Fix SPLIT PARTITION bound-overlap bug and other improvements
Hi,
While testing ALTER TABLE ... SPLIT PARTITION, I found a bug and a few behaviors and messages that seem worth improving.
0. A bound-overlap bug
I numbered this item as 0 because I found it after finishing items 1, 2, and 3. While doing a final verification before sending this email, I was surprised to find that the partitioned table ended up with two overlapping partitions.
Here is a simple repro:
```
evantest=# drop table t;
DROP TABLE
evantest=# CREATE TABLE t (i int) PARTITION BY RANGE(i);
CREATE TABLE
evantest=# CREATE TABLE p0a PARTITION OF t FOR VALUES FROM (0) TO (51);
CREATE TABLE
evantest=# CREATE TABLE p0b PARTITION OF t FOR VALUES FROM (51) TO (100);
CREATE TABLE
evantest=# \d+ t;
Partitioned table "public.t"
Column | Type | Collation | Nullable | Default | Storage | Compression | Stats target | Description
--------+---------+-----------+----------+---------+---------+-------------+--------------+-------------
i | integer | | | | plain | | |
Partition key: RANGE (i)
Partitions:
p0a FOR VALUES FROM (0) TO (51)
p0b FOR VALUES FROM (51) TO (100)
evantest=# ALTER TABLE t SPLIT PARTITION p0a INTO
evantest-# (PARTITION p0a FOR VALUES FROM (0) TO (53),
evantest(# PARTITION pdef DEFAULT);
ALTER TABLE
evantest=#
evantest=#
evantest=# \d+ t;
Partitioned table "public.t"
Column | Type | Collation | Nullable | Default | Storage | Compression | Stats target | Description
--------+---------+-----------+----------+---------+---------+-------------+--------------+-------------
i | integer | | | | plain | | |
Partition key: RANGE (i)
Partitions:
p0a FOR VALUES FROM (0) TO (53)
p0b FOR VALUES FROM (51) TO (100)
pdef DEFAULT
```
As shown above, p0a and p0b now overlap. I think this is a real bug.
The problem seems to be in check_partition_bounds_for_split_range(). If the only non-default replacement partition is both first and last, the function checks only the lower bound because it uses if (first) ... else .... It never checks the upper bound in that case.
1. The documentation about splitting with a DEFAULT partition is a bit unclear
Since SPLIT PARTITION allows one of the new partitions to be specified as DEFAULT, I wondered whether that new DEFAULT partition means the remaining part of the split partition's bound, or the partitioned table's global DEFAULT partition.
The current documentation says:
```
<para>
This form splits a single partition of the target table into new
partitions. Hash-partitioned target table is not supported.
Only a simple, non-partitioned partition can be split.
If the split partition is the <literal>DEFAULT</literal> partition,
one of the new partitions must be <literal>DEFAULT</literal>.
If the partitioned table does not have a <literal>DEFAULT</literal>
partition, a <literal>DEFAULT</literal> partition can be defined as one
of the new partitions.
</para>
<para>
The bounds of new partitions should not overlap with those of new or
existing partitions (except <replaceable class="parameter">partition_name</replaceable>).
The combined bounds of new partitions <literal>
<replaceable class="parameter">partition_name1</replaceable>,
<replaceable class="parameter">partition_name2</replaceable>[, ...]
</literal> should be equal to the bounds of the split partition
<replaceable class="parameter">partition_name</replaceable>.
One of the new partitions can have the same name as the split partition
<replaceable class="parameter">partition_name</replaceable>
(this is suitable in case of splitting the <literal>DEFAULT</literal>
partition: after the split, the <literal>DEFAULT</literal> partition
remains with the same name, but its partition bound changes).
</para>
```
From the first paragraph, it seems that the new DEFAULT partition is a table-level default partition. However, the second paragraph says that the combined bounds of the new partitions should be equal to the bounds of the split partition, which can make it look as if the new DEFAULT partition only covers the remaining part of the split partition's bound.
My tests show that the new DEFAULT partition is the partitioned table's global DEFAULT partition. So I think the documentation can be improved to make that clearer.
