From 64f318bbd5a43e59749657c3972c4c6eaa58af8f Mon Sep 17 00:00:00 2001
From: Tom Lane <tgl@sss.pgh.pa.us>
Date: Tue, 4 Jul 2023 14:03:40 -0400
Subject: [PATCH v2 2/3] Rewrite ArrayCount() to make dimensionality checks
 more complete.

ArrayCount has had a few layers of revision over the years,
reaching a point where it's really quite baroque and unintelligible.
There are actually four different arrays holding dimensions, which
can be reduced to two with little effort.  It was also close to
impossible to figure out why the dimension-counting actually worked
correctly.  Rewrite to make it perhaps a little less opaque.
In particular, a new element or subarray is now counted when we see
the start of the element or subarray, not at some later point where
we might not even be at the same nest_level anymore.

Also, install guards to catch non-rectangularity cases that were
previously missed, along the same lines as recent plperl and plpython
fixes: we were not checking that all scalar elements appear at the
same nesting depth.  Thanks to the very different logic, this seems
not to have caused any of the sort of internal errors that the PLs
suffered from, but it's still wrong.

On the other hand, the behavior of plperl and plpython suggests
that we should allow empty sub-arrays, as in '{{},{}}'; those
languages interpret equivalent constructs such as '[[],[]]' as
an empty (hence zero-dimensional) array, and it seems unclear
why array_in should not.  Remove the hack that was installed to
reject that case.

Also, fix things so that it's possible to write an empty array
with explicit bounds spec, along the lines of '[1:0]={}'.
Before, that was rejected with a complaint about upper bound
less than lower, but I don't see a good reason not to allow it.

In passing, remove ArrayParseState.ARRAY_ELEM_COMPLETED, a state
that the state machine never entered yet wasted cycles checking for.

(This patch looks a bit odd because I tried to avoid reindenting
existing code, despite having removed one level of loop.  That
seemed to me to make it easier to read the code changes.  A
follow-up pgindent run is needed.)
---
 src/backend/utils/adt/arrayfuncs.c   | 241 +++++++++++++++++----------
 src/test/regress/expected/arrays.out |  68 +++++++-
 src/test/regress/sql/arrays.sql      |  13 +-
 3 files changed, 230 insertions(+), 92 deletions(-)

diff --git a/src/backend/utils/adt/arrayfuncs.c b/src/backend/utils/adt/arrayfuncs.c
index 050568808a..4fb052b23e 100644
--- a/src/backend/utils/adt/arrayfuncs.c
+++ b/src/backend/utils/adt/arrayfuncs.c
@@ -57,7 +57,6 @@ typedef enum
 	ARRAY_NO_LEVEL,
 	ARRAY_LEVEL_STARTED,
 	ARRAY_ELEM_STARTED,
-	ARRAY_ELEM_COMPLETED,
 	ARRAY_QUOTED_ELEM_STARTED,
 	ARRAY_QUOTED_ELEM_COMPLETED,
 	ARRAY_ELEM_DELIMITED,
@@ -302,10 +301,18 @@ array_in(PG_FUNCTION_ARGS)
 		*q = '\0';
 		ub = atoi(p);
 		p = q + 1;
-		if (ub < lBound[ndim])
+
+		/* Upper bound of INT_MAX is disallowed, cf ArrayCheckBounds() */
+		if (ub == INT_MAX)
+			ereturn(escontext, (Datum) 0,
+					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+					 errmsg("array upper bound is too large: %d",
+							ub)));
+		/* Now it's safe to compute ub + 1 */
+		if (ub + 1 < lBound[ndim])
 			ereturn(escontext, (Datum) 0,
 					(errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
-					 errmsg("upper bound cannot be less than lower bound")));
+					 errmsg("upper bound cannot be less than lower bound minus one")));
 
