Michael Paquier [Mon, 2 Mar 2026 00:38:37 +0000 (09:38 +0900)]
Fix set of issues with extended statistics on expressions
This commit addresses two defects regarding extended statistics on
expressions:
- When building extended statistics in lookup_var_attr_stats(), the call
to examine_attribute() did not account for the possibility of a NULL
return value. This can happen depending on the behavior of a typanalyze
callback — for example, if the callback returns false, if no rows are
sampled, or if no statistics are computed. In such cases, the code
attempted to build MCV, dependency, and ndistinct statistics using a
NULL pointer, incorrectly assuming valid statistics were available,
which could lead to a server crash.
- When loading extended statistics for expressions,
statext_expressions_load() did not account for NULL entries in the
pg_statistic array storing expression statistics. Such NULL entries can
be generated when statistics collection fails for an expression, as may
occur during the final step of serialize_expr_stats(). An extended
statistics object defining N expressions requires N corresponding
elements in the pg_statistic array stored for the expressions, and some
of these elements can be NULL. This situation is reachable when a
typanalyze callback returns true, but sets stats_valid to indicate that
no useful statistics could be computed.
While these scenarios cannot occur with in-core typanalyze callbacks, as
far as I have analyzed, they can be triggered by custom data types with
custom typanalyze implementations, at least.
No tests are added in this commit. A follow-up commit will introduce a
test module that can be extended to cover similar edge cases if
additional issues are discovered. This takes care of the core of the
problem.
Attribute and relation statistics already offer similar protections:
- ANALYZE detects and skips the build of invalid statistics.
- Invalid catalog data is handled defensively when loading statistics.
This issue exists since the support for extended statistics on
expressions has been added, down to v14 as of a4d75c86bf15. Backpatch
to all supported stable branches.
Author: Michael Paquier <michael@paquier.xyz> Reviewed-by: Corey Huinker <corey.huinker@gmail.com> Reviewed-by: Chao Li <li.evan.chao@gmail.com>
Discussion: https://postgr.es/m/aaDrJsE1I5mrE-QF@paquier.xyz
Backpatch-through: 14
Tom Lane [Sun, 1 Mar 2026 17:56:55 +0000 (12:56 -0500)]
Correctly calculate "MCV frequency" for a unique column.
In commit bd3e3e9e5, I over-hastily used 1 / rel->rows as the assumed
frequency of entries in a column that ANALYZE has found to be unique.
However, rel->rows is the number of table rows that are estimated to
pass the query's restriction conditions, so that we got a too-large
result if the query has selective restrictions. What I should have
used is 1 / rel->tuples, since that is the estimated total number of
table rows. The pre-existing code path that digs a frequency out of
the histogram produces a frequency relative to the whole table, so
surely this new alternative code path must do so as well. Any
correction needed on the basis of selectivity must be done by the
user of the mcv_freq value.
Fixing this causes all the regression test plans changed by bd3e3e9e5
to revert to what they had been, except for the first change in
join.out. As I correctly argued in bd3e3e9e5, in that test case we
have no stats and should not risk a hash join. Evidently I was less
correct to argue that the other changes were improvements.
Reported-by: Joel Jacobson <joel@compiler.org> Diagnosed-by: Tender Wang <tndrwang@gmail.com>
Author: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/341b723c-da45-4058-9446-1514dedb17c1@app.fastmail.com
Fujii Masao [Sat, 28 Feb 2026 14:56:46 +0000 (23:56 +0900)]
psql: Show comments in \dRp+, \dRs+, and \dX+ psql meta-commands.
Previously, the psql meta-commands that list publications, subscriptions,
and extended statistics did not display their associated comments,
whereas other \d meta-commands did. This made it inconvenient for users
to view these objects together with their descriptions.
This commit improves \dRp+ and \dRs+ to include comments for publications
and subscriptions. It also extends the \dX meta-command to accept the + option,
allowing comments for extended statistics to be shown when requested.
Author: Fujii Masao <masao.fujii@gmail.com> Co-authored-by: Jim Jones <jim.jones@uni-muenster.de> Reviewed-by: Matheus Alcantara <matheusssilv97@gmail.com> Reviewed-by: Chao Li <li.evan.chao@gmail.com>
Discussion: https://postgr.es/m/CAHGQGwGL4JqiKA26fnGx-cTM=VzoTs_uzqejvj4Fawyr4uLUUw@mail.gmail.com
Tom Lane [Fri, 27 Feb 2026 20:20:16 +0000 (15:20 -0500)]
Doc: improve user docs and code comments about EXISTS(SELECT * ...).