2. I found this hint message confusing:
```
evantest=# ALTER TABLE t SPLIT PARTITION p0a INTO (PARTITION p0a1 FOR VALUES FROM (0) TO (5), PARTITION p0a2 FOR VALUES FROM (6) to (51), PARTITION pdef DEFAULT);
ERROR: upper bound of partition "p0a2" is greater than upper bound of split partition "p0a"
LINE 1: ...0) TO (5), PARTITION p0a2 FOR VALUES FROM (6) to (51), PARTI...
^
HINT: ALTER TABLE ... SPLIT PARTITION require combined bounds of new partitions must exactly match the bound of the split partition.
```
In this command, one of the new explicit partition bounds exceeds the original partition bound, but the command also specifies a DEFAULT partition. In this case, the combined explicit bounds do not need to exactly match the original partition bound, they only need to stay within it. So the hint is not quite accurate for this case.
3. SPLIT PARTITION currently provides another way to add a DEFAULT partition:
```
evantest=# \d+ t;
Partitioned table "public.t"
Column | Type | Collation | Nullable | Default | Storage | Compression | Stats target | Description
--------+---------+-----------+----------+---------+---------+-------------+--------------+-------------
i | integer | | | | plain | | |
Partition key: RANGE (i)
Partitions:
p0a FOR VALUES FROM (0) TO (50)
p0b FOR VALUES FROM (51) TO (100)
p200 FOR VALUES FROM (200) TO (300)
evantest=# ALTER TABLE t SPLIT PARTITION p0a INTO (PARTITION p0a FOR VALUES FROM (0) TO (50), PARTITION pdef DEFAULT);
ALTER TABLE
```
Here, the new p0a partition has the same bound as the original p0a partition, so no real split happens. The command effectively only adds a new DEFAULT partition. However, it still goes through the full split-partition path: creating a new partition, moving data, attaching the new partition, and dropping the old partition.
Initially, I considered adding a fast path for this case so that it would only add the new DEFAULT partition. But after thinking about it more, I think it is better to reject this degenerate form instead. We already has direct ways to add a DEFAULT partition:
```
CREATE TABLE p PARTITION OF t DEFAULT;
or
CREATE TABLE p;
ALTER TABLE t ATTACH PARTITION p DEFAULT;
```
So I do not think SPLIT PARTITION needs to become another syntax for adding a DEFAULT partition when no actual split is performed. Accepting this form could also raise another question later: if this is allowed, why does the user have to repeat the original bound at all? Why not allow something like this?
```
ALTER TABLE t SPLIT PARTITION p0a INTO (PARTITION pdef DEFAULT);
```
That seems like an awkward direction.
The attached patch tries to address all these issues.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
Attachments:
v1-0001-Fix-SPLIT-PARTITION-validation-with-DEFAULT.patchapplication/octet-stream; name=v1-0001-Fix-SPLIT-PARTITION-validation-with-DEFAULT.patch; x-unix-mode=0644Download+171-19
On Wed, 13 May 2026 at 09:39, Chao Li <li.evan.chao@gmail.com> wrote:
Hi,
While testing ALTER TABLE ... SPLIT PARTITION, I found a bug and a few behaviors and messages that seem worth improving.
0. A bound-overlap bug
I numbered this item as 0 because I found it after finishing items 1, 2, and 3. While doing a final verification before sending this email, I was surprised to find that the partitioned table ended up with two overlapping partitions.
Here is a simple repro:
```
evantest=# drop table t;
DROP TABLE
evantest=# CREATE TABLE t (i int) PARTITION BY RANGE(i);
CREATE TABLE
evantest=# CREATE TABLE p0a PARTITION OF t FOR VALUES FROM (0) TO (51);
CREATE TABLE
evantest=# CREATE TABLE p0b PARTITION OF t FOR VALUES FROM (51) TO (100);
CREATE TABLE
evantest=# \d+ t;
Partitioned table "public.t"
Column | Type | Collation | Nullable | Default | Storage | Compression | Stats target | Description
--------+---------+-----------+----------+---------+---------+-------------+--------------+-------------
i | integer | | | | plain | | |
Partition key: RANGE (i)
Partitions:
p0a FOR VALUES FROM (0) TO (51)
p0b FOR VALUES FROM (51) TO (100)evantest=# ALTER TABLE t SPLIT PARTITION p0a INTO
evantest-# (PARTITION p0a FOR VALUES FROM (0) TO (53),
evantest(# PARTITION pdef DEFAULT);
ALTER TABLE
evantest=#
evantest=#
evantest=# \d+ t;
Partitioned table "public.t"
Column | Type | Collation | Nullable | Default | Storage | Compression | Stats target | Description
--------+---------+-----------+----------+---------+---------+-------------+--------------+-------------
i | integer | | | | plain | | |
Partition key: RANGE (i)
Partitions:
p0a FOR VALUES FROM (0) TO (53)
p0b FOR VALUES FROM (51) TO (100)
pdef DEFAULT
```As shown above, p0a and p0b now overlap. I think this is a real bug.