 		dim[ndim] = ub - lBound[ndim] + 1;
 		ndim++;
@@ -461,7 +468,9 @@ array_isspace(char ch)
  *	 syntax-checking the array structure decoration (braces and delimiters).
  *
  * Returns number of dimensions as function result.  The axis lengths are
- * returned in dim[], which must be of size MAXDIM.
+ * returned in dim[], which must be of size MAXDIM.  This function does not
+ * special-case empty arrays: it will return a positive result with some
+ * zero axis length(s).
  *
  * If we detect an error, fill *escontext with error details and return -1
  * (unless escontext isn't provided, in which case errors will be thrown).
@@ -472,31 +481,22 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 	int			nest_level = 0,
 				i;
 	int			ndim = 1,
-				temp[MAXDIM],
-				nelems[MAXDIM],
-				nelems_last[MAXDIM];
+				nelems[MAXDIM];
+	bool		ndim_frozen = false;
 	bool		in_quotes = false;
 	bool		eoArray = false;
-	bool		empty_array = true;
 	const char *ptr;
 	ArrayParseState parse_state = ARRAY_NO_LEVEL;
 
+	/* Initialize dim[] entries to -1 meaning "unknown" */
 	for (i = 0; i < MAXDIM; ++i)
-	{
-		temp[i] = dim[i] = nelems_last[i] = 0;
-		nelems[i] = 1;
-	}
+		dim[i] = -1;
 
+	/* Scan string until we reach closing brace */
 	ptr = str;
 	while (!eoArray)
 	{
-		bool		itemdone = false;
-
-		while (!itemdone)
-		{
-			if (parse_state == ARRAY_ELEM_STARTED ||
-				parse_state == ARRAY_QUOTED_ELEM_STARTED)
-				empty_array = false;
+		bool		new_element = false;
 
