-		 * don't need to look through multiple levels of subquery RTEs; if we
-		 * ever do, we could create a PartitionedChildRelInfo with the
-		 * accumulated list of partitioned_rels which would then be found when
-		 * populated our parent rel with paths.  For the present, that appears
-		 * to be unnecessary.)
-		 */
 		rte = planner_rt_fetch(rel->relid, root);
-		switch (rte->rtekind)
-		{
-			case RTE_RELATION:
-				if (rte->relkind == RELKIND_PARTITIONED_TABLE)
-					partitioned_rels =
-						get_partitioned_child_rels(root, rel->relid);
-				break;
-			case RTE_SUBQUERY:
-				build_partitioned_rels = true;
-				break;
-			default:
-				elog(ERROR, "unexpected rtekind: %d", (int) rte->rtekind);
-		}
+		if (rte->rtekind == RTE_RELATION &&
+			rte->relkind == RELKIND_PARTITIONED_TABLE)
+		partitioned_rels = rel->live_partitioned_rels;
 	}
 	else if (rel->reloptkind == RELOPT_JOINREL && rel->part_scheme)
 	{
 		/*
-		 * Associate PartitionedChildRelInfo of the root partitioned tables
-		 * being joined with the root partitioned join (indicated by
-		 * RELOPT_JOINREL).
+		 * For joinrel consisting of root partitioned tables, get
+		 * partitioned_rels list by combining live_partitioned_rels of the
+		 * component partitioned tables.
 		 */
 		partitioned_rels = get_partitioned_child_rels_for_join(root,
 															   rel->relids);
@@ -1390,17 +1738,12 @@ add_paths_to_append_rel(PlannerInfo *root, RelOptInfo *rel,
 		ListCell   *lcp;
 
 		/*
-		 * If we need to build partitioned_rels, accumulate the partitioned
-		 * rels for this child.
+		 * Accumulate the live partitioned children of this child, if it's
+		 * itself partitioned rel.
 		 */
-		if (build_partitioned_rels)
-		{
-			List	   *cprels;
-
-			cprels = get_partitioned_child_rels(root, childrel->relid);
+		if (childrel->part_scheme)
 			partitioned_rels = list_concat(partitioned_rels,
-										   list_copy(cprels));
-		}
+										   childrel->live_partitioned_rels);
 
 		/*
 		 * If child has an unparameterized cheapest-total path, add that to
diff --git a/src/backend/optimizer/path/indxpath.c b/src/backend/optimizer/path/indxpath.c
index f35380391a..f4203ce200 100644
--- a/src/backend/optimizer/path/indxpath.c
+++ b/src/backend/optimizer/path/indxpath.c
@@ -40,9 +40,6 @@
 #include "utils/selfuncs.h"
 
 
-#define IsBooleanOpfamily(opfamily) \
-	((opfamily) == BOOL_BTREE_FAM_OID || (opfamily) == BOOL_HASH_FAM_OID)
-
 #define IndexCollMatchesExprColl(idxcollation, exprcollation) \
 	((idxcollation) == InvalidOid || (idxcollation) == (exprcollation))
 
diff --git a/src/backend/optimizer/path/joinrels.c b/src/backend/optimizer/path/joinrels.c
index 2b868c52de..3e943391b1 100644
--- a/src/backend/optimizer/path/joinrels.c
+++ b/src/backend/optimizer/path/joinrels.c
@@ -1396,6 +1396,30 @@ try_partition_wise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 											   child_rel2->relids);
 
 		/*
+		 * If either child_rel1 or child_rel2 is not a live partition, they'd
+		 * not have been touched by set_append_rel_size.  So, its RelOptInfo
+		 * would be missing some information that set_append_rel_size sets for
+		 * live partitions, such as the target list, child EQ members, etc.
+		 * We need to make the RelOptInfo of even the dead partitions look
+		 * minimally valid and as having a valid dummy path attached to it.
+		 */
+		if (IS_SIMPLE_REL(child_rel1) && child_rel1->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel1->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel1, child_rel1, appinfo);
+			mark_dummy_rel(child_rel1);
+		}
+
+		if (IS_SIMPLE_REL(child_rel2) && child_rel2->pathlist == NIL)
+		{
+			AppendRelInfo *appinfo = rel2->part_appinfos[cnt_parts];
+
+			set_basic_child_rel_properties(root, rel2, child_rel2, appinfo);
+			mark_dummy_rel(child_rel2);
+		}
+
+		/*
 		 * Construct restrictions applicable to the child join from those
 		 * applicable to the parent join.
 		 */
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index d58635c887..24d800d8b7 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -6182,14 +6182,24 @@ List *
 get_partitioned_child_rels_for_join(PlannerInfo *root, Relids join_relids)
 {
 	List	   *result = NIL;
-	ListCell   *l;
+	int			relid;
 
-	foreach(l, root->pcinfo_list)
+	relid = -1;
+	while ((relid = bms_next_member(join_relids, relid)) >= 0)
 	{
-		PartitionedChildRelInfo *pc = lfirst(l);
+		RelOptInfo *rel;
 
-		if (bms_is_member(pc->parent_relid, join_relids))
-			result = list_concat(result, list_copy(pc->child_rels));
+		/* Paranoia: ignore bogus relid indexes */
+		if (relid >= root->simple_rel_array_size)
+			continue;
+		rel = root->simple_rel_array[relid];
+		if (rel == NULL)
+			continue;
+		Assert(rel->relid == relid);	/* sanity check on array */
+		Assert(rel->part_scheme != NULL);
+		Assert(rel->live_partitioned_rels != NIL &&
+			   list_length(rel->live_partitioned_rels) > 0);
+		result = list_concat(result, list_copy(rel->live_partitioned_rels));
 	}
 
 	return result;
diff --git a/src/backend/optimizer/util/plancat.c b/src/backend/optimizer/util/plancat.c
index 9d35a41e22..e1ef936e68 100644
--- a/src/backend/optimizer/util/plancat.c
+++ b/src/backend/optimizer/util/plancat.c
@@ -1918,6 +1918,10 @@ find_partition_scheme(PlannerInfo *root, Relation relation)
 	memcpy(part_scheme->parttypbyval, partkey->parttypbyval,
 		   sizeof(bool) * partnatts);
 
+	part_scheme->partcollation = (Oid *) palloc(sizeof(Oid) * partnatts);
+	memcpy(part_scheme->partcollation, partkey->partcollation,
+		   sizeof(Oid) * partnatts);
+
 	/* Add the partitioning scheme to PlannerInfo. */
 	root->part_schemes = lappend(root->part_schemes, part_scheme);
 
diff --git a/src/backend/optimizer/util/relnode.c b/src/backend/optimizer/util/relnode.c
index 3bd1063aa8..b06696b7f0 100644
--- a/src/backend/optimizer/util/relnode.c
+++ b/src/backend/optimizer/util/relnode.c
@@ -18,6 +18,7 @@
 
 #include "miscadmin.h"
 #include "catalog/partition.h"
+#include "catalog/pg_class.h"
 #include "optimizer/clauses.h"
 #include "optimizer/cost.h"
 #include "optimizer/pathnode.h"
@@ -154,9 +155,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 	rel->part_scheme = NULL;
 	rel->nparts = 0;
 	rel->boundinfo = NULL;
+	rel->part_appinfos = NULL;
 	rel->part_rels = NULL;
 	rel->partexprs = NULL;
 	rel->nullable_partexprs = NULL;
+	rel->live_part_appinfos = NIL;
+	rel->live_partitioned_rels = NIL;
 
 	/*
 	 * Pass top parent's relids down the inheritance hierarchy. If the parent
@@ -233,8 +237,12 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 		int			cnt_parts = 0;
 
 		if (nparts > 0)
+		{
+			rel->part_appinfos = (AppendRelInfo **)
+				palloc(sizeof(AppendRelInfo *) * nparts);
 			rel->part_rels = (RelOptInfo **)
 				palloc(sizeof(RelOptInfo *) * nparts);
+		}
 
 		foreach(l, root->append_rel_list)
 		{
@@ -258,6 +266,7 @@ build_simple_rel(PlannerInfo *root, int relid, RelOptInfo *parent)
 			 * also match the PartitionDesc.  See expand_partitioned_rtentry.
 			 */
 			Assert(cnt_parts < nparts);
+			rel->part_appinfos[cnt_parts] = appinfo;
 			rel->part_rels[cnt_parts] = childrel;
 			cnt_parts++;
 		}
@@ -567,6 +576,7 @@ build_join_rel(PlannerInfo *root,
 	joinrel->part_scheme = NULL;
 	joinrel->nparts = 0;
 	joinrel->boundinfo = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -735,6 +745,7 @@ build_child_join_rel(PlannerInfo *root, RelOptInfo *outer_rel,
 	joinrel->has_eclass_joins = false;
 	joinrel->top_parent_relids = NULL;
 	joinrel->part_scheme = NULL;
+	joinrel->part_appinfos = NULL;
 	joinrel->part_rels = NULL;
 	joinrel->partexprs = NULL;
 	joinrel->nullable_partexprs = NULL;
@@ -1747,3 +1758,93 @@ build_joinrel_partition_info(RelOptInfo *joinrel, RelOptInfo *outer_rel,
 		joinrel->nullable_partexprs[cnt] = nullable_partexpr;
 	}
 }
+
+/*
+ * Initialize some basic properties of child rel from the parent rel, such
+ * target list, equivalence class members, etc.
+ */
+void
+set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo)
+{
+	AttrNumber	attno;
+
+	if (rel->part_scheme)
+	{
+		/*
+		 * We need attr_needed data for building targetlist of a join relation
+		 * representing join between matching partitions for partition-wise
+		 * join. A given attribute of a child will be needed in the same
+		 * highest joinrel where the corresponding attribute of parent is
+		 * needed. Hence it suffices to use the same Relids set for parent and
+		 * child.
+		 */
+		for (attno = rel->min_attr; attno <= rel->max_attr; attno++)
+		{
+			int			index = attno - rel->min_attr;
+			Relids		attr_needed = rel->attr_needed[index];
+
+			/* System attributes do not need translation. */
+			if (attno <= 0)
+			{
+				Assert(rel->min_attr == childrel->min_attr);
+				childrel->attr_needed[index] = attr_needed;
+			}
+			else
+			{
+				Var		   *var = list_nth_node(Var,
+												appinfo->translated_vars,
+												attno - 1);
+				int			child_index;
+
+				/*
+				 * Ignore any column dropped from the parent.  Corresponding
+				 * Var won't have any translation. It won't have attr_needed
+				 * information, since it can not be referenced in the query.
+				 */
+				if (var == NULL)
+				{
+					Assert(attr_needed == NULL);
+					continue;
+				}
+
+				child_index = var->varattno - childrel->min_attr;
+				childrel->attr_needed[child_index] = attr_needed;
+			}
+		}
+	}
+
+	/*
+	 * Copy/Modify targetlist. Even if this child is deemed empty, we need
+	 * its targetlist in case it falls on nullable side in a child-join
+	 * because of partition-wise join.
+	 *
+	 * NB: the resulting childrel->reltarget->exprs may contain arbitrary
+	 * expressions, which otherwise would not occur in a rel's targetlist.
+	 * Code that might be looking at an appendrel child must cope with
+	 * such.  (Normally, a rel's targetlist would only include Vars and
+	 * PlaceHolderVars.)  XXX we do not bother to update the cost or width
+	 * fields of childrel->reltarget; not clear if that would be useful.
+	 */
+	childrel->reltarget->exprs = (List *)
+			adjust_appendrel_attrs(root,
+								   (Node *) rel->reltarget->exprs,
+								   1, &appinfo);
+
+	/*
+	 * We have to make child entries in the EquivalenceClass data
+	 * structures as well.  This is needed either if the parent
+	 * participates in some eclass joins (because we will want to consider
+	 * inner-indexscan joins on the individual children) or if the parent
+	 * has useful pathkeys (because we should try to build MergeAppend
+	 * paths that produce those sort orderings). Even if this child is
+	 * deemed dummy, it may fall on nullable side in a child-join, which
+	 * in turn may participate in a MergeAppend, where we will need the
+	 * EquivalenceClass data structures.
+	 */
+	if (rel->has_eclass_joins || has_useful_pathkeys(root, rel))
+		add_child_rel_equivalences(root, appinfo, rel, childrel);
+	childrel->has_eclass_joins = rel->has_eclass_joins;
+}
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 945ac0239d..4a1ce92569 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -108,4 +108,6 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 						   PartitionBoundSpec *new_spec);
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
+/* For partition-pruning */
+Bitmapset get_partitions_from_clauses(Relation relation, List *partclauses);
 #endif							/* PARTITION_H */
diff --git a/src/include/catalog/pg_opfamily.h b/src/include/catalog/pg_opfamily.h
index 0d0ba7c66a..f2fddeceb8 100644
--- a/src/include/catalog/pg_opfamily.h
+++ b/src/include/catalog/pg_opfamily.h
@@ -187,4 +187,7 @@ DATA(insert OID = 4082 (	3580	pg_lsn_minmax_ops		PGNSP PGUID ));
 DATA(insert OID = 4104 (	3580	box_inclusion_ops		PGNSP PGUID ));
 DATA(insert OID = 5000 (	4000	box_ops		PGNSP PGUID ));
 