that's 100% real issue because of planner partition pruning
```
reshke=# CREATE TABLE t (i int) PARTITION BY RANGE(i);
CREATE TABLE
reshke=# CREATE TABLE p0a PARTITION OF t FOR VALUES FROM (0) TO (51);
CREATE TABLE
reshke=# insert into t values (50);
INSERT 0 1
reshke=# CREATE TABLE p0b PARTITION OF t FOR VALUES FROM (51) TO (100);
CREATE TABLE
reshke=# insert into t values (51);
INSERT 0 1
reshke=# insert into t values (51);
INSERT 0 1
reshke=# insert into t values (51);
INSERT 0 1
reshke=# ALTER TABLE t SPLIT PARTITION p0a INTO(PARTITION p0a FOR
VALUES FROM (0) TO (53),PARTITION pdef DEFAULT);
ALTER TABLE
reshke=# table t;
i
----
50
51
51
51
(4 rows)
reshke=# select * from t where i = 51;
i
---
(0 rows)
reshke=# set enable_partition_pruning to off;
SET
reshke=# select * from t where i = 51;
i
----
51
51
51
(3 rows)
reshke=#
``
--
Best regards,
Kirill Reshke
Kirill Reshke <reshkekirill@gmail.com> 于2026年5月13日周三 13:08写道：
On Wed, 13 May 2026 at 09:39, Chao Li <li.evan.chao@gmail.com> wrote:
Hi,
While testing ALTER TABLE ... SPLIT PARTITION, I found a bug and a few
behaviors and messages that seem worth improving.
0. A bound-overlap bug
I numbered this item as 0 because I found it after finishing items 1, 2,
and 3. While doing a final verification before sending this email, I was
surprised to find that the partitioned table ended up with two overlapping
partitions.Here is a simple repro:
```
evantest=# drop table t;
DROP TABLE
evantest=# CREATE TABLE t (i int) PARTITION BY RANGE(i);
CREATE TABLE
evantest=# CREATE TABLE p0a PARTITION OF t FOR VALUES FROM (0) TO (51);
CREATE TABLE
evantest=# CREATE TABLE p0b PARTITION OF t FOR VALUES FROM (51) TO (100);
CREATE TABLE
evantest=# \d+ t;
Partitioned table "public.t"
Column | Type | Collation | Nullable | Default | Storage |Compression | Stats target | Description
--------+---------+-----------+----------+---------+---------+-------------+--------------+-------------
i | integer | | | | plain |
| |
Partition key: RANGE (i)
Partitions:
p0a FOR VALUES FROM (0) TO (51)
p0b FOR VALUES FROM (51) TO (100)evantest=# ALTER TABLE t SPLIT PARTITION p0a INTO
evantest-# (PARTITION p0a FOR VALUES FROM (0) TO (53),
evantest(# PARTITION pdef DEFAULT);
ALTER TABLE
evantest=#
evantest=#
evantest=# \d+ t;
Partitioned table "public.t"
Column | Type | Collation | Nullable | Default | Storage |Compression | Stats target | Description
--------+---------+-----------+----------+---------+---------+-------------+--------------+-------------
i | integer | | | | plain |
| |
Partition key: RANGE (i)
Partitions:
p0a FOR VALUES FROM (0) TO (53)
p0b FOR VALUES FROM (51) TO (100)
pdef DEFAULT
```As shown above, p0a and p0b now overlap. I think this is a real bug.