 			switch (*ptr)
 			{
@@ -513,17 +513,25 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 					 * start, or after an element delimiter. In any case we
 					 * now must be past an element start.
 					 */
-					if (parse_state != ARRAY_LEVEL_STARTED &&
-						parse_state != ARRAY_ELEM_STARTED &&
-						parse_state != ARRAY_QUOTED_ELEM_STARTED &&
-						parse_state != ARRAY_ELEM_DELIMITED)
+				switch (parse_state)
+				{
+					case ARRAY_LEVEL_STARTED:
+					case ARRAY_ELEM_DELIMITED:
+						/* start new unquoted element */
+						parse_state = ARRAY_ELEM_STARTED;
+						new_element = true;
+						break;
+					case ARRAY_ELEM_STARTED:
+					case ARRAY_QUOTED_ELEM_STARTED:
+						/* already in element */
+						break;
+					default:
 						ereturn(escontext, -1,
 								(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 								 errmsg("malformed array literal: \"%s\"", str),
 								 errdetail("Unexpected \"%c\" character.",
 										   '\\')));
-					if (parse_state != ARRAY_QUOTED_ELEM_STARTED)
-						parse_state = ARRAY_ELEM_STARTED;
+				}
 					/* skip the escaped character */
 					if (*(ptr + 1))
 						ptr++;
@@ -540,18 +548,28 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 					 * element start, or after an element delimiter. In any
 					 * case we now must be past an element start.
 					 */
-					if (parse_state != ARRAY_LEVEL_STARTED &&
-						parse_state != ARRAY_QUOTED_ELEM_STARTED &&
-						parse_state != ARRAY_ELEM_DELIMITED)
+				switch (parse_state)
+				{
+					case ARRAY_LEVEL_STARTED:
+					case ARRAY_ELEM_DELIMITED:
+						/* start new quoted element */
+						Assert(!in_quotes);
+						in_quotes = true;
+						parse_state = ARRAY_QUOTED_ELEM_STARTED;
+						new_element = true;
+						break;
+					case ARRAY_QUOTED_ELEM_STARTED:
+						/* already in element, end it */
+						Assert(in_quotes);
+						in_quotes = false;
+						parse_state = ARRAY_QUOTED_ELEM_COMPLETED;
+						break;
+					default:
 						ereturn(escontext, -1,
 								(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 								 errmsg("malformed array literal: \"%s\"", str),
 								 errdetail("Unexpected array element.")));
-					in_quotes = !in_quotes;
-					if (in_quotes)
-						parse_state = ARRAY_QUOTED_ELEM_STARTED;
-					else
-						parse_state = ARRAY_QUOTED_ELEM_COMPLETED;
+				}
 					break;
 				case '{':
 					if (!in_quotes)
@@ -559,27 +577,52 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 						/*
 						 * A left brace can occur if no nesting has occurred
 						 * yet, after a level start, or after a level
-						 * delimiter.
+						 * delimiter.  If we see one after an element
+						 * delimiter, we have something like "{1,{2}}", so
+						 * complain about non-rectangularity.
 						 */
-						if (parse_state != ARRAY_NO_LEVEL &&
-							parse_state != ARRAY_LEVEL_STARTED &&
-							parse_state != ARRAY_LEVEL_DELIMITED)
+					switch (parse_state)
+					{
+						case ARRAY_NO_LEVEL:
+						case ARRAY_LEVEL_STARTED:
+						case ARRAY_LEVEL_DELIMITED:
+							/* okay */
+							break;
+						case ARRAY_ELEM_DELIMITED:
+							ereturn(escontext, -1,
+									(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+									 errmsg("malformed array literal: \"%s\"", str),
+									 errdetail("Multidimensional arrays must have sub-arrays with matching dimensions.")));
+						default:
 							ereturn(escontext, -1,
 									(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 									 errmsg("malformed array literal: \"%s\"", str),
 									 errdetail("Unexpected \"%c\" character.",
 											   '{')));
+					}
 						parse_state = ARRAY_LEVEL_STARTED;
+					/* Nested sub-arrays count as elements of outer level */
+					if (nest_level > 0)
+						nelems[nest_level - 1]++;
+					/* Initialize element counting in the new level */
 						if (nest_level >= MAXDIM)
 							ereturn(escontext, -1,
 									(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
 									 errmsg("number of array dimensions (%d) exceeds the maximum allowed (%d)",
 											nest_level + 1, MAXDIM)));
-						temp[nest_level] = 0;
+					nelems[nest_level] = 0;
 						nest_level++;
 						if (ndim < nest_level)
+					{
+						/* Can't increase ndim once it's frozen */
+						if (ndim_frozen)
+							ereturn(escontext, -1,
+									(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+									 errmsg("malformed array literal: \"%s\"", str),
+									 errdetail("Multidimensional arrays must have sub-arrays with matching dimensions.")));
 							ndim = nest_level;
 					}
+				}
 					break;
 				case '}':
 					if (!in_quotes)
@@ -587,46 +630,56 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 						/*
 						 * A right brace can occur after an element start, an
 						 * element completion, a quoted element completion, or
-						 * a level completion.
+						 * a level completion.  We also allow it after a level
+						 * start, that is an empty sub-array "{}" --- but that
+						 * freezes the number of dimensions and all such
+						 * sub-arrays must be at the same level, just like
+						 * sub-arrays containing elements.
 						 */
-						if (parse_state != ARRAY_ELEM_STARTED &&
-							parse_state != ARRAY_ELEM_COMPLETED &&
-							parse_state != ARRAY_QUOTED_ELEM_COMPLETED &&
-							parse_state != ARRAY_LEVEL_COMPLETED &&
-							!(nest_level == 1 && parse_state == ARRAY_LEVEL_STARTED))
+					switch (parse_state)
+					{
+						case ARRAY_ELEM_STARTED:
+						case ARRAY_QUOTED_ELEM_COMPLETED:
+						case ARRAY_LEVEL_COMPLETED:
+							/* okay */
+							break;
+						case ARRAY_LEVEL_STARTED:
+							/* empty sub-array: OK if at correct nest_level */
+							ndim_frozen = true;
+							if (nest_level != ndim)
 							ereturn(escontext, -1,
 									(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 									 errmsg("malformed array literal: \"%s\"", str),
-									 errdetail("Unexpected \"%c\" character.",
-											   '}')));
-						parse_state = ARRAY_LEVEL_COMPLETED;
-						if (nest_level == 0)
+										 errdetail("Multidimensional arrays must have sub-arrays with matching dimensions.")));
+							break;
+						default:
 							ereturn(escontext, -1,
 									(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 									 errmsg("malformed array literal: \"%s\"", str),
-									 errdetail("Unmatched \"%c\" character.", '}')));
+									 errdetail("Unexpected \"%c\" character.",
+											   '}')));
+					}
+					parse_state = ARRAY_LEVEL_COMPLETED;
+					/* The parse state check assured we're in a level. */
+					Assert(nest_level > 0);
 						nest_level--;
 