+#define IsBooleanOpfamily(opfamily) \
+	((opfamily) == BOOL_BTREE_FAM_OID || (opfamily) == BOOL_HASH_FAM_OID)
+
 #endif							/* PG_OPFAMILY_H */
diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index e085cefb7b..ecf70a66c4 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -342,6 +342,10 @@ typedef struct PlannerInfo
  * partition bounds. Since partition key data types and the opclass declared
  * input data types are expected to be binary compatible (per ResolveOpClass),
  * both of those should have same byval and length properties.
+ *
+ * Since partitioning might be using a collation for a given partition key
+ * column that is not same as the collation implied by column's type, store
+ * the same separately.
  */
 typedef struct PartitionSchemeData
 {
@@ -349,7 +353,8 @@ typedef struct PartitionSchemeData
 	int16		partnatts;		/* number of partition attributes */
 	Oid		   *partopfamily;	/* OIDs of operator families */
 	Oid		   *partopcintype;	/* OIDs of opclass declared input data types */
-	Oid		   *parttypcoll;	/* OIDs of collations of partition keys. */
+	Oid		   *parttypcoll;	/* OIDs of partition key type collation. */
+	Oid		   *partcollation;	/* OIDs of partitioning collation */
 
 	/* Cached information about partition key data types. */
 	int16	   *parttyplen;
@@ -529,6 +534,7 @@ typedef struct PartitionSchemeData *PartitionScheme;
  * 		part_scheme - Partitioning scheme of the relation
  * 		boundinfo - Partition bounds
  * 		nparts - Number of partitions
+ *		part_appinfos - AppendRelInfo of each partition
  * 		part_rels - RelOptInfos for each partition
  * 		partexprs, nullable_partexprs - Partition key expressions
  *
@@ -657,10 +663,27 @@ typedef struct RelOptInfo
 	PartitionScheme part_scheme;	/* Partitioning scheme. */
 	int			nparts;			/* number of partitions */
 	struct PartitionBoundInfoData *boundinfo;	/* Partition bounds */
-	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of partitions,
-									 * stored in the same order of bounds */
+	struct AppendRelInfo  **part_appinfos;	/* Array of AppendRelInfos of
+											 * of partitioned, stored in the
+											 * same order as of bounds */
+	struct RelOptInfo **part_rels;	/* Array of RelOptInfos of *all*
+									 * partitions, stored in the same order as
+									 * of bounds */
 	List	  **partexprs;		/* Non-nullable partition key expressions. */
 	List	  **nullable_partexprs;	/* Nullable partition key expressions. */
+
+
+	/*
+	 * List of AppendRelInfo's of the table's partitions that survive a
+	 * query's clauses.
+	 */
+	List	   *live_part_appinfos;
+
+	/*
+	 * RT indexes of live partitions that are partitioned tables themselves.
+	 * This includes the RT index of the table itself.
+	 */
+	List	   *live_partitioned_rels;
 } RelOptInfo;
 
 /*
diff --git a/src/include/optimizer/pathnode.h b/src/include/optimizer/pathnode.h
index e9ed16ad32..c1f2fc93cd 100644
--- a/src/include/optimizer/pathnode.h
+++ b/src/include/optimizer/pathnode.h
@@ -296,5 +296,9 @@ extern RelOptInfo *build_child_join_rel(PlannerInfo *root,
 					 RelOptInfo *outer_rel, RelOptInfo *inner_rel,
 					 RelOptInfo *parent_joinrel, List *restrictlist,
 					 SpecialJoinInfo *sjinfo, JoinType jointype);
+extern void set_basic_child_rel_properties(PlannerInfo *root,
+							   RelOptInfo *rel,
+							   RelOptInfo *childrel,
+							   AppendRelInfo *appinfo);
 
 #endif							/* PATHNODE_H */
-- 
2.11.0

From 7bd8adf227aa280104af146f6f42650b8893779d Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 13:48:13 +0900
Subject: [PATCH 4/7] Implement get_partitions_from_clauses

This now actually processes partclauses and classifies them into
a set of keys that can be used to look up partitions in the
partition descriptor, although there is still no support for the
latter.
---
 src/backend/catalog/partition.c | 949 +++++++++++++++++++++++++++++++++++++++-
 src/include/catalog/partition.h |   3 +-
 2 files changed, 947 insertions(+), 5 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 5e601dd0a4..17b6a8a258 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -27,6 +27,8 @@
 #include "catalog/pg_inherits.h"
 #include "catalog/pg_inherits_fn.h"
 #include "catalog/pg_opclass.h"
+#include "catalog/pg_operator.h"
+#include "catalog/pg_opfamily.h"
 #include "catalog/pg_partitioned_table.h"
 #include "catalog/pg_type.h"
 #include "commands/tablecmds.h"
@@ -37,6 +39,8 @@
 #include "nodes/parsenodes.h"
 #include "optimizer/clauses.h"
 #include "optimizer/planmain.h"
+#include "optimizer/planner.h"
+#include "optimizer/predtest.h"
 #include "optimizer/prep.h"
 #include "optimizer/var.h"
 #include "rewrite/rewriteManip.h"
@@ -111,6 +115,67 @@ typedef struct PartitionRangeBound
 	bool		lower;			/* this is the lower (vs upper) bound */
 } PartitionRangeBound;
 
+/*
+ * Information about a clause matched with a partition key column kept to
+ * avoid recomputing the same in remove_redundant_clauses().
+ */
+typedef struct
+{
+	OpExpr *op;
+	Expr   *constarg;
+
+	/* cached info. */
+	bool	valid_cache;	/* Is the following information initialized? */
+	int		op_strategy;
+	Oid		op_subtype;
+	FmgrInfo op_func;
+} PartClause;
+
+/*
+ * PartScanKeyInfo
+ *		Bounding scan keys to look up a table's partitions obtained from
+ *		mutually-ANDed clauses containing partitioning-compatible operators
+ */
+typedef struct PartScanKeyInfo
+{
+	/*
+	 * Constants constituting the *whole* partition key compared using
+	 * partitioning-compatible equality operator(s).  When n_eqkeys > 0, other
+	 * keys (minkeys and maxkeys) are irrelevant.
+	 */
+	Datum	eqkeys[PARTITION_MAX_KEYS];
+	int		n_eqkeys;
+
+	/*
+	 * Constants that constitute the lower bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using > or >=
+	 * operator compatible with partitioning, making this the lower bound in
+	 * a range query.
+	 */
+	Datum	minkeys[PARTITION_MAX_KEYS];
+	int		n_minkeys;
+	bool	min_incl;
+
+	/*
+	 * Constants that constitute the upper bound on the partition key or a
+	 * prefix thereof.  The last of those constants is compared using < or <=
+	 * operator compatible with partitioning, making this the upper bound in
+	 * a range query.
+	 */
+	Datum	maxkeys[PARTITION_MAX_KEYS];
+	int		n_maxkeys;
+	bool	max_incl;
+
+	/*
+	 * Specifies the type of NullTest that was applied to each of the
+	 * partition key columns or -1 if none was applied.  Partitioning handles
+	 * null partition keys specially depending on the partitioning method in
+	 * use, so get_partitions_for_keys can return partitions according to
+	 * the nullness condition for partition keys.
+	 */
+	NullTestType	keynullness[PARTITION_MAX_KEYS];
+} PartScanKeyInfo;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -150,6 +215,21 @@ static int partition_bound_bsearch(PartitionKey key,
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
+static Bitmapset *get_partitions_from_clauses_guts(Relation relation,
+								List *clauses);
+static int classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses);
+static void remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse);
+static bool partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result);
+static bool partkey_datum_from_expr(const Expr *expr, Datum *value);
+static Bitmapset *get_partitions_for_keys(Relation rel,
+						PartScanKeyInfo *keys);
+
 /*
  * RelationBuildPartitionDesc
  *		Form rel's partition descriptor
@@ -1422,7 +1502,7 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 }
 
 /*
- * get_partitions_using_clauses
+ * get_partitions_from_clauses
  *		Determine the set of partitions of relation that will satisfy all
  *		the clauses contained in partclauses
  *
@@ -1432,16 +1512,877 @@ get_partition_dispatch_recurse(Relation rel, Relation parent,
 Bitmapset *
 get_partitions_from_clauses(Relation relation, List *partclauses)
 {
-	PartitionDesc	partdesc = RelationGetPartitionDesc(relation);
-	Bitmapset	   *result = NULL;
+	Bitmapset	   *result;
+	List		   *partconstr = RelationGetPartitionQual(relation);
+
+	if (partconstr)
+		partconstr = (List *) expression_planner((Expr *) partconstr);
+	partclauses = list_concat(partclauses, partconstr);
+
+	result = get_partitions_from_clauses_guts(relation, partclauses);
 