that's 100% real issue because of planner partition pruning
```
reshke=# CREATE TABLE t (i int) PARTITION BY RANGE(i);
CREATE TABLE
reshke=# CREATE TABLE p0a PARTITION OF t FOR VALUES FROM (0) TO (51);
CREATE TABLE
reshke=# insert into t values (50);
INSERT 0 1
reshke=# CREATE TABLE p0b PARTITION OF t FOR VALUES FROM (51) TO (100);
CREATE TABLE
reshke=# insert into t values (51);
INSERT 0 1
reshke=# insert into t values (51);
INSERT 0 1
reshke=# insert into t values (51);
INSERT 0 1
reshke=# ALTER TABLE t SPLIT PARTITION p0a INTO(PARTITION p0a FOR
VALUES FROM (0) TO (53),PARTITION pdef DEFAULT);
ALTER TABLE
reshke=# table t;
i
----
50
51
51
51
(4 rows)
reshke=# select * from t where i = 51;
i
---
(0 rows)reshke=# set enable_partition_pruning to off;
SET
reshke=# select * from t where i = 51;
i
----
51
51
51
(3 rows)reshke=#
``--
Best regards,
Kirill ReshkeI agree this is a bug. I applied the patch locally and confirmed the bug
is fixed with the patch.
Overall, the patch looks good to me.
Regards,
Zhenwei Shang
Hi, Chao Li!
Thank you for the bug report, test script, and fix!
0. A bound-overlap bug
I think this fix should be applied without much discussion:
------------------------------------------------------------------------
diff --git a/src/backend/partitioning/partbounds.c
b/src/backend/partitioning/partbounds.c
index 9b4277a4987..8b8f90569fe 100644
--- a/src/backend/partitioning/partbounds.c
+++ b/src/backend/partitioning/partbounds.c
@@ -5419,7 +5419,8 @@ check_partition_bounds_for_split_range(Relation
parent,
"ALTER
TABLE ... SPLIT PARTITION"),parser_errposition(pstate, exprLocation((Node *) datum)));
}
- else
+
+ if (last)
{
PartitionRangeBound *split_upper;
------------------------------------------------------------------------1. The documentation about splitting with a DEFAULT partition is a
bit unclear
...
2. I found this hint message confusing:
...
Unfortunately, I cannot comment on these points; it would be good to get
the opinion of people who know English well.
3. SPLIT PARTITION currently provides another way to add a DEFAULT
partition:
...
Agreed, this is another way to add a DEFAULT partition. But I'm not sure
that this way should be disabled (using the special function
check_split_partition_not_same_bound)...
Maybe it's better to keep it "as is"?
--
With best regards,
Dmitry Koval
Postgres Professional: http://postgrespro.com
On May 14, 2026, at 04:47, Dmitry Koval <d.koval@postgrespro.ru> wrote:
Hi, Chao Li!
Thank you for the bug report, test script, and fix!
0. A bound-overlap bug
I think this fix should be applied without much discussion: ------------------------------------------------------------------------ diff --git a/src/backend/partitioning/partbounds.c b/src/backend/partitioning/partbounds.c index 9b4277a4987..8b8f90569fe 100644 --- a/src/backend/partitioning/partbounds.c +++ b/src/backend/partitioning/partbounds.c @@ -5419,7 +5419,8 @@ check_partition_bounds_for_split_range(Relation parent, "ALTER TABLE ... SPLIT PARTITION"), parser_errposition(pstate, exprLocation((Node *) datum))); } - else + + if (last) { PartitionRangeBound *split_upper; ------------------------------------------------------------------------
Thanks for your confirmation.
1. The documentation about splitting with a DEFAULT partition is a bit unclear
...
2. I found this hint message confusing:
...Unfortunately, I cannot comment on these points; it would be good to get the opinion of people who know English well.
I want to add one more point about these two changes.
There is a code comment saying that when a DEFAULT partition is specified, the new partition's lower bound may be greater than the original lower bound:
```
/*
* The lower bound of "spec" must equal the lower bound of the
* split partition. However, if one of the new partitions is
* DEFAULT, then it is ok for the new partition's lower bound to
* be greater than that of the split partition.