-						if (nelems_last[nest_level] != 0 &&
-							nelems[nest_level] != nelems_last[nest_level])
+					if (dim[nest_level] < 0)
+					{
+						/* Save length of first sub-array of this level */
+						dim[nest_level] = nelems[nest_level];
+					}
+					else if (nelems[nest_level] != dim[nest_level])
+					{
+						/* Subsequent sub-arrays must have same length */
 							ereturn(escontext, -1,
 									(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 									 errmsg("malformed array literal: \"%s\"", str),
-									 errdetail("Multidimensional arrays must have "
-											   "sub-arrays with matching "
-											   "dimensions.")));
-						nelems_last[nest_level] = nelems[nest_level];
-						nelems[nest_level] = 1;
-						if (nest_level == 0)
-							eoArray = itemdone = true;
-						else
-						{
-							/*
-							 * We don't set itemdone here; see comments in
-							 * ReadArrayStr
-							 */
-							temp[nest_level - 1]++;
+								 errdetail("Multidimensional arrays must have sub-arrays with matching dimensions.")));
 						}
+					/* Done if this is the outermost level's '}' */
+					if (nest_level == 0)
+						eoArray = true;
 					}
 					break;
 				default:
@@ -635,12 +688,11 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 						if (*ptr == typdelim)
 						{
 							/*
-							 * Delimiters can occur after an element start, an
-							 * element completion, a quoted element
-							 * completion, or a level completion.
+							 * Delimiters can occur after an element start, a
+							 * quoted element completion, or a level
+							 * completion.
 							 */
 							if (parse_state != ARRAY_ELEM_STARTED &&
-								parse_state != ARRAY_ELEM_COMPLETED &&
 								parse_state != ARRAY_QUOTED_ELEM_COMPLETED &&
 								parse_state != ARRAY_LEVEL_COMPLETED)
 								ereturn(escontext, -1,
@@ -652,8 +704,6 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 								parse_state = ARRAY_LEVEL_DELIMITED;
 							else
 								parse_state = ARRAY_ELEM_DELIMITED;
-							itemdone = true;
-							nelems[nest_level - 1]++;
 						}
 						else if (!array_isspace(*ptr))
 						{
@@ -662,23 +712,51 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 							 * level start, after an element start, or after
 							 * an element delimiter. In any case we now must
 							 * be past an element start.
+							 *
+							 * If it's a space character, we can ignore it; it
+							 * might be data or not, but it doesn't change the
+							 * parsing state.
 							 */
-							if (parse_state != ARRAY_LEVEL_STARTED &&
-								parse_state != ARRAY_ELEM_STARTED &&
-								parse_state != ARRAY_ELEM_DELIMITED)
+						switch (parse_state)
+						{
+							case ARRAY_LEVEL_STARTED:
+							case ARRAY_ELEM_DELIMITED:
+								/* start new unquoted element */
+								parse_state = ARRAY_ELEM_STARTED;
+								new_element = true;
+								break;
+							case ARRAY_ELEM_STARTED:
+								/* already in element */
+								break;
+							default:
 								ereturn(escontext, -1,
 										(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 										 errmsg("malformed array literal: \"%s\"", str),
 										 errdetail("Unexpected array element.")));
-							parse_state = ARRAY_ELEM_STARTED;
+						}
 						}
 					}
 					break;
 			}
-			if (!itemdone)
-				ptr++;
+
+		/* To reduce duplication, all new-element cases go through here. */
+		if (new_element)
+		{
+			/*
+			 * Once we have found an element, the number of dimensions can no
+			 * longer increase, and subsequent elements must all be at the
+			 * same nesting depth.
+			 */
+			ndim_frozen = true;
+			if (nest_level != ndim)
+				ereturn(escontext, -1,
+						(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+						 errmsg("malformed array literal: \"%s\"", str),
+						 errdetail("Multidimensional arrays must have sub-arrays with matching dimensions.")));
+			/* Count the new element */
+			nelems[nest_level - 1]++;
 		}
-		temp[ndim - 1]++;
+
 		ptr++;
 	}
 