-	result = bms_add_range(result, 0, partdesc->nparts - 1);
 	return result;
 }
 
 /* Module-local functions */
 
 /*
+ * get_partitions_from_clauses_guts
+ *		Determine relation's partitions that satisfy *all* of the clauses
+ *		in the list
+ *
+ * Return value is a Bitmapset containing the indexes of selected partitions.
+ */
+static Bitmapset *
+get_partitions_from_clauses_guts(Relation relation, List *clauses)
+{
+	PartitionDesc partdesc = RelationGetPartitionDesc(relation);
+	Bitmapset *result = NULL;
+	PartScanKeyInfo keys;
+	int		nkeys;
+	bool	constfalse;
+	List *or_clauses;
+	ListCell *lc;
+
+	nkeys = classify_partition_bounding_keys(relation, clauses,
+											 &keys, &constfalse,
+											 &or_clauses);
+	/*
+	 * Only look up in the partition decriptor if the query provides
+	 * constraints on the keys at all.
+	 */
+	if (nkeys > 0 && !constfalse)
+		result = get_partitions_for_keys(relation, &keys);
+	else if (!constfalse)
+		/* No constraints on the keys, so, return *all* partitions. */
+		result = bms_add_range(result, 0, partdesc->nparts - 1);
+
+	foreach(lc, or_clauses)
+	{
+		BoolExpr *or = (BoolExpr *) lfirst(lc);
+		ListCell *lc1;
+		Bitmapset *or_partset = NULL;
+
+		foreach(lc1, or->args)
+		{
+			Expr *orarg = lfirst(lc1);
+			Bitmapset *arg_partset;
+
+			arg_partset = get_partitions_from_clauses_guts(relation,
+														   list_make1(orarg));
+
+			/* Combine partition sets obtained from mutually ORed clauses. */
+			or_partset = bms_union(or_partset, arg_partset);
+		}
+
+		/* Combine partition sets obtained from mutually ANDed clauses. */
+		result = bms_intersect(result, or_partset);
+	}
+
+	return result;
+}
+
+/*
+ * classify_partition_bounding_keys
+ *		Classify partition clauses into equal, min, max keys, along with any
+ *		Nullness constraints and return that information in the output
+ *		argument keys (number of keys is the return value)
+ *
+ * Clauses in the provided list are implicitly ANDed, each of which is known
+ * to match some partition key column.  Map them to individual key columns
+ * and for each column, determine the equal bound or "best" min and max bound.
+ * For example, of a > 1, a > 2, and a >= 5, "5" is the best min bound for
+ * for the column a, which also happens to be an inclusive bound.
+ *
+ * For multi-column keys, an equal bound is returned only if all the columns
+ * are constrained by equality clauses.  Min and maximum bounds could contain
+ * bound values for only a prefix of key columns.
+ *
+ * If the list contains a pseudo-constant clause, *constfalse is set to true
+ * and no keys are set.  It is also set if we encounter mutually contradictory
+ * clauses in this function ourselves, for example, having both a > 1 and
+ * a = 0 in the list.
+ *
+ * All the OR clauses encountered in the list are added to *or_clauses.  It's
+ * the responsibility of the caller to process the argument clauses of each of
+ * the OR clauses, which would involve recursively calling this function.
+ */
+static int
+classify_partition_bounding_keys(Relation relation, List *clauses,
+								 PartScanKeyInfo *keys, bool *constfalse,
+								 List **or_clauses)
+{
+	PartitionKey partkey = RelationGetPartitionKey(relation);
+	int		i;
+	ListCell *lc;
+	List   *keyclauses_all[PARTITION_MAX_KEYS],
+		   *keyclauses[PARTITION_MAX_KEYS];
+	bool	only_or_clauses = true;
+	Expr   *eqkey_exprs[PARTITION_MAX_KEYS],
+		   *minkey_exprs[PARTITION_MAX_KEYS],
+		   *maxkey_exprs[PARTITION_MAX_KEYS];
+	NullTestType keynullness[PARTITION_MAX_KEYS];
+	bool	need_next_eq,
+			need_next_min,
+			need_next_max,
+			eqkey_set[PARTITION_MAX_KEYS],
+			minkey_set[PARTITION_MAX_KEYS],
+			maxkey_set[PARTITION_MAX_KEYS],
+			min_incl,
+			max_incl;
+	int		n_eqkeys = 0,
+			n_minkeys = 0,
+			n_maxkeys = 0,
+			n_keynullness = 0,
+			n_total = 0;
+
+	*or_clauses = NIL;
+	*constfalse = false;
+	memset(keyclauses_all, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	/* -1 represents an invalid value of NullTestType. */
+	memset(keynullness, -1, PARTITION_MAX_KEYS * sizeof(NullTestType));
+
+	foreach(lc, clauses)
+	{
+		Expr	   *clause;
+		ListCell   *partexprs_item;
+
+		if (IsA(lfirst(lc), RestrictInfo))
+		{
+			RestrictInfo *rinfo = lfirst(lc);
+
+			clause = rinfo->clause;
+			if (rinfo->pseudoconstant &&
+				!DatumGetBool(((Const *) clause)->constvalue))
+			{
+				*constfalse = true;
+				continue;
+			}
+		}
+		else
+			clause = (Expr *) lfirst(lc);
+
+		/* Get the BoolExpr's out of the way.*/
+		if (IsA(clause, BoolExpr))
+		{
+			if (or_clause((Node *) clause))
+			{
+				*or_clauses = lappend(*or_clauses, clause);
+				continue;
+			}
+			else if (and_clause((Node *) clause))
+			{
+				clauses = list_concat(clauses,
+									  list_copy(((BoolExpr *) clause)->args));
+				continue;
+			}
+			/* Fall-through for a NOT clause, which is handled below. */
+		}
+
+		partexprs_item = list_head(partkey->partexprs);
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			Oid		partopfamily = partkey->partopfamily[i],
+					partcoll = partkey->partcollation[i];
+			AttrNumber	partattno = partkey->partattrs[i];
+			Expr *partexpr = NULL;
+			PartClause *pc;
+
+			if (partattno == 0)
+			{
+				partexpr = lfirst(partexprs_item);
+				partexprs_item = lnext(partexprs_item);
+			}
+
+			if (IsA(clause, OpExpr))
+			{
+				OpExpr	   *opclause;
+				Expr	   *leftop,
+						   *rightop;
+
+				opclause = (OpExpr *) clause;
+				leftop = linitial(opclause->args);
+				if (IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+				rightop = lsecond(opclause->args);
+				/* Skip if leftop doesn't match this partition key column. */
+				if ((!IsA(leftop, Var) ||
+					 ((Var *) leftop)->varattno != partattno) &&
+					!equal(leftop, partexpr))
+					continue;
+
+				/*
+				 * Deal with <> operators that the planner allows if it finds
+				 * out that <>'s negator is indeed a valid partopfamily member.
+				 * Make an equivalent OR expression and add to the *or_clauses
+				 * list.  That is, we convert a <> opclause into
+				 * (leftop < rightop) OR (leftop > rightop).
+				 */
+				if (!op_in_opfamily(opclause->opno, partopfamily) &&
+					(partkey->strategy == PARTITION_STRATEGY_RANGE ||
+					 partkey->strategy == PARTITION_STRATEGY_LIST))
+				{
+					Expr   *ltexpr,
+						   *gtexpr;
+					Oid		negator,
+							ltop,
+							gtop;
+					int		strategy;
+					Oid		lefttype,
+							righttype;
+
+					negator = get_negator(opclause->opno);
+					get_op_opfamily_properties(negator, partopfamily, false,
+											   &strategy,
+											   &lefttype, &righttype);
+					if (strategy != BTEqualStrategyNumber)
+						elog(LOG, "unexpected negator of '<>' operator");
+					ltop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTLessStrategyNumber);
+					gtop = get_opfamily_member(partopfamily,
+											   lefttype, righttype,
+											   BTGreaterStrategyNumber);
+					ltexpr = make_opclause(ltop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					gtexpr = make_opclause(gtop, BOOLOID, false,
+										   (Expr *) leftop, (Expr *) rightop,
+										   InvalidOid, partcoll);
+					*or_clauses = lappend(*or_clauses,
+									makeBoolExpr(OR_EXPR,
+												 list_make2(ltexpr, gtexpr),
+												 -1));
+					continue;
+				}
+
+				pc = palloc0(sizeof(PartClause));
+				pc->op = opclause;
+				pc->constarg = rightop;
+				keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+
+				/* A strict operator implies NOT NULL argument. */
+				if (keynullness[i] == -1)
+				{
+					keynullness[i] = IS_NOT_NULL;
+					n_keynullness++;
+				}
+				only_or_clauses = false;
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
+				Oid		saop_op = saop->opno;
+				Oid		saop_opfuncid = saop->opfuncid;
+				Oid		saop_coll = saop->inputcollid;
+				Node   *leftop = (Node *) linitial(saop->args),
+					   *rightop = (Node *) lsecond(saop->args);
+				List   *elem_exprs,
+					   *elem_clauses;
+				ListCell *lc1;
+				bool	negated = false;
+
+				/*
+				 * Planner must have accepted this saop iff saop_op's negator
+				 * was found to be a valid partopfamily member.
+				 */
+				if (!op_in_opfamily(saop_op, partopfamily))
+					negated = true;
+
+				/*
+				 * First generate a list of Const nodes, one for each array
+				 * element.
+				 */
+				elem_exprs = NIL;
+				if (IsA(rightop, Const))
+				{
+					Const *arr = (Const *) lsecond(saop->args);
+					ArrayType *arrval = DatumGetArrayTypeP(arr->constvalue);
+					int16	elemlen;
+					bool	elembyval;
+					char	elemalign;
+					Datum  *elem_values;
+					bool   *elem_nulls;
+					int		num_elems;
+
+					get_typlenbyvalalign(ARR_ELEMTYPE(arrval),
+										 &elemlen, &elembyval, &elemalign);
+					deconstruct_array(arrval,
+									  ARR_ELEMTYPE(arrval),
+									  elemlen, elembyval, elemalign,
+									  &elem_values, &elem_nulls,
+									  &num_elems);
+					for (i = 0; i < num_elems; i++)
+					{
+						if (!elem_nulls[i])
+							elem_exprs = lappend(elem_exprs,
+											makeConst(ARR_ELEMTYPE(arrval),
+											-1, arr->constcollid,
+											elemlen, elem_values[i],
+											false, elembyval));
+						else
+							elem_exprs = lappend(elem_exprs,
+											makeNullConst(ARR_ELEMTYPE(arrval),
+														  -1,
+														  arr->constcollid));
+					}
+				}
+				else
+				{
+					ArrayExpr *arrexpr = castNode(ArrayExpr, rightop);
+
+					elem_exprs = list_copy(arrexpr->elements);
+				}
+
+				/*
+				 * Now generate a list of clauses, one for each array element,
+				 * of the form: saop_leftop saop_op elem_expr
+				 */
+				elem_clauses = NIL;
+				foreach(lc1, elem_exprs)
+				{
+					Const  *rightop = castNode(Const, lfirst(lc1));
+					Expr   *elem_clause;
+
+					if (rightop->constisnull)
+					{
+						NullTest *nulltest = makeNode(NullTest);
+
+						nulltest->arg = (Expr *) leftop;
+						nulltest->nulltesttype = !negated ? IS_NULL
+															  : IS_NOT_NULL;
+						nulltest->argisrow = false;
+						nulltest->location = -1;
+						elem_clause = (Expr *) nulltest;
+					}
+					else
+					{
+						OpExpr *opexpr = makeNode(OpExpr);
+
+						opexpr->opno = saop_op;
+						opexpr->opfuncid = saop_opfuncid;
+						opexpr->opresulttype = BOOLOID;
+						opexpr->opretset = false;
+						opexpr->opcollid = InvalidOid;
+						opexpr->inputcollid = saop_coll;
+						opexpr->args = list_make2(leftop, rightop);
+						opexpr->location = -1;
+						elem_clause = (Expr *) opexpr;
+					}
+
+					elem_clauses = lappend(elem_clauses, elem_clause);
+				}
+
+				/*
+				 * Build the OR clause if needed or add the clauses to the end
+				 * of the list that's being processed currently.
+				 */
+				if (saop->useOr)
+					*or_clauses = lappend(*or_clauses,
+										  makeBoolExpr(OR_EXPR, elem_clauses,
+													   -1));
+				else
+					clauses = list_concat(clauses, elem_clauses);
+			}
+			else if (IsA(clause, NullTest))
+			{
+				NullTest *nulltest = (NullTest *) clause;
+				Expr *arg = nulltest->arg;
+
+				/* Does leftop match with this partition key column? */
+				if ((IsA(arg, Var) && partattno != 0 &&
+					((Var *) arg)->varattno == partattno) ||
+					!equal(arg, partexpr))
+				{
+					if (keynullness[i] == -1)
+					{
+						keynullness[i] = nulltest->nulltesttype;
+						n_keynullness++;
+					}
+					only_or_clauses = false;
+				}
+			}
+			/*
+			 * Boolean conditions have a special shape, which would've been
+			 * accepted if the partitioning opfamily accepts Boolean
+			 * conditions.
+			 */
+			else if (IsBooleanOpfamily(partopfamily) &&
+					 (IsA(clause, BooleanTest) ||
+					  IsA(clause, Var) ||
+					  not_clause((Node *) clause)))
+			{
+				Expr   *leftop,
+					   *rightop;
+
+				pc = palloc0(sizeof(PartClause));
+
+				if (IsA(clause, BooleanTest))
+				{
+					BooleanTest *btest = (BooleanTest *) clause;
+
+					leftop = btest->arg;
+					rightop = (btest->booltesttype == IS_TRUE ||
+							   btest->booltesttype == IS_NOT_FALSE)
+									? (Expr *) makeBoolConst(true, false)
+									: (Expr *) makeBoolConst(false, false);
+				}
+				else
+				{
+					leftop = IsA(clause, Var)
+								? (Expr *) clause
+								: (Expr *) get_notclausearg((Expr *) clause);
+					rightop = IsA(clause, Var)
+									? (Expr *) makeBoolConst(true, false)
+									: (Expr *) makeBoolConst(false, false);
+				}
+				pc->op = (OpExpr *) make_opclause(BooleanEqualOperator,
+										   BOOLOID, false,
+										   leftop, rightop,
+										   InvalidOid, InvalidOid);
+				pc->constarg = rightop;
+				keyclauses_all[i] = lappend(keyclauses_all[i], pc);
+				only_or_clauses = false;
+			}
+		}
+	}
+
+	/* Return if no work to do below. */
+	if (only_or_clauses || *constfalse)
+		return 0;
+
+	/*
+	 * Redundant key elimination using btree-semantics based tricks.
+	 *
+	 * Only list and range partitioning use btree operator semantics, so
+	 * skip otherwise.   Also, if there are expressions whose value is yet
+	 * unknown, skip this step, because we need to compare actual values
+	 * below.
+	 */
+	memset(keyclauses, 0, PARTITION_MAX_KEYS * sizeof(List *));
+	if (partkey->strategy == PARTITION_STRATEGY_LIST ||
+		partkey->strategy == PARTITION_STRATEGY_RANGE)
+	{
+		for (i = 0; i < partkey->partnatts; i++)
+		{
+			remove_redundant_clauses(partkey, i,
+									 keyclauses_all[i],
+									 &keyclauses[i],
+									 constfalse);
+			if (*constfalse)
+				return 0;
+		}
+	}
+
+	/*
+	 * Now, generate the bounding tuples that can serve as equal, min, and
+	 * max keys.  An equal bounding key must contain all partition key
+	 * columns, whereas a prefix of all partition key columns is admissible
+	 * as min and max keys.
+	 */
+	memset(eqkey_exprs, 0, sizeof(eqkey_exprs));
+	memset(minkey_exprs, 0, sizeof(minkey_exprs));
+	memset(maxkey_exprs, 0, sizeof(maxkey_exprs));
+	memset(eqkey_set, false, sizeof(eqkey_set));
+	memset(minkey_set, false, sizeof(minkey_set));
+	memset(maxkey_set, false, sizeof(maxkey_set));
+
+	need_next_eq = true;
+	need_next_min = true;
+	need_next_max = true;
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		/*
+		 * If no scan key existed for the previous column, we are done.
+		 */
+		if (i > n_eqkeys)
+			need_next_eq = false;
+
+		if (i > n_minkeys)
+			need_next_min = false;
+
+		if (i > n_maxkeys)
+			need_next_max = false;
+
+		foreach(lc, keyclauses[i])
+		{
+			PartClause *clause = lfirst(lc);
+			int		strategy = clause->op_strategy;
+
+			switch (strategy)
+			{
+				case BTLessStrategyNumber:
+				case BTLessEqualStrategyNumber:
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = (strategy == BTLessEqualStrategyNumber);
+
+						if (strategy == BTLessStrategyNumber)
+							need_next_eq = need_next_max = false;
+					}
+					break;
+
+				case BTGreaterStrategyNumber:
+				case BTGreaterEqualStrategyNumber:
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = (strategy == BTGreaterEqualStrategyNumber);
+
+						if (strategy == BTGreaterStrategyNumber)
+							need_next_eq = need_next_min = false;
+					}
+					break;
+
+				case BTEqualStrategyNumber:
+					if (need_next_eq)
+					{
+						eqkey_exprs[i] = clause->constarg;
+						if (!eqkey_set[i])
+							n_eqkeys++;
+						eqkey_set[i] = true;
+					}
+
+					if (need_next_min)
+					{
+						minkey_exprs[i] = clause->constarg;
+						if (!minkey_set[i])
+							n_minkeys++;
+						minkey_set[i] = true;
+						min_incl = true;
+					}
+
+					if (need_next_max)
+					{
+						maxkey_exprs[i] = clause->constarg;
+						if (!maxkey_set[i])
+							n_maxkeys++;
+						maxkey_set[i] = true;
+						max_incl = true;
+					}
+					break;
+
+				/*
+				 * Ideally, never get here, because 1. we don't support
+				 * operators that are not btree operators and 2. clauses
+				 * containing '<>' which are not listed in the btree operator
+				 * families have already been handled by the higher-level
+				 * code.
+				 */
+				default:
+					break;
+			}
+		}
+	}
+
+	/*
+	 * If we have equal keys for all the partition key columns, then mark
+	 * their copies in minkeys and maxkeys as invalid, so that we perform
+	 * partition lookup using only eqkeys.  Don't pass as the equal key
+	 * otherwise.
+	 */
+	if (n_eqkeys == partkey->partnatts)
+		n_minkeys = n_maxkeys = 0;
+	else
+		n_eqkeys = 0;
+
+	/* Populate keys. */
+	memset(keys, 0, sizeof(PartScanKeyInfo));
+	if (n_eqkeys + n_minkeys + n_maxkeys + n_keynullness > 0)
+	{
+		Datum	value;
+		int		n_datums_resolved;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_eqkeys; i++)
+		{
+			if (partkey_datum_from_expr(eqkey_exprs[i], &value))
+			{
+				keys->eqkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_eqkeys = n_datums_resolved;
+		n_total += keys->n_eqkeys;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_minkeys; i++)
+		{
+			if (partkey_datum_from_expr(minkey_exprs[i], &value))
+			{
+				keys->minkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_minkeys = n_datums_resolved;
+		n_total += keys->n_minkeys;
+		keys->min_incl = min_incl;
+
+		n_datums_resolved = 0;
+		for (i = 0; i < n_maxkeys; i++)
+		{
+			if (partkey_datum_from_expr(maxkey_exprs[i], &value))
+			{
+				keys->maxkeys[i] = value;
+				n_datums_resolved++;
+			}
+		}
+		keys->n_maxkeys = n_datums_resolved;
+		n_total += keys->n_maxkeys;
+		keys->max_incl = max_incl;
+
+		for (i = 0; i < partkey->partnatts; i++)
+			keys->keynullness[i] = keynullness[i];
+		n_total += n_keynullness;
+	}
+
+	return n_total;
+}
+
+/*
+ * partkey_datum_from_expr
+ *		Extract constant value from expr and set *datum to that value
+ */
+static bool
+partkey_datum_from_expr(const Expr *expr, Datum *value)
+{
+	/*
+	 * Add more expression types here as needed to support higher-level
+	 * code.
+	 */
+	switch (nodeTag(expr))
+	{
+		case T_Const:
+			*value = ((Const *) expr)->constvalue;
+			return true;
+
+		default:
+			return false;
+	}
+
+	Assert(false);	/* don't ever get here */
+	return false;
+}
+
+static void
+remove_redundant_clauses(PartitionKey partkey,
+						 int partattoff, List *all_clauses,
+						 List **result, bool *constfalse)
+{
+	Oid			partopfamily = partkey->partopfamily[partattoff];
+	Oid			partopcintype = partkey->partopcintype[partattoff];
+	PartClause *xform[BTMaxStrategyNumber];
+	ListCell *lc;
+	int		s;
+	bool	test_result;
+
+	*result = NIL;
+
+	/*
+	 * xform[s] points to the currently best scan key of strategy type s+1; it
+	 * is NULL if we haven't yet found such a key for this attr.
+	 */
+	memset(xform, 0, sizeof(xform));
+	foreach(lc, all_clauses)
+	{
+		PartClause *cur = lfirst(lc);
+
+		if (!cur->valid_cache)
+		{
+			Oid		lefttype;
+			get_op_opfamily_properties(cur->op->opno, partopfamily, false,
+									   &cur->op_strategy,
+									   &lefttype,
+									   &cur->op_subtype);
+			fmgr_info(get_opcode(cur->op->opno), &cur->op_func);
+			cur->valid_cache = true;
+		}
+
+		s = cur->op_strategy - 1;
+		/* Have we seen a clause of this strategy before?. */
+		if (xform[s] == NULL)
+		{
+			/* nope, so assign. */
+			xform[s] = cur;
+		}
+		else
+		{
+			/* yup, keep only the more restrictive key. */
+			if (partition_cmp_args(partopfamily, partopcintype,
+								   cur, cur, xform[s],
+								   &test_result))
+			{
+				if (test_result)
+					xform[s] = cur;
+				else if (s == BTEqualStrategyNumber - 1)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* else the old key is more restrictive, keep around. */
+			}
+			else
+			{
+				/*
+				 * we couldn't determine which one is more restrictive.  Keep
+				 * the previous one in xform[s] and push this one directly
+				 * to the output list.
+				 */
+				*result = lappend(*result, cur);
+			}
+		}
+	}
+
+	/* Finished processing all clauses.  Now compare across strategies. */
+	if (xform[BTEqualStrategyNumber - 1])
+	{
+		PartClause *eq = xform[BTEqualStrategyNumber - 1];
+
+		for (s = BTMaxStrategyNumber; --s >= 0;)
+		{
+			PartClause *chk = xform[s];
+
+			if (!chk || s == (BTEqualStrategyNumber - 1))
+				continue;
+
+			if (partition_cmp_args(partopfamily, partopcintype, chk, eq, chk,
+								   &test_result))
+			{
+				if (!test_result)
+				{
+					*constfalse = true;
+					return;
+				}
+				/* discard the redundant key. */
+				xform[s] = NULL;
+			}
+		}
+	}
+
+	/* try to keep only one of <, <= */
+	if (xform[BTLessStrategyNumber - 1] &&
+		xform[BTLessEqualStrategyNumber - 1])
+	{
+		PartClause *lt = xform[BTLessStrategyNumber - 1],
+				   *le = xform[BTLessEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, le, lt, le,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTLessEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTLessStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/* try to keep only one of >, >= */
+	if (xform[BTGreaterStrategyNumber - 1] &&
+		xform[BTGreaterEqualStrategyNumber - 1])
+	{
+		PartClause *gt = xform[BTGreaterStrategyNumber - 1],
+				   *ge = xform[BTGreaterEqualStrategyNumber - 1];
+
+		if (partition_cmp_args(partopfamily, partopcintype, ge, gt, ge,
+							   &test_result))
+		{
+			if (test_result)
+				xform[BTGreaterEqualStrategyNumber - 1] = NULL;
+			else
+				xform[BTGreaterStrategyNumber - 1] = NULL;
+		}
+	}
+
+	/*
+	 * xform now contains "best" clauses for i'th partition key column
+	 * for given btree strategy number.  Copy them to keyclauses[i].
+	 */
+	for (s = BTMaxStrategyNumber; --s >= 0;)
+		if (xform[s])
+			*result = lappend(*result, xform[s]);
+}
+
+static bool
+partition_cmp_args(Oid partopfamily, Oid partopcintype,
+				   PartClause *op, PartClause *leftarg, PartClause *rightarg,
+				   bool *result)
+{
+	Datum	leftarg_const,
+			rightarg_const;
+
+	Assert(op->valid_cache && leftarg->valid_cache && rightarg->valid_cache);
+	if (!partkey_datum_from_expr(leftarg->constarg, &leftarg_const))
+		return false;
+	if (!partkey_datum_from_expr(rightarg->constarg, &rightarg_const))
+		return false;
+
+	/*
+	 * If the leftarg and rightarg clauses' constants are both of the type
+	 * expected by "op" clause's operator, then compare then using the
+	 * latter's comparison function.
+	 */
+	if (leftarg->op_subtype == partopcintype &&
+		rightarg->op_subtype == op->op_subtype)
+	{
+		*result = DatumGetBool(FunctionCall2Coll(&op->op_func,
+												 op->op->inputcollid,
+												 leftarg_const,
+												 rightarg_const));
+		return true;
+	}
+	else
+	{
+		/* Otherwise, look one up in the partitioning operator family. */
+		Oid		cmp_op = get_opfamily_member(partopfamily,
+											 leftarg->op_subtype,
+											 rightarg->op_subtype,
+											 op->op_strategy);
+		if (OidIsValid(cmp_op))
+		{
+			*result = DatumGetBool(OidFunctionCall2Coll(get_opcode(cmp_op),
+														op->op->inputcollid,
+														leftarg_const,
+														rightarg_const));
+			return true;
+		}
+	}
+
+	/* Couldn't do the comparison. */
+	*result = false;
+	return false;
+}
+
+/*
+ * get_partitions_for_keys
+ *		Returns the partitions that will need to be scanned for the given
+ *		bounding keys
+ *
+ * Input:
+ *	See the comments above the definition of PartScanKeyInfo to see what
+ *	kind of information is received here.
+ *
+ * Outputs:
+ *	Partition set satisfying the keys.
+ */
+static Bitmapset *
+get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
+{
+	PartitionDesc partdesc = RelationGetPartitionDesc(rel);
+	Bitmapset *result = NULL;
+
+	result = bms_add_range(result, 0, partdesc->nparts - 1);
+
+	return result;
+}
+
+/*
  * get_partition_operator
  *
  * Return oid of the operator of given strategy for a given partition key
diff --git a/src/include/catalog/partition.h b/src/include/catalog/partition.h
index 4a1ce92569..81c626fa4a 100644
--- a/src/include/catalog/partition.h
+++ b/src/include/catalog/partition.h
@@ -109,5 +109,6 @@ extern void check_default_allows_bound(Relation parent, Relation defaultRel,
 extern List *get_proposed_default_constraint(List *new_part_constaints);
 