*/
```
This also indicates that the original hint message mentioning “exactly match" is wrong for the DEFAULT case.
3. SPLIT PARTITION currently provides another way to add a DEFAULT partition:
...Agreed, this is another way to add a DEFAULT partition. But I'm not sure that this way should be disabled (using the special function check_split_partition_not_same_bound)...
Maybe it's better to keep it "as is"?
Yeah, this may be worth more discussion. But I think we should either reject this usage or add a fast path to avoid unnecessary creation of a new partition, data movement, etc. Otherwise, it feels more like a misuse of SPLIT PARTITION rather than a useful new alternative.
To make this patch easier to process, I split it into 4 commits:
0001 - Fixes the bound-overlap bug
0002 - Fix the incorrect HINT message for the DEFAULT case
0003 - Fix the incorrect description about combined bound in the SGML doc
0004 - Reject only-create-default-partition usage
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
Attachments:
v2-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patchapplication/octet-stream; name=v2-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch; x-unix-mode=0644Download+33-2
v2-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patchapplication/octet-stream; name=v2-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch; x-unix-mode=0644Download+3-6
v2-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patchapplication/octet-stream; name=v2-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch; x-unix-mode=0644Download+14-10
v2-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchapplication/octet-stream; name=v2-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch; x-unix-mode=0644Download+103-1
Hi, Chao!
On Thu, May 14, 2026 at 9:59 AM Chao Li <li.evan.chao@gmail.com> wrote:
To make this patch easier to process, I split it into 4 commits:
0001 - Fixes the bound-overlap bug
0002 - Fix the incorrect HINT message for the DEFAULT case
0003 - Fix the incorrect description about combined bound in the SGML doc
0004 - Reject only-create-default-partition usage
Thank you for your work. I've revised the patchset.
0002 - I've also fixed gramma of hints in other branches
0004 - In the check_split_partition_not_same_bound(), calling
partition_bounds_create() and partition_bounds_equal() looks a bit
heavyweight. It doesn't matter much performance-wise, but it feels
like start processing from scratch while we're on quite late stage
already. I've replaced that with more lightweight check. Also I
removed dealing with memory context. This code implies small
non-repetitive memory allocations which only lives during DDL
operation, no need to wrap them with memory context as we don't do so
in other places.
Any objections if I commit this?
------
Regards,
Alexander Korotkov
Supabase
On May 18, 2026, at 05:45, Alexander Korotkov <aekorotkov@gmail.com> wrote:
Hi, Chao!
On Thu, May 14, 2026 at 9:59 AM Chao Li <li.evan.chao@gmail.com> wrote:
To make this patch easier to process, I split it into 4 commits:
0001 - Fixes the bound-overlap bug
0002 - Fix the incorrect HINT message for the DEFAULT case
0003 - Fix the incorrect description about combined bound in the SGML doc
0004 - Reject only-create-default-partition usageThank you for your work. I've revised the patchset.
0002 - I've also fixed gramma of hints in other branches
0004 - In the check_split_partition_not_same_bound(), calling
partition_bounds_create() and partition_bounds_equal() looks a bit
heavyweight. It doesn't matter much performance-wise, but it feels
like start processing from scratch while we're on quite late stage
already. I've replaced that with more lightweight check. Also I
removed dealing with memory context. This code implies small
non-repetitive memory allocations which only lives during DDL
operation, no need to wrap them with memory context as we don't do so
in other places.Any objections if I commit this?
------
Regards,
Alexander Korotkov
Supabase
Hi Alexander,
Thanks for the revisions. I think you may have missed the attachments, so I cannot review the changes.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
On Mon, May 18, 2026 at 2:15 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 18, 2026, at 05:45, Alexander Korotkov <aekorotkov@gmail.com> wrote:
Hi, Chao!
On Thu, May 14, 2026 at 9:59 AM Chao Li <li.evan.chao@gmail.com> wrote:
To make this patch easier to process, I split it into 4 commits:
0001 - Fixes the bound-overlap bug
0002 - Fix the incorrect HINT message for the DEFAULT case
0003 - Fix the incorrect description about combined bound in the SGML doc
0004 - Reject only-create-default-partition usageThank you for your work. I've revised the patchset.