@@ -692,13 +770,6 @@ ArrayCount(const char *str, int *dim, char typdelim, Node *escontext)
 					 errdetail("Junk after closing right brace.")));
 	}
 
-	/* special case for an empty array */
-	if (empty_array)
-		return 0;
-
-	for (i = 0; i < ndim; ++i)
-		dim[i] = temp[i];
-
 	return ndim;
 }
 
diff --git a/src/test/regress/expected/arrays.out b/src/test/regress/expected/arrays.out
index 7064391468..26d4281d94 100644
--- a/src/test/regress/expected/arrays.out
+++ b/src/test/regress/expected/arrays.out
@@ -1475,12 +1475,7 @@ select '{{1,{2}},{2,3}}'::text[];
 ERROR:  malformed array literal: "{{1,{2}},{2,3}}"
 LINE 1: select '{{1,{2}},{2,3}}'::text[];
                ^
-DETAIL:  Unexpected "{" character.
-select '{{},{}}'::text[];
-ERROR:  malformed array literal: "{{},{}}"
-LINE 1: select '{{},{}}'::text[];
-               ^
-DETAIL:  Unexpected "}" character.
+DETAIL:  Multidimensional arrays must have sub-arrays with matching dimensions.
 select E'{{1,2},\\{2,3}}'::text[];
 ERROR:  malformed array literal: "{{1,2},\{2,3}}"
 LINE 1: select E'{{1,2},\\{2,3}}'::text[];
@@ -1501,6 +1496,49 @@ ERROR:  malformed array literal: "{ }}"
 LINE 1: select '{ }}'::text[];
                ^
 DETAIL:  Junk after closing right brace.
+select '{{1},{{2}}}'::text[];
+ERROR:  malformed array literal: "{{1},{{2}}}"
+LINE 1: select '{{1},{{2}}}'::text[];
+               ^
+DETAIL:  Multidimensional arrays must have sub-arrays with matching dimensions.
+select '{{{1}},{2}}'::text[];
+ERROR:  malformed array literal: "{{{1}},{2}}"
+LINE 1: select '{{{1}},{2}}'::text[];
+               ^
+DETAIL:  Multidimensional arrays must have sub-arrays with matching dimensions.
+select '{{},{{}}}'::text[];
+ERROR:  malformed array literal: "{{},{{}}}"
+LINE 1: select '{{},{{}}}'::text[];
+               ^
+DETAIL:  Multidimensional arrays must have sub-arrays with matching dimensions.
+select '{{{}},{}}'::text[];
+ERROR:  malformed array literal: "{{{}},{}}"
+LINE 1: select '{{{}},{}}'::text[];
+               ^
+DETAIL:  Multidimensional arrays must have sub-arrays with matching dimensions.
+select '{{1},{}}'::text[];
+ERROR:  malformed array literal: "{{1},{}}"
+LINE 1: select '{{1},{}}'::text[];
+               ^
+DETAIL:  Multidimensional arrays must have sub-arrays with matching dimensions.
+select '{{},{1}}'::text[];
+ERROR:  malformed array literal: "{{},{1}}"
+LINE 1: select '{{},{1}}'::text[];
+               ^
+DETAIL:  Multidimensional arrays must have sub-arrays with matching dimensions.
+select '[2147483646:2147483647]={1,2}'::int[];
+ERROR:  array upper bound is too large: 2147483647
+LINE 1: select '[2147483646:2147483647]={1,2}'::int[];
+               ^
+select '[1:-1]={}'::int[];
+ERROR:  upper bound cannot be less than lower bound minus one
+LINE 1: select '[1:-1]={}'::int[];
+               ^
+select '[1:0]={1}'::int[];
+ERROR:  malformed array literal: "[1:0]={1}"
+LINE 1: select '[1:0]={1}'::int[];
+               ^
+DETAIL:  Specified array dimensions do not match array contents.
 select array[];
 ERROR:  cannot determine type of empty array
 LINE 1: select array[];
@@ -1514,6 +1552,12 @@ select '{}'::text[];
  {}
 (1 row)
 