Point out that Postgres automatically optimizes away the target list
of an EXISTS' subquery, except in weird cases such as target lists
containing set-returning functions. Thus, both common conventions
EXISTS(SELECT * FROM ...) and EXISTS(SELECT 1 FROM ...) are
overhead-free and there's little reason to prefer one over the other.
In the code comments, mention that the SQL spec says that
EXISTS(SELECT * FROM ...) should be interpreted as EXISTS(SELECT
some-literal FROM ...), but we don't choose to do it exactly that way.
Author: Peter Eisentraut <peter@eisentraut.org> Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/9b301c70-3909-4f0f-98ca-9e3c4d142f3e@eisentraut.org
Tom Lane [Fri, 27 Feb 2026 17:54:02 +0000 (12:54 -0500)]
Don't flatten join alias Vars that are stored within a GROUP RTE.
The RTE's groupexprs list is used for deparsing views, and for that
usage it must contain the original alias Vars; else we can get
incorrect SQL output. But since commit 247dea89f,
parseCheckAggregates put the GROUP BY expressions through
flatten_join_alias_vars before building the RTE_GROUP RTE.
Changing the order of operations there is enough to fix it.
This patch unfortunately can do nothing for already-created views:
if they use a coding pattern that is subject to the bug, they will
deparse incorrectly and hence present a dump/reload hazard in the
future. The only fix is to recreate the view from the original SQL.
But the trouble cases seem to be quite narrow. AFAICT the output
was only wrong for "SELECT ... t1 LEFT JOIN t2 USING (x) GROUP BY x"
where t1.x and t2.x were not of identical data types and t1.x was
the side that required an implicit coercion. If there was no hidden
coercion, or if the join was plain, RIGHT, or FULL, the deparsed
output was uglier than intended but not functionally wrong.
Reported-by: Swirl Smog Dowry <swirl-smog-dowry@duck.com>
Author: Tom Lane <tgl@sss.pgh.pa.us> Reviewed-by: Richard Guo <guofenglinux@gmail.com>
Discussion: https://postgr.es/m/CA+-gibjCg_vjcq3hWTM0sLs3_TUZ6Q9rkv8+pe2yJrdh4o4uoQ@mail.gmail.com
Backpatch-through: 18
John Naylor [Fri, 27 Feb 2026 13:30:41 +0000 (20:30 +0700)]
Centralize detection of x86 CPU features
We now maintain an array of booleans that indicate which features were
detected at runtime. When code wants to check for a given feature,
the array is automatically checked if it has been initialized and if
not, a single function checks all features at once.
Move all x86 feature detection to pg_cpu_x86.c, and move the CRC
function choosing logic to the file where the hardware-specific
functions are defined, consistent with more recent hardware-specific
files in src/port.
Andrew Dunstan [Fri, 27 Feb 2026 12:14:06 +0000 (07:14 -0500)]
Clean up nodes that are no longer of use in 007_pgdumpall.pl
Oversight in commit 763aaa06f03. When nodes are going out of scope, we
should stop the underlying postmasters rather than waiting for the
script to end.
Michael Paquier [Fri, 27 Feb 2026 09:59:41 +0000 (18:59 +0900)]
Use pg_malloc_object() and pg_alloc_array() variants in frontend code
This commit updates the frontend tools (src/bin/, contrib/ and
src/test/) to use the memory allocation variants based on
pg_malloc_object() and pg_malloc_array() in various code paths. This
does not cover all the allocations, but a good chunk of them.
Like all the changes of this kind (31d3847a37be, etc.), this should
encourage any future code to use this new style.
Author: Andreas Karlsson <andreas@proxel.se>
Discussion: https://postgr.es/m/cfb645da-6b3a-4f22-9bcc-5bc46b0e9c61@proxel.se
heapam_scan_analyze_next_tuple() doesn't distinguish between dead and
recently dead tuples when counting them, so it doesn't need OldestXmin.
GetOldestNonRemovableTransactionId() isn't free, so removing it is a
win.
Looking at other table AMs implementing table_scan_analyze_next_tuple(),
we couldn't find one using OldestXmin either, so remove it from the
callback.