0002 - I've also fixed gramma of hints in other branches
0004 - In the check_split_partition_not_same_bound(), calling
partition_bounds_create() and partition_bounds_equal() looks a bit
heavyweight. It doesn't matter much performance-wise, but it feels
like start processing from scratch while we're on quite late stage
already. I've replaced that with more lightweight check. Also I
removed dealing with memory context. This code implies small
non-repetitive memory allocations which only lives during DDL
operation, no need to wrap them with memory context as we don't do so
in other places.Any objections if I commit this?
------
Regards,
Alexander Korotkov
SupabaseHi Alexander,
Thanks for the revisions. I think you may have missed the attachments, so I cannot review the changes.
Sorry. Here it is.
------
Regards,
Alexander Korotkov
Supabase
Attachments:
v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patchapplication/octet-stream; name=v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patchDownload+14-10
v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patchapplication/octet-stream; name=v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patchDownload+33-2
v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patchapplication/octet-stream; name=v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patchDownload+13-16
v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchapplication/octet-stream; name=v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchDownload+125-1
On May 18, 2026, at 17:16, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:15 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 18, 2026, at 05:45, Alexander Korotkov <aekorotkov@gmail.com> wrote:
Hi, Chao!
On Thu, May 14, 2026 at 9:59 AM Chao Li <li.evan.chao@gmail.com> wrote:
To make this patch easier to process, I split it into 4 commits:
0001 - Fixes the bound-overlap bug
0002 - Fix the incorrect HINT message for the DEFAULT case
0003 - Fix the incorrect description about combined bound in the SGML doc
0004 - Reject only-create-default-partition usageThank you for your work. I've revised the patchset.
0002 - I've also fixed gramma of hints in other branches
0004 - In the check_split_partition_not_same_bound(), calling
partition_bounds_create() and partition_bounds_equal() looks a bit
heavyweight. It doesn't matter much performance-wise, but it feels
like start processing from scratch while we're on quite late stage
already. I've replaced that with more lightweight check. Also I
removed dealing with memory context. This code implies small
non-repetitive memory allocations which only lives during DDL
operation, no need to wrap them with memory context as we don't do so
in other places.Any objections if I commit this?
------
Regards,
Alexander Korotkov
SupabaseHi Alexander,
Thanks for the revisions. I think you may have missed the attachments, so I cannot review the changes.
Sorry. Here it is.
------
Regards,
Alexander Korotkov
Supabase
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
Supabase
On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
Supabase
My suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```
In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
Attachments:
v4-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patchapplication/octet-stream; name=v4-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch; x-unix-mode=0644Download+33-2
v4-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patchapplication/octet-stream; name=v4-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch; x-unix-mode=0644Download+13-16
v4-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patchapplication/octet-stream; name=v4-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch; x-unix-mode=0644Download+14-10
v4-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchapplication/octet-stream; name=v4-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch; x-unix-mode=0644Download+256-1
Hi, Chao!
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003. Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()). I've revised the remaining patch:
made function header comment a bit more detailed and added additional
regression tests. Please, check.
------
Regards,
Alexander Korotkov
Supabase
Attachments:
v5-0001-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchapplication/octet-stream; name=v5-0001-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchDownload+261-1
On May 19, 2026, at 19:00, Alexander Korotkov <aekorotkov@gmail.com> wrote:
Hi, Chao!
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003.
Thanks for pushing them.
Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()).
Yes, I realized that while reviewing v3. That’s reason I didn’t get back v2 and only worked again based on v3.
I've revised the remaining patch:
made function header comment a bit more detailed
This part looks good to me.
and added additional
regression tests. Please, check.
But I don’t see any change for regression test between v4 and v5. Maybe you forgot to save your changes?
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
Hi, Chao!
On Wed, May 20, 2026 at 2:37 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 19, 2026, at 19:00, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003.
Thanks for pushing them.
Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()).Yes, I realized that while reviewing v3. That’s reason I didn’t get back v2 and only worked again based on v3.
I've revised the remaining patch:
made function header comment a bit more detailedThis part looks good to me.
and added additional
regression tests. Please, check.But I don’t see any change for regression test between v4 and v5. Maybe you forgot to save your changes?