+select '{{},{}}'::text[];
+ text 
+------
+ {}
+(1 row)
+
 select '{{{1,2,3,4},{2,3,4,5}},{{3,4,5,6},{4,5,6,7}}}'::text[];
                      text                      
 -----------------------------------------------
@@ -1559,6 +1603,18 @@ select '[0:1]={1.1,2.2}'::float8[];
  [0:1]={1.1,2.2}
 (1 row)
 
+select '[2147483646:2147483646]={1}'::int[];
+            int4             
+-----------------------------
+ [2147483646:2147483646]={1}
+(1 row)
+
+select '[1:0]={}'::int[];
+ int4 
+------
+ {}
+(1 row)
+
 -- all of the above should be accepted
 -- tests for array aggregates
 CREATE TEMP TABLE arraggtest ( f1 INT[], f2 TEXT[][], f3 FLOAT[]);
diff --git a/src/test/regress/sql/arrays.sql b/src/test/regress/sql/arrays.sql
index f1375621e0..b7e2b180cc 100644
--- a/src/test/regress/sql/arrays.sql
+++ b/src/test/regress/sql/arrays.sql
@@ -454,16 +454,25 @@ select 'foo' ilike all (array['F%', '%O']); -- t
 
 -- none of the following should be accepted
 select '{{1,{2}},{2,3}}'::text[];
-select '{{},{}}'::text[];
 select E'{{1,2},\\{2,3}}'::text[];
 select '{{"1 2" x},{3}}'::text[];
 select '{}}'::text[];
 select '{ }}'::text[];
+select '{{1},{{2}}}'::text[];
+select '{{{1}},{2}}'::text[];
+select '{{},{{}}}'::text[];
+select '{{{}},{}}'::text[];
+select '{{1},{}}'::text[];
+select '{{},{1}}'::text[];
+select '[2147483646:2147483647]={1,2}'::int[];
+select '[1:-1]={}'::int[];
+select '[1:0]={1}'::int[];
 select array[];
 -- none of the above should be accepted
 
 -- all of the following should be accepted
 select '{}'::text[];
+select '{{},{}}'::text[];
 select '{{{1,2,3,4},{2,3,4,5}},{{3,4,5,6},{4,5,6,7}}}'::text[];
 select '{0 second  ,0 second}'::interval[];
 select '{ { "," } , { 3 } }'::text[];
@@ -474,6 +483,8 @@ select '{
          }'::interval[];
 select array[]::text[];
 select '[0:1]={1.1,2.2}'::float8[];
+select '[2147483646:2147483646]={1}'::int[];
+select '[1:0]={}'::int[];
 -- all of the above should be accepted
 
 -- tests for array aggregates
-- 
2.39.3