Melanie Plageman [Thu, 26 Feb 2026 20:33:36 +0000 (15:33 -0500)]
Simplify visibility check in heap_page_would_be_all_visible()
heap_page_would_be_all_visible() does not need to distinguish between
HEAPTUPLE_RECENTLY_DEAD and HEAPTUPLE_DEAD tuples: any tuple in a state
other than HEAPTUPLE_LIVE means the page is not all-visible and
heap_page_would_be_all_visible() returns false.
Given that, calling HeapTupleSatisfiesVacuum() is unnecessary, since it
performs extra work to distinguish between dead and recently dead tuples
using OldestXmin. Replace it with the more minimal
HeapTupleSatisfiesVacuumHorizon().
Author: Melanie Plageman <melanieplageman@gmail.com> Reviewed-by: Kirill Reshke <reshkekirill@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Chao Li <li.evan.chao@gmail.com>
Discussion: https://postgr.es/m/CALdSSPjvhGXihT_9f-GJabYU%3D_PjrFDUxYaURuTbfLyQM6TErg%40mail.gmail.com
Jeff Davis [Thu, 26 Feb 2026 20:23:22 +0000 (12:23 -0800)]
Fix more multibyte issues in ltree.
Commit 84d5efa7e3 missed some multibyte issues caused by short-circuit
logic in the callers. The callers assumed that if the predicate string
is longer than the label string, then it couldn't possibly be a match,
but it can be when using case-insensitive matching (LVAR_INCASE) if
casefolding changes the byte length.
Fix by refactoring to get rid of the short-circuit logic as well as
the function pointer, and consolidate the logic in a replacement
function ltree_label_match().
Melanie Plageman [Thu, 26 Feb 2026 20:04:49 +0000 (15:04 -0500)]
Rename LVRelState VM-related logging counters
The LVRelState fields that track newly all-visible/all-frozen pages were
previously named vm_new_visible_pages, vm_new_frozen_pages, and
vm_new_visible_frozen_pages. The correct terminology is all-visible and
all-frozen; omitting “all” was open to misinterpretation, as the page
isn't visible or invisible, rather all the tuples on the page are
visible to all running and future transactions. Rename the members
accordingly.
Author: Melanie Plageman <melanieplageman@gmail.com> Suggested-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/bqc4kh5midfn44gnjiqez3bjqv4zogydguvdn446riw45jcf3y%404ez66il7ebvk
Andres Freund [Thu, 26 Feb 2026 15:19:23 +0000 (10:19 -0500)]
instrumentation: Rename INSTR_TIME_LT macro to INSTR_TIME_GT
This was incorrectly named "LT" for "larger than" in e5a5e0a90750d66, but
that is against existing conventions, where "LT" means "less than".
Clarify by using "GT" for "greater than" in macro name, and add a missing
comment at the top of instr_time.h to note the macro's existence.
Reported by: Peter Smith <smithpb2250@gmail.com>
Author: Lukas Fittl <lukas@fittl.com> Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/CAHut%2BPut94CTpjQsqOJHdHkgJ2ZXq%2BqVSfMEcmDKLiWLW-hPfA%40mail.gmail.com
Andrew Dunstan [Mon, 23 Feb 2026 20:08:55 +0000 (15:08 -0500)]
Add non-text output formats to pg_dumpall
pg_dumpall can now produce output in custom, directory, or tar formats
in addition to plain text SQL scripts. When using non-text formats,
pg_dumpall creates a directory containing:
- toc.glo: global data (roles and tablespaces) in custom format
- map.dat: mapping between database OIDs and names
- databases/: subdirectory with per-database archives named by OID
pg_restore is extended to handle these pg_dumpall archives, restoring
globals and then each database. The --globals-only option can be used
to restore only the global objects.
This enables parallel restore of pg_dumpall output and selective
restoration of individual databases from a cluster-wide backup.
Noah Misch [Thu, 26 Feb 2026 02:13:22 +0000 (18:13 -0800)]
EUC_CN, EUC_JP, EUC_KR, EUC_TW: Skip U+00A0 tests instead of failing.
Settings that ran the new test euc_kr.sql to completion would fail these
older src/pl tests. Use alternative expected outputs, for which psql
\gset and \if have reduced the maintenance burden. This fixes
"LANG=ko_KR.euckr LC_MESSAGES=C make check-world". (LC_MESSAGES=C fixes
IO::Pty usage in tests 010_tab_completion and 001_password.) That file
is new in commit c67bef3f3252a3a38bf347f9f119944176a796ce. Back-patch
to v14, like that commit.