Sorry, I just mess up, no changes in tests.
I'm going to push this if no objection.
------
Regards,
Alexander Korotkov
Supabase
On May 20, 2026, at 14:19, Alexander Korotkov <aekorotkov@gmail.com> wrote:
Hi, Chao!
On Wed, May 20, 2026 at 2:37 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 19, 2026, at 19:00, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003.
Thanks for pushing them.
Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()).Yes, I realized that while reviewing v3. That’s reason I didn’t get back v2 and only worked again based on v3.
I've revised the remaining patch:
made function header comment a bit more detailedThis part looks good to me.
and added additional
regression tests. Please, check.But I don’t see any change for regression test between v4 and v5. Maybe you forgot to save your changes?
Sorry, I just mess up, no changes in tests.
I'm going to push this if no objection.
No worries. Then v5 looks good to me.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
On Wed, May 20, 2026 at 9:29 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 20, 2026, at 14:19, Alexander Korotkov <aekorotkov@gmail.com> wrote:
Hi, Chao!
On Wed, May 20, 2026 at 2:37 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 19, 2026, at 19:00, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003.
Thanks for pushing them.
Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()).Yes, I realized that while reviewing v3. That’s reason I didn’t get back v2 and only worked again based on v3.
I've revised the remaining patch:
made function header comment a bit more detailedThis part looks good to me.
and added additional
regression tests. Please, check.But I don’t see any change for regression test between v4 and v5. Maybe you forgot to save your changes?
Sorry, I just mess up, no changes in tests.
I'm going to push this if no objection.No worries. Then v5 looks good to me.
Thank you, pushed.
------
Regards,
Alexander Korotkov
Supabase
On Wed, May 20, 2026 at 2:46 PM Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 9:29 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 20, 2026, at 14:19, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 2:37 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 19, 2026, at 19:00, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003.
Thanks for pushing them.
Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()).Yes, I realized that while reviewing v3. That’s reason I didn’t get back v2 and only worked again based on v3.
I've revised the remaining patch:
made function header comment a bit more detailedThis part looks good to me.
and added additional
regression tests. Please, check.But I don’t see any change for regression test between v4 and v5. Maybe you forgot to save your changes?
Sorry, I just mess up, no changes in tests.
I'm going to push this if no objection.No worries. Then v5 looks good to me.
Thank you, pushed.
Uhhh, most of buildfarm animals don't support locales used in our
tests. I've to revert that,
------
Regards,
Alexander Korotkov
Supabase
On Wed, May 20, 2026 at 11:29 PM Alexander Korotkov
<aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 2:46 PM Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 9:29 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 20, 2026, at 14:19, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 2:37 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 19, 2026, at 19:00, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003.
Thanks for pushing them.
Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()).Yes, I realized that while reviewing v3. That’s reason I didn’t get back v2 and only worked again based on v3.
I've revised the remaining patch:
made function header comment a bit more detailedThis part looks good to me.
and added additional
regression tests. Please, check.But I don’t see any change for regression test between v4 and v5. Maybe you forgot to save your changes?
Sorry, I just mess up, no changes in tests.
I'm going to push this if no objection.No worries. Then v5 looks good to me.
Thank you, pushed.
Uhhh, most of buildfarm animals don't support locales used in our
tests. I've to revert that,
The another attempt is attached. Now use -0.0 and 0.0 as binary
different but logically equivalent values, no locale dependence.
------
Regards,
Alexander Korotkov
Supabase
Attachments:
v6-0001-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchapplication/octet-stream; name=v6-0001-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patchDownload+259-1
On May 21, 2026, at 05:17, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 11:29 PM Alexander Korotkov
<aekorotkov@gmail.com> wrote:On Wed, May 20, 2026 at 2:46 PM Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 9:29 AM Chao Li <li.evan.chao@gmail.com> wrote:
On May 20, 2026, at 14:19, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Wed, May 20, 2026 at 2:37 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 19, 2026, at 19:00, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Tue, May 19, 2026 at 5:50 AM Chao Li <li.evan.chao@gmail.com> wrote:On May 18, 2026, at 20:04, Alexander Korotkov <aekorotkov@gmail.com> wrote:
On Mon, May 18, 2026 at 2:57 PM Chao Li <li.evan.chao@gmail.com> wrote:
<v3-0003-Clarify-SPLIT-PARTITION-bound-requirements-in-doc.patch><v3-0001-Fix-SPLIT-PARTITION-range-bound-validation-with-D.patch><v3-0002-Fix-SPLIT-PARTITION-hint-for-DEFAULT-partition-bo.patch><v3-0004-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
v3-0001 through v3-0003 look good to me.