 /* For partition-pruning */
-Bitmapset get_partitions_from_clauses(Relation relation, List *partclauses);
+extern Bitmapset *get_partitions_from_clauses(Relation relation,
+							List *partclauses);
 #endif							/* PARTITION_H */
-- 
2.11.0

From 47f02c8c1a3346ad4ad91a5544b98eeaa44caaec Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 22 Aug 2017 11:33:27 +0900
Subject: [PATCH 5/7] Some interface changes for partition_bound_{cmp/bsearch}

Introduces a notion of PartitionBoundCmpArg, which replaces the set
of arguments void *probe and bool probe_is_bound of both
partition_bound_cmp and partition_bound_bsearch.  It wasn't possible
before to specify the number of datums when a non-bound type of
probe is passed.  Slightly tweaking the existing interface to allow
specifying the same seems awkward.  So, instead encapsulate that
into PartitionBoundCmpArg.  Also, modify partition_rbound_datum_cmp
to compare caller-specifed number of datums, instead of
key->partnatts datums.
---
 src/backend/catalog/partition.c | 135 ++++++++++++++++++++++++++++------------
 1 file changed, 96 insertions(+), 39 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index 17b6a8a258..a2068ae422 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -176,6 +176,30 @@ typedef struct PartScanKeyInfo
 	NullTestType	keynullness[PARTITION_MAX_KEYS];
 } PartScanKeyInfo;
 
+/*
+ * PartitionBoundCmpArg - Caller-defined argument to be passed to
+ *						  partition_bound_cmp()
+ *
+ * The first (fixed) argument involved in a comparison is the partition bound
+ * found in the catalog, while an instance of the following struct describes
+ * either a new partition bound being compared against existing bounds
+ * (is_bound is true in that case and either lbound or rbound is set), or a
+ * new tuple's partition key specified in datums (where ndatums = number of
+ * partition key columns specified in the query).
+ */
+typedef struct PartitionBoundCmpArg
+{
+	bool	is_bound;
+	union
+	{
+		PartitionListValue	   *lbound;
+		PartitionRangeBound	   *rbound;
+	}	bound;
+
+	Datum  *datums;
+	int		ndatums;
+} PartitionBoundCmpArg;
+
 static int32 qsort_partition_list_value_cmp(const void *a, const void *b,
 							   void *arg);
 static int32 qsort_partition_rbound_cmp(const void *a, const void *b,
@@ -204,14 +228,15 @@ static int32 partition_rbound_cmp(PartitionKey key,
 					 bool lower1, PartitionRangeBound *b2);
 static int32 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums);
+						   Datum *tuple_datums, int n_tuple_datums);
 
 static int32 partition_bound_cmp(PartitionKey key,
 					PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound);
+					int offset, PartitionBoundCmpArg *arg);
 static int partition_bound_bsearch(PartitionKey key,
 						PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal);
+						PartitionBoundCmpArg *arg,
+						bool *is_equal);
 static void get_partition_dispatch_recurse(Relation rel, Relation parent,
 							   List **pds, List **leaf_part_oids);
 
@@ -903,10 +928,16 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int			offset;
 							bool		equal;
-
+							PartitionListValue lbound;
+							PartitionBoundCmpArg arg;
+
+							memset(&lbound, 0, sizeof(PartitionListValue));
+							lbound.value = val->constvalue;
+							memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+							arg.is_bound = true;
+							arg.bound.lbound = &lbound;
 							offset = partition_bound_bsearch(key, boundinfo,
-															 &val->constvalue,
-															 true, &equal);
+															 &arg, &equal);
 							if (offset >= 0 && equal)
 							{
 								overlap = true;
@@ -957,6 +988,7 @@ check_new_partition_bound(char *relname, Relation parent,
 					PartitionBoundInfo boundinfo = partdesc->boundinfo;
 					int			offset;
 					bool		equal;
+					PartitionBoundCmpArg	arg;
 
 					Assert(boundinfo &&
 						   boundinfo->strategy == PARTITION_STRATEGY_RANGE &&
@@ -978,8 +1010,11 @@ check_new_partition_bound(char *relname, Relation parent,
 					 * since the index array is initialised with an extra -1
 					 * at the end.
 					 */
-					offset = partition_bound_bsearch(key, boundinfo, lower,
-													 true, &equal);
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = true;
+					arg.bound.rbound = lower;
+					offset = partition_bound_bsearch(key, boundinfo, &arg,
+													 &equal);
 
 					if (boundinfo->indexes[offset + 1] < 0)
 					{
@@ -993,9 +1028,9 @@ check_new_partition_bound(char *relname, Relation parent,
 						{
 							int32		cmpval;
 