Fujii Masao [Thu, 26 Feb 2026 00:01:52 +0000 (09:01 +0900)]
doc: Clarify INCLUDING COMMENTS behavior in CREATE TABLE LIKE.
The documentation for the INCLUDING COMMENTS option of the LIKE clause
in CREATE TABLE was inaccurate and incomplete. It stated that comments for
copied columns, constraints, and indexes are copied, but regarding comments
on constraints in reality only comments on CHECK and NOT NULL constraints
are copied; comments on other constraints (such as primary keys) are not.
In addition, comments on extended statistics are copied, but this was not
documented.
The CREATE FOREIGN TABLE documentation had a similar omission: comments
on extended statistics are also copied, but this was not mentioned.
This commit updates the documentation to clarify the actual behavior.
The CREATE TABLE reference now specifies that comments on copied columns,
CHECK constraints, NOT NULL constraints, indexes, and extended statistics are
copied. The CREATE FOREIGN TABLE reference now notes that comments on
extended statistics are copied as well.
Backpatch to all supported versions. Documentation updates related to
CREATE FOREIGN TABLE LIKE and NOT NULL constraint comment copying are
not applied to v17 and earlier, since those features were introduced in v18.
Fujii Masao [Wed, 25 Feb 2026 23:46:12 +0000 (08:46 +0900)]
Fix ProcWakeup() resetting wrong waitStart field.
Previously, when one process woke another that was waiting on a lock,
ProcWakeup() incorrectly cleared its own waitStart field (i.e.,
MyProc->waitStart) instead of that of the process being awakened.
As a result, the awakened process retained a stale lock-wait start timestamp.
This did not cause user-visible issues. pg_locks.waitstart was reported as
NULL for the awakened process (i.e., when pg_locks.granted is true),
regardless of the waitStart value.
Tom Lane [Wed, 25 Feb 2026 15:51:42 +0000 (10:51 -0500)]
Stabilize output of new isolation test insert-conflict-do-update-4.
The test added by commit 4b760a181 assumed that a table's physical
row order would be predictable after an UPDATE. But a non-heap table
AM might produce some other order. Even with heap AM, the assumption
seems risky; compare a3fd53bab for instance. Adding an ORDER BY is
cheap insurance and doesn't break any goal of the test.
Author: Pavel Borisov <pashkin.elfe@gmail.com> Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/CALT9ZEHcE6tpvumScYPO6pGk_ASjTjWojLkodHnk33dvRPHXVw@mail.gmail.com
Backpatch-through: 14
Richard Guo [Wed, 25 Feb 2026 02:13:21 +0000 (11:13 +0900)]
Fix unsafe RTE_GROUP removal in simplify_EXISTS_query
When simplify_EXISTS_query removes the GROUP BY clauses from an EXISTS
subquery, it previously deleted the RTE_GROUP RTE directly from the
subquery's range table.
This approach is dangerous because deleting an RTE from the middle of
the rtable list shifts the index of any subsequent RTE, which can
silently corrupt any Var nodes in the query tree that reference those
later relations. (Currently, this direct removal has not caused
problems because the RTE_GROUP RTE happens to always be the last entry
in the rtable list. However, relying on that is extremely fragile and
seems like trouble waiting to happen.)
Instead of deleting the RTE_GROUP RTE, this patch converts it in-place
to be RTE_RESULT type and clears its groupexprs list. This preserves
the length and indexing of the rtable list, ensuring all Var
references remain intact.
Reported-by: Tom Lane <tgl@sss.pgh.pa.us>
Author: Richard Guo <guofenglinux@gmail.com> Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/3472344.1771858107@sss.pgh.pa.us
Backpatch-through: 18
John Naylor [Wed, 25 Feb 2026 01:44:59 +0000 (08:44 +0700)]
Fix USE_SLICING_BY_8_CRC32C builds on x86
A future commit will move the CRC function choosing logic to the
file where the hardware-specific functions are defined, but until
then add guards for builds without those functions. Oversight in
commit b9278871f.
Per buildfarm animal rhinoceros
Reported-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/4014992.1771963187@sss.pgh.pa.us
Jacob Champion [Tue, 24 Feb 2026 22:01:37 +0000 (14:01 -0800)]
pg_upgrade: Use max_protocol_version=3.0 for older servers
The grease patch in 4966bd3ed found its first problem: prior to the
February 2018 patch releases, no server knew how to negotiate protocol
versions, so pg_upgrade needs to take that into account when speaking to
those older servers.