For v3-0004, I have a suspicion, but it's late here and my brain is getting slow, so I would like to study it more tomorrow.
Sure, take your time.
------
Regards,
Alexander Korotkov
SupabaseMy suspicion was that check_split_partition_not_same_bound() now has two paths. The RANGE path honors collation, while the LIST path does not. So I spent some time creating a test that uses a case-insensitive collation:
```
evantest=# create collation case_insensitive (provider=icu, locale='und-u-ks-level2', deterministic = false);
CREATE COLLATION
evantest=# create table t (b text collate case_insensitive) partition by list (b);
CREATE TABLE
evantest=# create table tp_ab partition of t for values in ('a', 'b');
CREATE TABLE
evantest=# alter table t split partition tp_ab into
evantest-# (partition tp_a for values in ('a', 'A'),
evantest(# partition tp_default default);
ERROR: cannot split partition "tp_ab" only to add a DEFAULT partition
LINE 2: (partition tp_a for values in ('a', 'A'),
^
DETAIL: The non-DEFAULT partition would keep the same partition bound.
HINT: Use CREATE TABLE ... PARTITION OF ... DEFAULT to add a DEFAULT partition.
```In this test, the split partition’s bound is ('a', 'b'), and the new partition’s bound is ('a', 'A'). Their list lengths are both 2, but the two bounds are actually different, because 'a' and 'A' are considered equal by the collation.
So, in the LIST path, since check_partition_bounds_for_split_list() has already ensured that the new partition’s bound is contained within the split partition’s bound, we need to check the reverse direction as well. Whether the split partition’s bound is also contained in the new partition’s bound. If yes, the two bounds are identical.
See the attached v4 for my changes for 0004. 0001-0003 are unchanged. Since 0001 and 0003 are independent of 0004, maybe they can be pushed first.
I've pushed 0001-0003.
Thanks for pushing them.
Thank you for discovering the collation issue
in 0004. Note that original approach of using
partition_bounds_equal() can't handle different collations too (as it
internally uses datumIsEqual()).Yes, I realized that while reviewing v3. That’s reason I didn’t get back v2 and only worked again based on v3.
I've revised the remaining patch:
made function header comment a bit more detailedThis part looks good to me.
and added additional
regression tests. Please, check.But I don’t see any change for regression test between v4 and v5. Maybe you forgot to save your changes?
Sorry, I just mess up, no changes in tests.
I'm going to push this if no objection.No worries. Then v5 looks good to me.
Thank you, pushed.
Uhhh, most of buildfarm animals don't support locales used in our
tests. I've to revert that,The another attempt is attached. Now use -0.0 and 0.0 as binary
different but logically equivalent values, no locale dependence.------
Regards,
Alexander Korotkov
Supabase
<v6-0001-Reject-degenerate-SPLIT-PARTITION-with-DEFAULT-pa.patch>
Thank you very much for taking care of that. This was a lesson learned for me.
Actually, when I added the ICU test, I did think about whether regression tests support ICU. So I searched the existing tests for “icu” and found some examples. For instance, src/test/regress/sql/collate.icu.utf8.sql has tests like:
```
RESET icu_validation_level;
CREATE COLLATION testx (provider = icu, locale = '@colStrength=primary;nonsense=yes'); DROP COLLATION testx;
CREATE COLLATION testx (provider = icu, locale = 'nonsense-nowhere'); DROP COLLATION testx;
```
I added v6 to CF to get a CI run and check whether the buildfarm is happy with v6. The CI test has passed now. See [1]https://commitfest.postgresql.org/patch/6796/. The new test using “-0.0" also looks good to me.
[1]: https://commitfest.postgresql.org/patch/6796/
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/