+							arg.bound.rbound = upper;
 							cmpval = partition_bound_cmp(key, boundinfo,
-														 offset + 1, upper,
-														 true);
+														 offset + 1, &arg);
 							if (cmpval < 0)
 							{
 								/*
@@ -3382,12 +3417,15 @@ get_partition_for_tuple(PartitionDispatch *pd,
 				{
 					bool		equal = false;
 					int			cur_offset;
+					PartitionBoundCmpArg	arg;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
+														 &arg, &equal);
 					if (cur_offset >= 0 && equal)
 						cur_index = partdesc->boundinfo->indexes[cur_offset];
 				}
@@ -3399,6 +3437,7 @@ get_partition_for_tuple(PartitionDispatch *pd,
 								range_partkey_has_null = false;
 					int			cur_offset;
 					int			i;
+					PartitionBoundCmpArg	arg;
 
 					/*
 					 * No range includes NULL, so this will be accepted by the
@@ -3429,12 +3468,13 @@ get_partition_for_tuple(PartitionDispatch *pd,
 					if (range_partkey_has_null)
 						break;
 
+					memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+					arg.is_bound = false;
+					arg.datums = values;
+					arg.ndatums = key->partnatts;
 					cur_offset = partition_bound_bsearch(key,
 														 partdesc->boundinfo,
-														 values,
-														 false,
-														 &equal);
-
+														 &arg, &equal);
 					/*
 					 * The offset returned is such that the bound at
 					 * cur_offset is less than or equal to the tuple value, so
@@ -3631,12 +3671,12 @@ partition_rbound_cmp(PartitionKey key,
 static int32
 partition_rbound_datum_cmp(PartitionKey key,
 						   Datum *rb_datums, PartitionRangeDatumKind *rb_kind,
-						   Datum *tuple_datums)
+						   Datum *tuple_datums, int n_tuple_datums)
 {
 	int			i;
 	int32		cmpval = -1;
 
-	for (i = 0; i < key->partnatts; i++)
+	for (i = 0; i < n_tuple_datums; i++)
 	{
 		if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE)
 			return -1;
@@ -3658,11 +3698,11 @@ partition_rbound_datum_cmp(PartitionKey key,
  * partition_bound_cmp
  *
  * Return whether the bound at offset in boundinfo is <, =, or > the argument
- * specified in *probe.
+ * specified in *arg.
  */
 static int32
 partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
-					int offset, void *probe, bool probe_is_bound)
+					int offset, PartitionBoundCmpArg *arg)
 {
 	Datum	   *bound_datums = boundinfo->datums[offset];
 	int32		cmpval = -1;
@@ -3670,17 +3710,35 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 	switch (key->strategy)
 	{
 		case PARTITION_STRATEGY_LIST:
-			cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
-													 key->partcollation[0],
-													 bound_datums[0],
-													 *(Datum *) probe));
-			break;
+			{
+				Datum	listdatum;
+
+				if (arg->is_bound)
+					listdatum = arg->bound.lbound->value;
+				else
+				{
+					if (arg->ndatums >= 1)
+						listdatum = arg->datums[0];
+					/*
+					 * If no tuple datum to compare with the bound, consider
+					 * the latter to be greater.
+					 */
+					else
+						return 1;
+				}
+
+				cmpval = DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0],
+														 key->partcollation[0],
+														 bound_datums[0],
+														 listdatum));
+				break;
+			}
 
 		case PARTITION_STRATEGY_RANGE:
 			{
 				PartitionRangeDatumKind *kind = boundinfo->kind[offset];
 
-				if (probe_is_bound)
+				if (arg->is_bound)
 				{
 					/*
 					 * We need to pass whether the existing bound is a lower
@@ -3691,12 +3749,13 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 					cmpval = partition_rbound_cmp(key,
 												  bound_datums, kind, lower,
-												  (PartitionRangeBound *) probe);
+												  arg->bound.rbound);
 				}
 				else
 					cmpval = partition_rbound_datum_cmp(key,
 														bound_datums, kind,
-														(Datum *) probe);
+														arg->datums,
+														arg->ndatums);
 				break;
 			}
 
@@ -3710,20 +3769,19 @@ partition_bound_cmp(PartitionKey key, PartitionBoundInfo boundinfo,
 
 /*
  * Binary search on a collection of partition bounds. Returns greatest
- * bound in array boundinfo->datums which is less than or equal to *probe.
- * If all bounds in the array are greater than *probe, -1 is returned.
+ * bound in array boundinfo->datums which is less than or equal to *arg.
+ * If all bounds in the array are greater than *arg, -1 is returned.
  *
- * *probe could either be a partition bound or a Datum array representing
- * the partition key of a tuple being routed; probe_is_bound tells which.
- * We pass that down to the comparison function so that it can interpret the
- * contents of *probe accordingly.
+ * *arg could either be a partition bound or a Datum array representing
+ * the partition key of a tuple being routed.  We simply pass that down to
+ * partition_bound_cmp where it is interpreted appropriately.
  *
  * *is_equal is set to whether the bound at the returned index is equal with
- * *probe.
+ * *arg.
  */
 static int
 partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
-						void *probe, bool probe_is_bound, bool *is_equal)
+						PartitionBoundCmpArg *arg, bool *is_equal)
 {
 	int			lo,
 				hi,
@@ -3736,8 +3794,7 @@ partition_bound_bsearch(PartitionKey key, PartitionBoundInfo boundinfo,
 		int32		cmpval;
 
 		mid = (lo + hi + 1) / 2;
-		cmpval = partition_bound_cmp(key, boundinfo, mid, probe,
-									 probe_is_bound);
+		cmpval = partition_bound_cmp(key, boundinfo, mid, arg);
 		if (cmpval <= 0)
 		{
 			lo = mid;
-- 
2.11.0

From 473dc4917eccb402b570961c92754ea342d9445d Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 31 Oct 2017 16:26:55 +0900
Subject: [PATCH 6/7] Tweak default range partition's constraint a little

When using as a predicate, it's useful for it explicitly say that
the default range partition might contain nulls, because non-default
range partitions can't.
---
 src/backend/catalog/partition.c      | 29 +++++++++++++++++++++++------
 src/test/regress/expected/update.out |  2 +-
 2 files changed, 24 insertions(+), 7 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index a2068ae422..fb973ef20c 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2934,12 +2934,29 @@ get_qual_for_range(Relation parent, PartitionBoundSpec *spec,
 
 		if (or_expr_args != NIL)
 		{
-			/* OR all the non-default partition constraints; then negate it */
-			result = lappend(result,
-							 list_length(or_expr_args) > 1
-							 ? makeBoolExpr(OR_EXPR, or_expr_args, -1)
-							 : linitial(or_expr_args));
-			result = list_make1(makeBoolExpr(NOT_EXPR, result, -1));
+			Expr   *other_parts_constr;
+
+			/*
+			 * Combine the constraints obtained for non-default partitions
+			 * using OR.  As requested, each of the OR's args doesn't include
+			 * the NOT NULL test for partition keys (which is to avoid its
+			 * useless repetition).  Add the same now.
+			 */
+			other_parts_constr =
+						makeBoolExpr(AND_EXPR,
+							lappend(get_range_nulltest(key),
+									list_length(or_expr_args) > 1
+										? makeBoolExpr(OR_EXPR, or_expr_args,
+													   -1)
+										: linitial(or_expr_args)),
+									-1);
+
+			/*
+			 * Finally, the default partition contains everything *NOT*
+			 * contained in the non-default partitions.
+			 */
+			result = list_make1(makeBoolExpr(NOT_EXPR,
+										list_make1(other_parts_constr), -1));
 		}
 
 		return result;
diff --git a/src/test/regress/expected/update.out b/src/test/regress/expected/update.out
index cef70b1a1e..40217bdf9c 100644
--- a/src/test/regress/expected/update.out
+++ b/src/test/regress/expected/update.out
@@ -227,7 +227,7 @@ create table part_def partition of range_parted default;
  a      | text    |           |          |         | extended |              | 
  b      | integer |           |          |         | plain    |              | 
 Partition of: range_parted DEFAULT
-Partition constraint: (NOT (((a = 'a'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'a'::text) AND (b >= 10) AND (b < 20)) OR ((a = 'b'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'b'::text) AND (b >= 10) AND (b < 20))))
+Partition constraint: (NOT ((a IS NOT NULL) AND (b IS NOT NULL) AND (((a = 'a'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'a'::text) AND (b >= 10) AND (b < 20)) OR ((a = 'b'::text) AND (b >= 1) AND (b < 10)) OR ((a = 'b'::text) AND (b >= 10) AND (b < 20)))))
 
 insert into range_parted values ('c', 9);
 -- ok
-- 
2.11.0

From 6b04f3560b673005fa8b9a81262754227349af89 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Wed, 18 Oct 2017 17:14:53 +0900
Subject: [PATCH 7/7] Implement get_partitions_for_keys

Disable constraint_exclusion using internal partition constraints.
---
 src/backend/catalog/partition.c         | 355 +++++++++++++++++++++++++++++++-
 src/test/regress/expected/inherit.out   |   4 +-
 src/test/regress/expected/partition.out | 121 ++++-------
 3 files changed, 391 insertions(+), 89 deletions(-)

diff --git a/src/backend/catalog/partition.c b/src/backend/catalog/partition.c
index fb973ef20c..c58c7354c6 100644
--- a/src/backend/catalog/partition.c
+++ b/src/backend/catalog/partition.c
@@ -2409,10 +2409,361 @@ partition_cmp_args(Oid partopfamily, Oid partopcintype,
 static Bitmapset *
 get_partitions_for_keys(Relation rel, PartScanKeyInfo *keys)
 {
-	PartitionDesc partdesc = RelationGetPartitionDesc(rel);
+	PartitionKey	partkey = RelationGetPartitionKey(rel);
+	PartitionDesc	partdesc = RelationGetPartitionDesc(rel);
+	PartitionBoundInfo	boundinfo = partdesc->boundinfo;
 	Bitmapset *result = NULL;
+	PartitionBoundCmpArg	arg;
+	int		i,
+			eqoff,
+			minoff,
+			maxoff;
+	bool	is_equal;
+
+	/* Return an empty set if no partitions to see. */
+	if (partdesc->nparts == 0)
+		return NULL;
+
+	/*
+	 * Check if any of the scan keys are null.  If so, return the only
+	 * null-accepting partition if boundinfo says there is one.
+	 */
+	for (i = 0; i < partkey->partnatts; i++)
+	{
+		if (keys->keynullness[i] == IS_NULL)
+		{
+			int		other_idx = -1;
+
+			/*
+			 * Note that only one of the null-accepting partition and the
+			 * default partition can be holding null values at any given
+			 * time.
+			 */
+			if (partition_bound_accepts_nulls(boundinfo)||
+				partition_bound_has_default(boundinfo))
+				other_idx = partition_bound_accepts_nulls(boundinfo)
+								? boundinfo->null_index
+								: boundinfo->default_index;
+			if (other_idx >= 0)
+				result = bms_make_singleton(other_idx);
+
+			return result;
+		}
+	}
+
+	/*
+	 * If there are no datums to compare keys with, but there exists a
+	 * partition, the latter must be a partition that accepts only nulls
+	 * or a default partition.  If it is the former and we didn't already
+	 * return it as the only scannable partition, that means the query
+	 * doesn't want null values in its output.  So, all of what the query
+	 * wants instead must be in the default partition.
+	 */
+	if (boundinfo->ndatums == 0)
+	{
+		if (partition_bound_has_default(boundinfo))
+			result = bms_make_singleton(boundinfo->default_index);
+		return result;
+	}
+	/* No bounding keys, so just return all partitions. */
+	else if (keys->n_eqkeys + keys->n_minkeys + keys->n_maxkeys == 0)
+	{
+		result = bms_add_range(result, 0, partdesc->nparts - 1);
+		return result;
+	}
+
+	/* Valid keys->eqkeys must provide all partition keys. */
+	Assert(keys->n_eqkeys == 0 || keys->n_eqkeys == partkey->partnatts);
+	if (keys->n_eqkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->eqkeys;
+		arg.ndatums = keys->n_eqkeys;
+		eqoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		if (eqoff >= 0)
+		{
+			switch (partkey->strategy)
+			{
+				case PARTITION_STRATEGY_LIST:
+					/* For list partition, must exactly match the datum. */
+					if (!is_equal)
+						eqoff = -1;
+					break;
+
+				case PARTITION_STRATEGY_RANGE:
+					/*
+					 * eqoff is gives us the bound that is known to be <=
+					 * eqkeys given how partition_bound_bsearch works.  The
+					 * bound at eqoff + 1, then, would be the upper bound of
+					 * the only partition that needs to be scanned.
+					 */
+					if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+						eqoff += 1;
+			}
+		}
 