This will be true even after the grease feature is reverted; we don't
need anyone to trip over this again in the future. Backpatch so that all
supported versions of pg_upgrade can gracefully handle an update to the
default protocol version. (This is needed for any distributions that
link older binaries against newer libpqs, such as Debian.) Branches
prior to 18 need an additional version check, for the existence of
max_protocol_version.
Per buildfarm member crake.
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/CAOYmi%2B%3D4QhCjssfNEoZVK8LPtWxnfkwT5p-PAeoxtG9gpNjqOQ%40mail.gmail.com
Backpatch-through: 14
Álvaro Herrera [Tue, 24 Feb 2026 16:34:56 +0000 (17:34 +0100)]
Add backtrace support for Windows using DbgHelp API
Previously, backtrace generation on Windows would return an "unsupported"
message. With this commit, we rely on CaptureStackBackTrace() to capture
the call stack and the DbgHelp API (SymFromAddrW, SymGetLineFromAddrW64)
for symbol resolution.
Symbol handler initialization (SymInitialize) is performed once per
process and cached. If initialization fails, the report for it is
returned as the backtrace output. The symbol handler is cleaned up via
on_proc_exit() to release DbgHelp resources.
The implementation provides symbol names, offsets, and addresses. When
PDB files are available, it also includes source file names and line
numbers. Symbol names and file paths are converted from UTF-16 to the
database encoding using wchar2char(), which properly handles both UTF-8
and non-UTF-8 databases on Windows. When symbol information is
unavailable or encoding conversion fails, it falls back to displaying raw
addresses.
The implementation uses the explicit UTF16 versions of the DbgHelp
functions (SYMBOL_INFOW, SymFromAddrW, IMAGEHLP_LINEW64,
SymGetLineFromAddrW64) rather than the generic versions. This allows us
to rely on predictable encoding conversion, rather than using the
haphazard ANSI codepage that we'd get otherwise.
DbgHelp is apparently available on all Windows platforms we support, so
there are no version number checks.
Author: Bryan Green <dbryan.green@gmail.com> Reviewed-by: Euler Taveira <euler@eulerto.com> Reviewed-by: Jakub Wartak <jakub.wartak@enterprisedb.com> Reviewed-by: Greg Burd <greg@burd.me>
Discussion: https://postgr.es/m/a692c0fe-caca-4c08-9c5d-debfd0ef2504@gmail.com
Peter Eisentraut [Tue, 24 Feb 2026 09:56:17 +0000 (10:56 +0100)]
Make ALTER DOMAIN VALIDATE CONSTRAINT no-op when constraint is already validated
Currently, AlterDomainValidateConstraint will re-validate a constraint
that has already been validated, which would just waste cycles. This
operation should be a no-op when the constraint is already validated.
This also aligns with ATExecValidateConstraint.
Author: jian he <jian.universality@gmail.com>
Discussion: https://postgr.es/m/CACJufxG=-Dv9fPJHqkA9c-wGZ2dDOWOXSp-X-0K_G7r-DgaASw@mail.gmail.com
Peter Eisentraut [Tue, 24 Feb 2026 09:30:50 +0000 (10:30 +0100)]
Allow ALTER COLUMN SET EXPRESSION on virtual columns with CHECK constraints
Previously, changing the generation expression of a virtual column was
prohibited if the column was referenced by a CHECK constraint. This
lifts that restriction.
RememberAllDependentForRebuilding within ATExecSetExpression will
rebuild all the dependent constraints, later ATPostAlterTypeCleanup
queues the required AlterTableStmt operations for ALTER TABLE Phase 3
execution.
Overall, ALTER COLUMN SET EXPRESSION on virtual columns may require
scanning the table to re-verify any associated CHECK constraints, but
it does not require a table rewrite in ALTER TABLE Phase 3.
Author: jian he <jian.universality@gmail.com> Reviewed-by: Matheus Alcantara <matheusssilv97@gmail.com>
Discussion: https://postgr.es/m/CACJufxH3VETr7orF5rW29GnDk3n1wWbOE3WdkHYd3iPGrQ9E_A@mail.gmail.com
Michael Paquier [Tue, 24 Feb 2026 04:26:37 +0000 (13:26 +0900)]
Fix variety of typos and grammar mistakes
This commit includes a batch of fixes for various minor typos and
grammar mistakes, that have been proposed to the hackers mailing list
since the beginning of January.