-	result = bms_add_range(result, 0, partdesc->nparts - 1);
+		/*
+		 * Ask later code to include the default partition, because eqkeys
+		 * didn't identify a specific partition or identified a range
+		 * of unassigned values.
+		 */
+		if (eqoff >= 0 && boundinfo->indexes[eqoff] >= 0)
+			result = bms_make_singleton(boundinfo->indexes[eqoff]);
+		else if (partition_bound_has_default(boundinfo))
+			result = bms_make_singleton(boundinfo->default_index);
+
+		/* There are no minkeys and maxkeys when eqkeys is valid. */
+		return result;
+	}
+
+	/*
+	 * Find the leftmost bound that satisfies the query, i.e., one that
+	 * satisfies minkeys.
+	 */
+	minoff = 0;
+	if (keys->n_minkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->minkeys;
+		arg.ndatums = keys->n_minkeys;
+		minoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+
+				/*
+				 * minoff set to -1 means all datums are greater than
+				 * minkeys, which means all partitions satisfy minkeys.
+				 */
+				if (minoff == -1)
+					minoff = 0;
+
+				/*
+				 * minkeys matched one of the datums (because, is_equal), but
+				 * the query may have asked to exclude that value.  If so,
+				 * move to the bound on the right, which doesn't necessarily
+				 * mean we're excluding the list partition containing that
+				 * value, because there very well might be values in the range
+				 * thus selected that belong to the partition to which the
+				 * matched value (minkeys) also belongs.
+				 */
+				if (is_equal && !keys->min_incl)
+					minoff++;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If only a prefix of the whole partition key is provided,
+				 * there will be multiple partitions whose bound share the
+				 * same prefix.  If minkey is inclusive, we must make minoff
+				 * point to the leftmost such bound, making the result contain
+				 * all such partitions.  If it is exclusive, we must move
+				 * minoff to the right such that minoff points to the first
+				 * partition whose bound is greater than this prefix, thus
+				 * excluding all aforementioned partitions from appearing in
+				 * the result.
+				 */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->min_incl)
+							minoff -= 1;
+						else
+							minoff += 1;
+						if (minoff < 0 || minoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 minoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (!keys->min_incl)
+						minoff -= 1;
+				}
+
+				/*
+				 * At this point, minoff gives us the leftmost bound that is
+				 * known to be <= query's minkey.  The bound at minoff + 1 (if
+				 * there is one), then, would be the upper bound of the
+				 * leftmost partition that needs to be scanned.
+				 */
+				if (minoff < boundinfo->ndatums - 1)
+					minoff += 1;
+				break;
+		}
+	}
+
+	/*
+	 * Find the rightmost bound that satisfies the query, i.e., one that
+	 * satisfies maxkeys.
+	 */
+	if (partkey->strategy == PARTITION_STRATEGY_RANGE)
+		/* 1 more index than datums in this case */
+		maxoff = boundinfo->ndatums;
+	else
+		maxoff = boundinfo->ndatums - 1;
+	if (keys->n_maxkeys > 0)
+	{
+		memset(&arg, 0, sizeof(PartitionBoundCmpArg));
+		arg.datums = keys->maxkeys;
+		arg.ndatums = keys->n_maxkeys;
+		maxoff = partition_bound_bsearch(partkey, boundinfo, &arg, &is_equal);
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/* See the comment above for minkeys. */
+				if (is_equal && !keys->max_incl)
+					maxoff--;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/* See the comment above for minkeys. */
+				if (is_equal && arg.ndatums < partkey->partnatts)
+				{
+					int32	cmpval;
+
+					is_equal = false;
+					do
+					{
+						if (keys->max_incl)
+							maxoff += 1;
+						else
+							maxoff -= 1;
+						if (maxoff < 0 || maxoff >= boundinfo->ndatums)
+							break;
+						cmpval = partition_bound_cmp(partkey, boundinfo,
+													 maxoff, &arg);
+					} while (cmpval == 0);
+
+					/* Back up if went too far. */
+					if (keys->max_incl)
+						maxoff -= 1;
+				}
+
+				/*
+				 * At this point, maxoff gives us the rightmost bound that is
+				 * known to be <= query's maxkey.  The bound at maxoff + 1,
+				 * then, would be the upper bound of the rightmost partition
+				 * that needs to be scanned.  Although, if the bound is equal
+				 * to maxkeys and the latter is not inclusive, then the bound
+				 * at maxoff itself is the upper bound of the rightmost
+				 * partition that needs to be scanned.
+				 */
+				if (!is_equal || keys->max_incl)
+					maxoff += 1;
+
+				break;
+		}
+	}
+
+	/*
+	 * minoff or maxoff set to -1 means none of the datums in
+	 * PartitionBoundInfo satisfies both minkeys and maxkeys.  If both are set
+	 * to a valid datum offset, that means there exists at least some
+	 * datums (and hence partitions) satisfying both minkeys and maxkeys.
+	 */
+	if (minoff >= 0 && maxoff >= 0)
+	{
+		bool	list_include_def = false,
+				range_include_def = false;
+
+		switch (partkey->strategy)
+		{
+			case PARTITION_STRATEGY_LIST:
+				/*
+				 * Add to the other_parts, list partition indexes are not
+				 * monotonously increasing like range partitions' are.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+					result = bms_add_member(result, boundinfo->indexes[i]);
+
+				/*
+				 * For range queries, always include the default list
+				 * partition.  Because list partitions divide the key space
+				 * in a discontinuous manner, not all values in the given
+				 * range will have a partition assigned.
+				 */
+				list_include_def = true;
+				break;
+
+			case PARTITION_STRATEGY_RANGE:
+				/*
+				 * If the bound at minoff or maxoff looks like it's an upper
+				 * bound of an unassigned range of values, move to the
+				 * adjacent bound which must be the upper bound of the
+				 * leftmost or rightmost partition, respectively, that needs
+				 * to be scanned.
+				 *
+				 * By doing that, we skip over a portion of values that do
+				 * indeed satisfy the query, but don't have a valid partition
+				 * assigned. The default partition would've been included to
+				 * cover those values.  Although, if the original bound in
+				 * question is an infinite value, there would not be any
+				 * unassigned range to speak of, because the range is unbounded
+				 * in that direction by definition, so no need to include the
+				 * default.
+				 */
+				if (boundinfo->indexes[minoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_minkeys > 0)
+						lastkey = keys->n_minkeys - 1;
+					if (boundinfo->kind[minoff][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					minoff += 1;
+				}
+
+				if (maxoff >= 1 && boundinfo->indexes[maxoff] < 0)
+				{
+					int		lastkey = partkey->partnatts - 1;
+
+					if (keys->n_maxkeys > 0)
+						lastkey = keys->n_maxkeys - 1;
+					if (boundinfo->kind[maxoff - 1][lastkey] ==
+												PARTITION_RANGE_DATUM_VALUE)
+					{
+						range_include_def = true;
+					}
+					maxoff -= 1;
+				}
+
+				if (minoff <= maxoff)
+					result = bms_add_range(result,
+										   boundinfo->indexes[minoff],
+										   boundinfo->indexes[maxoff]);
+				/*
+				 * There might exist a range of values unassigned to any
+				 * non-default range partition between the datums at
+				 * minoff and maxoff.
+				 */
+				for (i = minoff; i <= maxoff; i++)
+				{
+					if (boundinfo->indexes[i] < 0)
+					{
+						range_include_def = true;
+						break;
+					}
+				}
+				break;
+		}
+
+		if ((list_include_def || range_include_def) &&
+			partition_bound_has_default(boundinfo))
+			result = bms_add_member(result, boundinfo->default_index);
+	}
+	else if (partition_bound_has_default(boundinfo))
+		result = bms_add_member(result, boundinfo->default_index);
 
 	return result;
 }
diff --git a/src/test/regress/expected/inherit.out b/src/test/regress/expected/inherit.out
index c698faff2f..661f137122 100644
--- a/src/test/regress/expected/inherit.out
+++ b/src/test/regress/expected/inherit.out
@@ -1713,11 +1713,9 @@ explain (costs off) select * from list_parted where a = 'ab' or a in (null, 'cd'
  Append
    ->  Seq Scan on part_ab_cd
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-   ->  Seq Scan on part_ef_gh
-         Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
    ->  Seq Scan on part_null_xy
          Filter: (((a)::text = 'ab'::text) OR ((a)::text = ANY ('{NULL,cd}'::text[])))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from list_parted where a = 'ab';
                 QUERY PLAN                
diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
index 484c6fe750..206c6de5b7 100644
--- a/src/test/regress/expected/partition.out
+++ b/src/test/regress/expected/partition.out
@@ -208,16 +208,14 @@ explain (costs off) select * from rlp where 1 > a;	/* commutates */
 (3 rows)
 
 explain (costs off) select * from rlp where a <= 1;
-              QUERY PLAN               
----------------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on rlp1
          Filter: (a <= 1)
    ->  Seq Scan on rlp2
          Filter: (a <= 1)
-   ->  Seq Scan on rlp_default_default
-         Filter: (a <= 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 1;
        QUERY PLAN        
@@ -519,15 +517,13 @@ explain (costs off) select * from rlp where a <= 31;
          Filter: (a <= 31)
    ->  Seq Scan on rlp5_1
          Filter: (a <= 31)
-   ->  Seq Scan on rlp5_default
-         Filter: (a <= 31)
    ->  Seq Scan on rlp_default_10
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_30
          Filter: (a <= 31)
    ->  Seq Scan on rlp_default_default
          Filter: (a <= 31)
-(29 rows)
+(27 rows)
 
 explain (costs off) select * from rlp where a = 1 or a = 7;
               QUERY PLAN              
@@ -575,9 +571,7 @@ explain (costs off) select * from rlp where a > 20 and a < 27;
          Filter: ((a > 20) AND (a < 27))
    ->  Seq Scan on rlp4_2
          Filter: ((a > 20) AND (a < 27))
-   ->  Seq Scan on rlp4_default
-         Filter: ((a > 20) AND (a < 27))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from rlp where a = 29;
            QUERY PLAN           
@@ -651,8 +645,6 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
                             QUERY PLAN                             
 -------------------------------------------------------------------
  Append
-   ->  Seq Scan on rlp2
-         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3abcd
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3efgh
@@ -661,7 +653,7 @@ explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a =
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
    ->  Seq Scan on rlp3_default
          Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
-(11 rows)
+(9 rows)
 
 -- multi-column keys
 create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
@@ -675,16 +667,14 @@ create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
 create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
 create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
 explain (costs off) select * from mc3p where a = 1;
-           QUERY PLAN           
---------------------------------
+       QUERY PLAN        
+-------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a = 1)
    ->  Seq Scan on mc3p1
          Filter: (a = 1)
-   ->  Seq Scan on mc3p_default
-         Filter: (a = 1)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
                  QUERY PLAN                 
@@ -702,9 +692,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
          Filter: ((a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
                        QUERY PLAN                       
@@ -714,9 +702,7 @@ explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
    ->  Seq Scan on mc3p1
          Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
                            QUERY PLAN                            
@@ -730,9 +716,7 @@ explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
    ->  Seq Scan on mc3p4
          Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a > 10;
            QUERY PLAN           
@@ -771,16 +755,14 @@ explain (costs off) select * from mc3p where a >= 10;
 (17 rows)
 
 explain (costs off) select * from mc3p where a < 10;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN        
+--------------------------
  Append
    ->  Seq Scan on mc3p0
          Filter: (a < 10)
    ->  Seq Scan on mc3p1
          Filter: (a < 10)
-   ->  Seq Scan on mc3p_default
-         Filter: (a < 10)
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
                   QUERY PLAN                   
@@ -792,9 +774,7 @@ explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
          Filter: ((a <= 10) AND (abs(b) < 10))
    ->  Seq Scan on mc3p2
          Filter: ((a <= 10) AND (abs(b) < 10))
-   ->  Seq Scan on mc3p_default
-         Filter: ((a <= 10) AND (abs(b) < 10))
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
                  QUERY PLAN                  
@@ -821,8 +801,8 @@ explain (costs off) select * from mc3p where a > 20;
 (3 rows)
 
 explain (costs off) select * from mc3p where a >= 20;
-           QUERY PLAN           
---------------------------------
+        QUERY PLAN         
+---------------------------
  Append
    ->  Seq Scan on mc3p5
          Filter: (a >= 20)
@@ -830,9 +810,7 @@ explain (costs off) select * from mc3p where a >= 20;
          Filter: (a >= 20)
    ->  Seq Scan on mc3p7
          Filter: (a >= 20)
-   ->  Seq Scan on mc3p_default
-         Filter: (a >= 20)
-(9 rows)
+(7 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
                                                            QUERY PLAN                                                            
@@ -872,9 +850,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
    ->  Seq Scan on mc3p5
          Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
-(11 rows)
+(9 rows)
 
 explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
                       QUERY PLAN                      
@@ -912,9 +888,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
    ->  Seq Scan on mc3p4
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
-(13 rows)
+(11 rows)
 
 explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
                                  QUERY PLAN                                  
@@ -926,9 +900,7 @@ explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 a
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
    ->  Seq Scan on mc3p2
          Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-   ->  Seq Scan on mc3p_default
-         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
-(9 rows)
+(7 rows)
 
 -- a simpler multi-column keys case
 create table mc2p (a int, b int) partition by range (a, b);
@@ -999,28 +971,20 @@ explain (costs off) select * from boolpart where a in (true, false);
 (5 rows)
 
 explain (costs off) select * from boolpart where a = false;
-             QUERY PLAN             
-------------------------------------
+          QUERY PLAN          
+------------------------------
  Append
    ->  Seq Scan on boolpart_f
          Filter: (NOT a)
-   ->  Seq Scan on boolpart_t
-         Filter: (NOT a)
-   ->  Seq Scan on boolpart_default
-         Filter: (NOT a)
-(7 rows)
+(3 rows)
 
 explain (costs off) select * from boolpart where not a = false;
-             QUERY PLAN             
-------------------------------------
+          QUERY PLAN          
+------------------------------
  Append
-   ->  Seq Scan on boolpart_f
-         Filter: a
    ->  Seq Scan on boolpart_t
          Filter: a
-   ->  Seq Scan on boolpart_default
-         Filter: a
-(7 rows)
+(3 rows)
 
 explain (costs off) select * from boolpart where a is true or a is not true;
                     QUERY PLAN                    
@@ -1030,33 +994,22 @@ explain (costs off) select * from boolpart where a is true or a is not true;
          Filter: ((a IS TRUE) OR (a IS NOT TRUE))
    ->  Seq Scan on boolpart_t
          Filter: ((a IS TRUE) OR (a IS NOT TRUE))
-   ->  Seq Scan on boolpart_default
-         Filter: ((a IS TRUE) OR (a IS NOT TRUE))
-(7 rows)
+(5 rows)
 
 explain (costs off) select * from boolpart where a is not true;
-             QUERY PLAN             
-------------------------------------
+           QUERY PLAN            
+---------------------------------
  Append
    ->  Seq Scan on boolpart_f
          Filter: (a IS NOT TRUE)
-   ->  Seq Scan on boolpart_t
-         Filter: (a IS NOT TRUE)
-   ->  Seq Scan on boolpart_default
-         Filter: (a IS NOT TRUE)
-(7 rows)
+(3 rows)
 
 explain (costs off) select * from boolpart where a is not true and a is not false;
-                       QUERY PLAN                       
---------------------------------------------------------
- Append
-   ->  Seq Scan on boolpart_f
-         Filter: ((a IS NOT TRUE) AND (a IS NOT FALSE))
-   ->  Seq Scan on boolpart_t
-         Filter: ((a IS NOT TRUE) AND (a IS NOT FALSE))
-   ->  Seq Scan on boolpart_default
-         Filter: ((a IS NOT TRUE) AND (a IS NOT FALSE))
-(7 rows)
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
 
 explain (costs off) select * from boolpart where a is unknown;
              QUERY PLAN             
-- 
2.11.0

From 1c8d55f8235f28135ab8f8f00e23afe469253e06 Mon Sep 17 00:00:00 2001
From: amit <amitlangote09@gmail.com>
Date: Tue, 5 Sep 2017 10:28:23 +0900
Subject: [PATCH 1/7] Add new tests for partition-pruning

---
 src/test/regress/expected/partition.out | 1085 +++++++++++++++++++++++++++++++
 src/test/regress/parallel_schedule      |    2 +-
 src/test/regress/serial_schedule        |    1 +
 src/test/regress/sql/partition.sql      |  155 +++++
 4 files changed, 1242 insertions(+), 1 deletion(-)
 create mode 100644 src/test/regress/expected/partition.out
 create mode 100644 src/test/regress/sql/partition.sql

diff --git a/src/test/regress/expected/partition.out b/src/test/regress/expected/partition.out
new file mode 100644
index 0000000000..484c6fe750
--- /dev/null
+++ b/src/test/regress/expected/partition.out
@@ -0,0 +1,1085 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+          QUERY PLAN          
+------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+   ->  Seq Scan on lp_bc
+   ->  Seq Scan on lp_ef
+   ->  Seq Scan on lp_g
+   ->  Seq Scan on lp_null
+   ->  Seq Scan on lp_default
+(7 rows)
+
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+                        QUERY PLAN                         
+-----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a < 'd'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a > 'a'::bpchar) AND (a <= 'd'::bpchar))
+(7 rows)
+
+explain (costs off) select * from lp where a = 'a';
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a = 'a'::bpchar)
+(3 rows)
+
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ('a'::bpchar = a)
+(3 rows)
+
+explain (costs off) select * from lp where a is not null;
+           QUERY PLAN            
+---------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_bc
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_ef
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_g
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on lp_default
+         Filter: (a IS NOT NULL)
+(11 rows)
+
+explain (costs off) select * from lp where a is null;
+         QUERY PLAN          
+-----------------------------
+ Append
+   ->  Seq Scan on lp_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+   ->  Seq Scan on lp_bc
+         Filter: ((a = 'a'::bpchar) OR (a = 'c'::bpchar))
+(5 rows)
+
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+                                   QUERY PLAN                                   
+--------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+   ->  Seq Scan on lp_bc
+         Filter: ((a IS NOT NULL) AND ((a = 'a'::bpchar) OR (a = 'c'::bpchar)))
+(5 rows)
+
+explain (costs off) select * from lp where a <> 'g';
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on lp_ad
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_bc
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_ef
+         Filter: (a <> 'g'::bpchar)
+   ->  Seq Scan on lp_default
+         Filter: (a <> 'g'::bpchar)
+(9 rows)
+
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+                         QUERY PLAN                          
+-------------------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_ef
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_g
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+   ->  Seq Scan on lp_default
+         Filter: ((a <> 'a'::bpchar) AND (a <> 'd'::bpchar))
+(9 rows)
+
+explain (costs off) select * from lp where a not in ('a', 'd');
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on lp_bc
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_ef
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_g
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+   ->  Seq Scan on lp_default
+         Filter: (a <> ALL ('{a,d}'::bpchar[]))
+(9 rows)
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: (a = 'a'::text COLLATE "C")
+(3 rows)
+
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on coll_pruning_a
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_b
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+   ->  Seq Scan on coll_pruning_def
+         Filter: ((a)::text = 'a'::text COLLATE "POSIX")
+(7 rows)
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp_default_null partition of rlp_default for values in (null);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+explain (costs off) select * from rlp where a < 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 1)
+(3 rows)
+
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (1 > a)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 1;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 1)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 1)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 1;
+       QUERY PLAN        
+-------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = 1)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+            QUERY PLAN             
+-----------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (a = '1'::bigint)
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp3_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_2
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp4_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_1
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp5_default
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_null
+         Filter: ((a)::numeric = '1'::numeric)
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a)::numeric = '1'::numeric)
+(31 rows)
+
+explain (costs off) select * from rlp where a <= 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 10)
+(9 rows)
+
+explain (costs off) select * from rlp where a > 10;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a > 10)
+   ->  Seq Scan on rlp3_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_2
+         Filter: (a > 10)
+   ->  Seq Scan on rlp4_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 10)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a > 10)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 10)
+(23 rows)
+
+explain (costs off) select * from rlp where a < 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a < 15)
+   ->  Seq Scan on rlp2
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a < 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a < 15)
+(9 rows)
+
+explain (costs off) select * from rlp where a <= 15;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 15)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 15)
+(17 rows)
+
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 15) AND ((b)::text = 'ab'::text))
+(17 rows)
+
+explain (costs off) select * from rlp where a = 16;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a = 16)
+   ->  Seq Scan on rlp3_default
+         Filter: (a = 16)
+(9 rows)
+
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+                                 QUERY PLAN                                 
+----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: ((a = 16) AND ((b)::text = ANY ('{not,in,here}'::text[])))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+                       QUERY PLAN                        
+---------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text < 'ab'::text) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+                        QUERY PLAN                        
+----------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: (((b)::text <= 'ab'::text) AND (a = 16))
+(5 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is null;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NULL) AND (a = 16))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 16 and b is not null;
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((b IS NOT NULL) AND (a = 16))
+   ->  Seq Scan on rlp3_default
+         Filter: ((b IS NOT NULL) AND (a = 16))
+(9 rows)
+
+explain (costs off) select * from rlp where a is null;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_null
+         Filter: (a IS NULL)
+(3 rows)
+
+explain (costs off) select * from rlp where a is not null;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp2
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp3_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_2
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp4_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp5_1
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp5_default
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a IS NOT NULL)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a IS NOT NULL)
+(29 rows)
+
+explain (costs off) select * from rlp where a > 30;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp5_1
+         Filter: (a > 30)
+   ->  Seq Scan on rlp5_default
+         Filter: (a > 30)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a > 30)
+(7 rows)
+
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+            QUERY PLAN            
+----------------------------------
+ Append
+   ->  Seq Scan on rlp_default_30
+         Filter: (a = 30)
+(3 rows)
+
+explain (costs off) select * from rlp where a <= 31;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3abcd
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3efgh
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3nullxy
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp3_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_2
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp4_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_1
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp5_default
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_10
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a <= 31)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a <= 31)
+(29 rows)
+
+explain (costs off) select * from rlp where a = 1 or a = 7;
+              QUERY PLAN              
+--------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR (a = 7))
+(3 rows)
+
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+                      QUERY PLAN                       
+-------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_null
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a = 1) OR ((b)::text = 'ab'::text))
+(25 rows)
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 20) AND (a < 27))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 20) AND (a < 27))
+(7 rows)
+
+explain (costs off) select * from rlp where a = 29;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a = 29)
+(3 rows)
+
+explain (costs off) select * from rlp where a >= 29;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on rlp4_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_1
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp5_default
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_30
+         Filter: (a >= 29)
+   ->  Seq Scan on rlp_default_default
+         Filter: (a >= 29)
+(11 rows)
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+               QUERY PLAN               
+----------------------------------------
+ Append
+   ->  Seq Scan on rlp_default_10
+         Filter: ((a > 1) AND (a = 10))
+(3 rows)
+
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+               QUERY PLAN                
+-----------------------------------------
+ Append
+   ->  Seq Scan on rlp3abcd
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3efgh
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3nullxy
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp3_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_2
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp4_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_1
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp5_default
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_30
+         Filter: ((a > 1) AND (a >= 15))
+   ->  Seq Scan on rlp_default_default
+         Filter: ((a > 1) AND (a >= 15))
+(23 rows)
+
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+        QUERY PLAN        
+--------------------------
+ Result
+   One-Time Filter: false
+(2 rows)
+
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+                            QUERY PLAN                             
+-------------------------------------------------------------------
+ Append
+   ->  Seq Scan on rlp2
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3abcd
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3efgh
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3nullxy
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+   ->  Seq Scan on rlp3_default
+         Filter: (((a = 1) AND (a = 3)) OR ((a > 1) AND (a = 15)))
+(11 rows)
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+explain (costs off) select * from mc3p where a = 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p1
+         Filter: (a = 1)
+   ->  Seq Scan on mc3p_default
+         Filter: (a = 1)
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) < 1))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+                 QUERY PLAN                 
+--------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((c < 8) AND (a = 1) AND (abs(b) = 1))
+(7 rows)
+
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+                           QUERY PLAN                            
+-----------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p3
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 10) AND (abs(b) >= 5) AND (abs(b) <= 35))
+(11 rows)
+
+explain (costs off) select * from mc3p where a > 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a > 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a > 10)
+(9 rows)
+
+explain (costs off) select * from mc3p where a >= 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p2
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p3
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p4
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 10)
+(17 rows)
+
+explain (costs off) select * from mc3p where a < 10;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p1
+         Filter: (a < 10)
+   ->  Seq Scan on mc3p_default
+         Filter: (a < 10)
+(7 rows)
+
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+                  QUERY PLAN                   
+-----------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p1
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p2
+         Filter: ((a <= 10) AND (abs(b) < 10))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a <= 10) AND (abs(b) < 10))
+(9 rows)
+
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+                 QUERY PLAN                  
+---------------------------------------------
+ Append
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 11) AND (abs(b) = 0))
+(3 rows)
+
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+                         QUERY PLAN                         
+------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 20) AND (c = 100) AND (abs(b) = 10))
+(3 rows)
+
+explain (costs off) select * from mc3p where a > 20;
+        QUERY PLAN        
+--------------------------
+ Append
+   ->  Seq Scan on mc3p7
+         Filter: (a > 20)
+(3 rows)
+
+explain (costs off) select * from mc3p where a >= 20;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc3p5
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p6
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p7
+         Filter: (a >= 20)
+   ->  Seq Scan on mc3p_default
+         Filter: (a >= 20)
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+                                                           QUERY PLAN                                                            
+---------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)))
+(7 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+                                                                 QUERY PLAN                                                                 
+--------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1))
+(9 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+                                                                      QUERY PLAN                                                                       
+-------------------------------------------------------------------------------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p5
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1) AND (c = 1)) OR ((a = 10) AND (abs(b) = 5) AND (c = 10)) OR ((a > 11) AND (a < 20)) OR (a < 1) OR (a = 1))
+(11 rows)
+
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p1
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p2
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p4
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p5
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p6
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p7
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+   ->  Seq Scan on mc3p_default
+         Filter: ((a = 1) OR (abs(b) = 1) OR (c = 1))
+(17 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+                                  QUERY PLAN                                  
+------------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p3
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p4
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 10)))
+(13 rows)
+
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Append
+   ->  Seq Scan on mc3p0
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p1
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p2
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+   ->  Seq Scan on mc3p_default
+         Filter: (((a = 1) AND (abs(b) = 1)) OR ((a = 10) AND (abs(b) = 9)))
+(9 rows)
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+explain (costs off) select * from mc2p where a < 2;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc2p0
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p1
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p2
+         Filter: (a < 2)
+   ->  Seq Scan on mc2p_default
+         Filter: (a < 2)
+(9 rows)
+
+explain (costs off) select * from mc2p where a = 2 and b < 1;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on mc2p3
+         Filter: ((b < 1) AND (a = 2))
+(3 rows)
+
+explain (costs off) select * from mc2p where a > 1;
+           QUERY PLAN           
+--------------------------------
+ Append
+   ->  Seq Scan on mc2p3
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p4
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p5
+         Filter: (a > 1)
+   ->  Seq Scan on mc2p_default
+         Filter: (a > 1)
+(9 rows)
+
+explain (costs off) select * from mc2p where a = 1 and b > 1;
+              QUERY PLAN               
+---------------------------------------
+ Append
+   ->  Seq Scan on mc2p2
+         Filter: ((b > 1) AND (a = 1))
+(3 rows)
+
+-- boolean partitioning
+create table boolpart (a bool) partition by list (a);
+create table boolpart_default partition of boolpart default;
+create table boolpart_t partition of boolpart for values in ('true');
+create table boolpart_f partition of boolpart for values in ('false');
+explain (costs off) select * from boolpart where a in (true, false);
+                   QUERY PLAN                   
+------------------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: (a = ANY ('{t,f}'::boolean[]))
+   ->  Seq Scan on boolpart_t
+         Filter: (a = ANY ('{t,f}'::boolean[]))
+(5 rows)
+
+explain (costs off) select * from boolpart where a = false;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: (NOT a)
+   ->  Seq Scan on boolpart_t
+         Filter: (NOT a)
+   ->  Seq Scan on boolpart_default
+         Filter: (NOT a)
+(7 rows)
+
+explain (costs off) select * from boolpart where not a = false;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: a
+   ->  Seq Scan on boolpart_t
+         Filter: a
+   ->  Seq Scan on boolpart_default
+         Filter: a
+(7 rows)
+
+explain (costs off) select * from boolpart where a is true or a is not true;
+                    QUERY PLAN                    
+--------------------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: ((a IS TRUE) OR (a IS NOT TRUE))
+   ->  Seq Scan on boolpart_t
+         Filter: ((a IS TRUE) OR (a IS NOT TRUE))
+   ->  Seq Scan on boolpart_default
+         Filter: ((a IS TRUE) OR (a IS NOT TRUE))
+(7 rows)
+
+explain (costs off) select * from boolpart where a is not true;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: (a IS NOT TRUE)
+   ->  Seq Scan on boolpart_t
+         Filter: (a IS NOT TRUE)
+   ->  Seq Scan on boolpart_default
+         Filter: (a IS NOT TRUE)
+(7 rows)
+
+explain (costs off) select * from boolpart where a is not true and a is not false;
+                       QUERY PLAN                       
+--------------------------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: ((a IS NOT TRUE) AND (a IS NOT FALSE))
+   ->  Seq Scan on boolpart_t
+         Filter: ((a IS NOT TRUE) AND (a IS NOT FALSE))
+   ->  Seq Scan on boolpart_default
+         Filter: ((a IS NOT TRUE) AND (a IS NOT FALSE))
+(7 rows)
+
+explain (costs off) select * from boolpart where a is unknown;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: (a IS UNKNOWN)
+   ->  Seq Scan on boolpart_t
+         Filter: (a IS UNKNOWN)
+   ->  Seq Scan on boolpart_default
+         Filter: (a IS UNKNOWN)
+(7 rows)
+
+explain (costs off) select * from boolpart where a is not unknown;
+             QUERY PLAN             
+------------------------------------
+ Append
+   ->  Seq Scan on boolpart_f
+         Filter: (a IS NOT UNKNOWN)
+   ->  Seq Scan on boolpart_t
+         Filter: (a IS NOT UNKNOWN)
+   ->  Seq Scan on boolpart_default
+         Filter: (a IS NOT UNKNOWN)
+(7 rows)
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p, boolpart;
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index aa5e6af621..38dfe618b5 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -60,7 +60,7 @@ test: create_index create_view
 # ----------
 # Another group of parallel tests
 # ----------
-test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func
+test: create_aggregate create_function_3 create_cast constraints triggers inherit create_table_like typed_table vacuum drop_if_exists updatable_views rolenames roleattributes create_am hash_func partition
 