Similar batches are planned on a bi-monthly basis depending on the
amount received, with the next one for the end of April.
Michael Paquier [Tue, 24 Feb 2026 03:54:23 +0000 (12:54 +0900)]
doc: Adjust some markups on pg_waldump page
Author: Peter Smith <smithpb2250@gmail.com> Reviewed-by: Chao Li <li.evan.chao@gmail.com> Reviewed-by: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
Discussion: https://postgr.es/m/CAHut+PuuPps9bUPvouU5dH=tOTiF8QBzQox5O7DqXeOFdda79Q@mail.gmail.com
Michael Paquier [Tue, 24 Feb 2026 03:34:42 +0000 (12:34 +0900)]
fe_utils: Sprinkle some pg_malloc_object() and pg_malloc_array()
The idea is to encourage more the use of these allocation routines
across the tree, as these offer stronger type safety guarantees than
pg_malloc() & friends (type cast in the result, sizeof() embedded).
This commit updates some code paths of src/fe_utils/.
Author: Henrik TJ <henrik@0x48.dk> Reviewed-by: Andreas Karlsson <andreas@proxel.se>
Discussion: https://postgr.es/m/6df1b64e-1314-9afd-41a3-3fefb76225e1@0x48.dk
Nathan Bossart [Mon, 23 Feb 2026 20:55:21 +0000 (14:55 -0600)]
Allow pg_{read,write}_all_data to access large objects.
Since the initial goal of pg_read_all_data was to be able to run
pg_dump as a non-superuser without explicitly granting access to
every object, it follows that it should allow reading all large
objects. For consistency, pg_write_all_data should allow writing
all large objects, too.
Tom Lane [Mon, 23 Feb 2026 20:30:44 +0000 (15:30 -0500)]
Work around lgamma(NaN) bug on AIX.
lgamma(NaN) should produce NaN, but on older versions of AIX
it reports an ERANGE error. While that's been fixed in the latest
version of libm, it'll take awhile for the fix to propagate. This
workaround is harmless even when the underlying bug does get fixed.
Peter Eisentraut [Mon, 23 Feb 2026 20:17:06 +0000 (21:17 +0100)]
Use LOCKMODE in parse_relation.c/.h
There were a couple of comments in parse_relation.c
> Note: properly, lockmode should be declared LOCKMODE not int, but that
> would require importing storage/lock.h into parse_relation.h. Since
> LOCKMODE is typedef'd as int anyway, that seems like overkill.
but actually LOCKMODE has been in storage/lockdefs.h for a while,
which is intentionally a more narrow header. So we can include that
one in parse_relation.h and just use LOCKMODE normally.
An alternative would be to add a duplicate typedef into
parse_relation.h, but that doesn't seem necessary here.
Reviewed-by: Andreas Karlsson <andreas@proxel.se> Reviewed-by: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
Discussion: https://www.postgresql.org/message-id/flat/4bcd65fb-2497-484c-bb41-83cb435eb64d%40eisentraut.org
Jacob Champion [Mon, 23 Feb 2026 18:48:20 +0000 (10:48 -0800)]
libpq: Grease the protocol by default
Send PG_PROTOCOL_GREASE and _pq_.test_protocol_negotiation, which were
introduced in commit d8d7c5dc8, by default, and fail the connection if
the server attempts to claim support for them. The hope is to provide
feedback to noncompliant implementations and gain confidence in our
ability to advance the protocol. (See the other commit for details.)
To help end users navigate the situation, a link to our documentation
that explains the behavior is displayed. We append this to the error
message when the NegotiateProtocolVersion response is incorrect, or when
the peer sends an error during startup that appears to be grease-
related.
It's still possible for users to connect to servers that don't support
protocol negotiation, by adding max_protocol_version=3.0 to their
connection strings. Only the default connection behavior is impacted.
This commit is tracked as a PG19 open item and will be reverted before
RC1. (The implementation here doesn't handle negotiation with later
server versions, so it can't be released into the wild as a
five-year-supported feature. But an improved implementation might be
able to do so, in the future...)