 # ----------
 # sanity_check does a vacuum, affecting the sort order of SELECT *
diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule
index 3866314a92..17d88e5ca9 100644
--- a/src/test/regress/serial_schedule
+++ b/src/test/regress/serial_schedule
@@ -71,6 +71,7 @@ test: create_cast
 test: constraints
 test: triggers
 test: inherit
+test: partition
 test: create_table_like
 test: typed_table
 test: vacuum
diff --git a/src/test/regress/sql/partition.sql b/src/test/regress/sql/partition.sql
new file mode 100644
index 0000000000..392cb9bbe9
--- /dev/null
+++ b/src/test/regress/sql/partition.sql
@@ -0,0 +1,155 @@
+--
+-- Test partitioning planner code
+--
+create table lp (a char) partition by list (a);
+create table lp_default partition of lp default;
+create table lp_ef partition of lp for values in ('e', 'f');
+create table lp_ad partition of lp for values in ('a', 'd');
+create table lp_bc partition of lp for values in ('b', 'c');
+create table lp_g partition of lp for values in ('g');
+create table lp_null partition of lp for values in (null);
+explain (costs off) select * from lp;
+explain (costs off) select * from lp where a > 'a' and a < 'd';
+explain (costs off) select * from lp where a > 'a' and a <= 'd';
+explain (costs off) select * from lp where a = 'a';
+explain (costs off) select * from lp where 'a' = a;	/* commutates */
+explain (costs off) select * from lp where a is not null;
+explain (costs off) select * from lp where a is null;
+explain (costs off) select * from lp where a = 'a' or a = 'c';
+explain (costs off) select * from lp where a is not null and (a = 'a' or a = 'c');
+explain (costs off) select * from lp where a <> 'g';
+explain (costs off) select * from lp where a <> 'a' and a <> 'd';
+explain (costs off) select * from lp where a not in ('a', 'd');
+
+-- collation matches the partitioning collation, pruning works
+create table coll_pruning (a text collate "C") partition by list (a);
+create table coll_pruning_a partition of coll_pruning for values in ('a');
+create table coll_pruning_b partition of coll_pruning for values in ('b');
+create table coll_pruning_def partition of coll_pruning default;
+explain (costs off) select * from coll_pruning where a collate "C" = 'a' collate "C";
+-- collation doesn't match the partitioning collation, no pruning occurs
+explain (costs off) select * from coll_pruning where a collate "POSIX" = 'a' collate "POSIX";
+
+create table rlp (a int, b varchar) partition by range (a);
+create table rlp_default partition of rlp default partition by list (a);
+create table rlp_default_default partition of rlp_default default;
+create table rlp_default_10 partition of rlp_default for values in (10);
+create table rlp_default_30 partition of rlp_default for values in (30);
+create table rlp_default_null partition of rlp_default for values in (null);
+create table rlp1 partition of rlp for values from (minvalue) to (1);
+create table rlp2 partition of rlp for values from (1) to (10);
+
+create table rlp3 (b varchar, a int) partition by list (b varchar_ops);
+create table rlp3_default partition of rlp3 default;
+create table rlp3abcd partition of rlp3 for values in ('ab', 'cd');
+create table rlp3efgh partition of rlp3 for values in ('ef', 'gh');
+create table rlp3nullxy partition of rlp3 for values in (null, 'xy');
+alter table rlp attach partition rlp3 for values from (15) to (20);
+
+create table rlp4 partition of rlp for values from (20) to (30) partition by range (a);
+create table rlp4_default partition of rlp4 default;
+create table rlp4_1 partition of rlp4 for values from (20) to (25);
+create table rlp4_2 partition of rlp4 for values from (25) to (29);
+
+create table rlp5 partition of rlp for values from (31) to (maxvalue) partition by range (a);
+create table rlp5_default partition of rlp5 default;
+create table rlp5_1 partition of rlp5 for values from (31) to (40);
+
+explain (costs off) select * from rlp where a < 1;
+explain (costs off) select * from rlp where 1 > a;	/* commutates */
+explain (costs off) select * from rlp where a <= 1;
+explain (costs off) select * from rlp where a = 1;
+explain (costs off) select * from rlp where a = 1::bigint;		/* same as above */
+explain (costs off) select * from rlp where a = 1::numeric;	/* no pruning */
+explain (costs off) select * from rlp where a <= 10;
+explain (costs off) select * from rlp where a > 10;
+explain (costs off) select * from rlp where a < 15;
+explain (costs off) select * from rlp where a <= 15;
+explain (costs off) select * from rlp where a > 15 and b = 'ab';
+explain (costs off) select * from rlp where a = 16;
+explain (costs off) select * from rlp where a = 16 and b in ('not', 'in', 'here');
+explain (costs off) select * from rlp where a = 16 and b < 'ab';
+explain (costs off) select * from rlp where a = 16 and b <= 'ab';
+explain (costs off) select * from rlp where a = 16 and b is null;
+explain (costs off) select * from rlp where a = 16 and b is not null;
+explain (costs off) select * from rlp where a is null;
+explain (costs off) select * from rlp where a is not null;
+explain (costs off) select * from rlp where a > 30;
+explain (costs off) select * from rlp where a = 30;	/* only default is scanned */
+explain (costs off) select * from rlp where a <= 31;
+explain (costs off) select * from rlp where a = 1 or a = 7;
+explain (costs off) select * from rlp where a = 1 or b = 'ab';
+
+explain (costs off) select * from rlp where a > 20 and a < 27;
+explain (costs off) select * from rlp where a = 29;
+explain (costs off) select * from rlp where a >= 29;
+
+-- redundant clauses are eliminated
+explain (costs off) select * from rlp where a > 1 and a = 10;	/* only default */
+explain (costs off) select * from rlp where a > 1 and a >=15;	/* rlp3 onwards, including default */
+explain (costs off) select * from rlp where a = 1 and a = 3;	/* empty */
+explain (costs off) select * from rlp where (a = 1 and a = 3) or (a > 1 and a = 15);
+
+-- multi-column keys
+create table mc3p (a int, b int, c int) partition by range (a, abs(b), c);
+create table mc3p_default partition of mc3p default;
+create table mc3p0 partition of mc3p for values from (minvalue, minvalue, minvalue) to (1, 1, 1);
+create table mc3p1 partition of mc3p for values from (1, 1, 1) to (10, 5, 10);
+create table mc3p2 partition of mc3p for values from (10, 5, 10) to (10, 10, 10);
+create table mc3p3 partition of mc3p for values from (10, 10, 10) to (10, 10, 20);
+create table mc3p4 partition of mc3p for values from (10, 10, 20) to (10, maxvalue, maxvalue);
+create table mc3p5 partition of mc3p for values from (11, 1, 1) to (20, 10, 10);
+create table mc3p6 partition of mc3p for values from (20, 10, 10) to (20, 20, 20);
+create table mc3p7 partition of mc3p for values from (20, 20, 20) to (maxvalue, maxvalue, maxvalue);
+
+explain (costs off) select * from mc3p where a = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) < 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1;
+explain (costs off) select * from mc3p where a = 1 and abs(b) = 1 and c < 8;
+explain (costs off) select * from mc3p where a = 10 and abs(b) between 5 and 35;
+explain (costs off) select * from mc3p where a > 10;
+explain (costs off) select * from mc3p where a >= 10;
+explain (costs off) select * from mc3p where a < 10;
+explain (costs off) select * from mc3p where a <= 10 and abs(b) < 10;
+explain (costs off) select * from mc3p where a = 11 and abs(b) = 0;	/* empty */
+explain (costs off) select * from mc3p where a = 20 and abs(b) = 10 and c = 100;
+explain (costs off) select * from mc3p where a > 20;
+explain (costs off) select * from mc3p where a >= 20;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1 and c = 1) or (a = 10 and abs(b) = 5 and c = 10) or (a > 11 and a < 20) or a < 1 or a = 1;
+explain (costs off) select * from mc3p where a = 1 or abs(b) = 1 or c = 1;
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 10);
+explain (costs off) select * from mc3p where (a = 1 and abs(b) = 1) or (a = 10 and abs(b) = 9);
+
+-- a simpler multi-column keys case
+create table mc2p (a int, b int) partition by range (a, b);
+create table mc2p_default partition of mc2p default;
+create table mc2p0 partition of mc2p for values from (minvalue, minvalue) to (1, minvalue);
+create table mc2p1 partition of mc2p for values from (1, minvalue) to (1, 1);
+create table mc2p2 partition of mc2p for values from (1, 1) to (1, maxvalue);
+create table mc2p3 partition of mc2p for values from (2, minvalue) to (2, 1);
+create table mc2p4 partition of mc2p for values from (2, 1) to (2, maxvalue);
+create table mc2p5 partition of mc2p for values from (2, maxvalue) to (maxvalue, maxvalue);
+
+explain (costs off) select * from mc2p where a < 2;
+explain (costs off) select * from mc2p where a = 2 and b < 1;
+explain (costs off) select * from mc2p where a > 1;
+explain (costs off) select * from mc2p where a = 1 and b > 1;
+
+-- boolean partitioning
+create table boolpart (a bool) partition by list (a);
+create table boolpart_default partition of boolpart default;
+create table boolpart_t partition of boolpart for values in ('true');
+create table boolpart_f partition of boolpart for values in ('false');
+
+explain (costs off) select * from boolpart where a in (true, false);
+explain (costs off) select * from boolpart where a = false;
+explain (costs off) select * from boolpart where not a = false;
+explain (costs off) select * from boolpart where a is true or a is not true;
+explain (costs off) select * from boolpart where a is not true;
+explain (costs off) select * from boolpart where a is not true and a is not false;
+explain (costs off) select * from boolpart where a is unknown;
+explain (costs off) select * from boolpart where a is not unknown;
+
+drop table lp, coll_pruning, rlp, mc3p, mc2p, boolpart;
-- 
2.11.0

diff --git a/src/backend/nodes/bitmapset.c b/src/backend/nodes/bitmapset.c
index 34d242b..7146ba1 100644
--- a/src/backend/nodes/bitmapset.c
+++ b/src/backend/nodes/bitmapset.c
@@ -789,7 +789,7 @@ bms_add_members(Bitmapset *a, const Bitmapset *b)
  *		Add members in the range of 'lower' to 'upper' to the set.
  *
  * Note this could also be done by calling bms_add_member in a loop, however,
- * using this function will be faster when the range is large as we work with
+ * using this function will be faster when the range is large as we're working
  * at the bitmapword level rather than at bit level.
  */
 Bitmapset *
@@ -834,7 +834,7 @@ bms_add_range(Bitmapset *a, int lower, int upper)
 	 */
 	while (wordnum <= uwordnum)
 	{
-		bitmapword mask = (bitmapword) ~0;
+		bitmapword mask = (~(bitmapword) 0);
 
 		/* If working on the lower word, zero out bits below 'lower'. */
 		if (wordnum == lwordnum)