Author: Jelte Fennema-Nio <postgres@jeltef.nl> Co-authored-by: Jacob Champion <jacob.champion@enterprisedb.com>
Discussion: https://postgr.es/m/DDPR5BPWH1RJ.1LWAK6QAURVAY%40jeltef.nl
Tom Lane [Mon, 23 Feb 2026 18:34:22 +0000 (13:34 -0500)]
Restore AIX support.
The concerns that led us to remove AIX support in commit 0b16bb877
have now been alleviated:
1. IBM has stepped forward to provide support, including buildfarm
animal(s).
2. AIX 7.2 and later seem to be fine with large pg_attribute_aligned
requirements. Since 7.1 is now EOL anyway, we can just cease to
support it.
3. Tossing xlc support overboard seems okay as well. It's a bit
sad to drop one of the few remaining non-gcc-alike compilers, but
working around xlc's bugs and idiosyncrasies doesn't seem justified
by the theoretical portability benefits.
4. Likewise, we can stop supporting 32-bit AIX builds. This is
not so much about whether we could build such executables as that
they're too much of a pain to manage in the field, due to limited
address space available for dynamic library loading.
5. We hit on a way to manage catalog column alignment that doesn't
require continuing developer effort (see commit ecae09725).
Hence, this commit reverts 0b16bb877 and some follow-on commits
such as e6bb491bf, except for not putting back XLC support nor
the changes related to catalog column alignment.
Some other notable changes from the way things were in v16:
Prefer unnamed POSIX semaphores on AIX, rather than the default
choice of SysV semaphores.
Include /opt/freeware/lib in -Wl,-blibpath, even when it is not
mentioned anywhere in LDFLAGS.
Remove platform-specific adjustment of MEMSET_LOOP_LIMIT; maybe
that's still the right thing, but it really ought to be re-tested.
Silence compiler warnings related to getpeereid(), wcstombs_l(),
and PAM conversation procs.
Accept "libpythonXXX.a" as an okay name for the Python shared
library (but only on AIX!).
Author: Aditya Kamath <Aditya.Kamath1@ibm.com>
Author: Srirama Kucherlapati <sriram.rk@in.ibm.com> Co-authored-by: Peter Eisentraut <peter@eisentraut.org> Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/CY5PR11MB63928CC05906F27FB10D74D0FD322@CY5PR11MB6392.namprd11.prod.outlook.com
Tom Lane [Mon, 23 Feb 2026 17:34:51 +0000 (12:34 -0500)]
Cope with AIX's alignment woes by using _Pragma("pack").
Because we assume that int64 and double have the same alignment
requirement, AIX's default behavior that alignof(double) = 4 while
alignof(int64) = 8 is a headache. There are two issues:
1. We align both int8 and float8 tuple columns per ALIGNOF_DOUBLE,
which is an ancient choice that can't be undone without breaking
pg_upgrade and creating some subtle SQL-level compatibility issues
too. However, the cost of that is just some marginal inefficiency
in fetching int8 values, which can't be too awful if the platform
architects were willing to pay the same costs for fetching float8s.
So our decision is to leave that alone. This patch makes our
alignment choices the same as they were pre-v17, namely that
ALIGNOF_DOUBLE and ALIGNOF_INT64_T are whatever the compiler prefers
and then MAXIMUM_ALIGNOF is the larger of the two. (On all supported
platforms other than AIX, all three values will be the same.)
2. We need to overlay C structs onto catalog tuples, and int8 fields
in those struct declarations may not be aligned to match this rule.
In the old branches we had some annoying rules about ordering catalog
columns to avoid alignment problems, but nobody wants to resurrect
those. However, there's a better answer: make the compiler construe
those struct declarations the way we need it to by using the pack(N)
pragma. This requires no manual effort to maintain going forward;
we only have to insert the pragma into all the catalog *.h files.
(As the catalogs stand at this writing, nothing actually changes
because we've not moved any affected columns since v16; hence no
catversion bump is required. The point of this is to not have
to worry about the issue going forward.)
We did not have this option when the AIX port was first made. This
patch depends on the C99 feature _Pragma(), as well as the pack(N)
pragma which dates to somewhere around gcc 4.0, and probably doesn't
exist in xlc at all. But now that we've agreed to toss xlc support
out the window, there doesn't seem to be a reason not to go this way.
In passing, I got rid of LONGALIGN[_DOWN] along with the configure
probes for ALIGNOF_LONG. We were not using those anywhere and it
seems highly unlikely that we'd do so in future. Instead supply
INT64ALIGN[_DOWN], which isn't used either but at least could
have a good reason to be used.
Nathan Bossart [Mon, 23 Feb 2026 17:22:04 +0000 (11:22 -0600)]
Warn upon successful MD5 password authentication.
This uses the "connection warning" infrastructure introduced by
commit 1d92e0c2cc to emit a WARNING when an MD5 password is used to
authenticate. MD5 password support was marked as deprecated in
v18 and will be removed in a future release of Postgres. These
warnings are on by default but can be turned off via the existing
md5_password_warnings parameter.
Reviewed-by: Andreas Karlsson <andreas@proxel.se> Reviewed-by: Xiangyu Liang <liangxiangyu_2013@163.com>
Discussion: https://postgr.es/m/aYzeAYEbodkkg5e-%40nathan
Peter Eisentraut [Mon, 23 Feb 2026 16:38:06 +0000 (17:38 +0100)]
Rename validate_relation_kind()
There are three static definitions of validate_relation_kind() in the
codebase, one each in table.c, indexam.c and sequence.c, validating that
the given relation is a table, an index or a sequence respectively.
The compiler knows which definition to use where because they are static.
But this could be confusing to a reader. Rename these functions so that
their names reflect the kind of relation they are validating. While at
it, also update the comments in table.c to clarify the definition of
table-like relkinds so that we don't have to maintain the exclusion list
as the set of relkinds undergoes changes.
Peter Eisentraut [Mon, 23 Feb 2026 16:26:29 +0000 (17:26 +0100)]
Flip logic in table validate_relation_kind
It instead of checking which relkinds it shouldn't be, explicitly list
the ones we accept. This is used to check which relkinds are accepted
in table_open() and related functions. Before this change, figuring
that out was always a few steps too complicated. This also makes
changes for new relkinds more explicit instead of accidental.
Finally, this makes this more aligned with the functions of the same
name in src/backend/access/index/indexam.c and
src/backend/access/sequence/sequence.c.
Andrew Dunstan [Sat, 21 Feb 2026 22:17:48 +0000 (17:17 -0500)]
Disallow CR and LF in database, role, and tablespace names
Previously, these characters could cause problems when passed through
shell commands, and were flagged with a comment in string_utils.c
suggesting they be rejected in a future major release.
The affected commands are CREATE DATABASE, CREATE ROLE, CREATE TABLESPACE,
ALTER DATABASE RENAME, ALTER ROLE RENAME, and ALTER TABLESPACE RENAME.
Also add a pg_upgrade check to detect these invalid names in clusters
being upgraded from pre-v19 versions, producing a report file listing
any offending objects that must be renamed before upgrading.
Tests have been modified accordingly.
Author: Mahendra Singh Thalor <mahi6run@gmail.com> Reviewed-By: Álvaro Herrera <alvherre@alvh.no-ip.org> Reviewed-By: Andrew Dunstan <andrew@dunslane.net> Reviewed-By: Tom Lane <tgl@sss.pgh.pa.us> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Srinath Reddy <srinath2133@gmail.com>
Discussion: https://postgr.es/m/CAKYtNApkOi4FY0S7+3jpTqnHVyyZ6Tbzhtbah-NBbY-mGsiKAQ@mail.gmail.com
Peter Eisentraut [Mon, 23 Feb 2026 15:25:54 +0000 (16:25 +0100)]
meson: allow disabling building/installation of static libraries.
We now support the common meson option -Ddefault_library, with values
'both' (the default), 'shared' (install only shared libraries), and
'static' (install only static libraries). The 'static' choice doesn't
actually work, since psql and other programs insist on linking to the
shared version of libpq, but it's there pro-forma. It could be built
out if we really wanted, but since we have never supported the
equivalent in the autoconf build system, there doesn't appear to be an
urgent need.
With an eye to re-supporting AIX, the internal implementation
distinguishes whether to install libpgport.a and other static-only
libraries from whether to build/install the static variant of
libraries that we can build both ways. This detail isn't exposed as a
meson option, though it could be if there's demand.
The Cirrus CI task SanityCheck now uses -Ddefault_library=shared to
save a little bit of build time (and to test this option).
Author: Peter Eisentraut <peter@eisentraut.org> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/e8aa97db-872b-4087-b073-f296baae948d@eisentraut.org
Nathan Bossart [Mon, 23 Feb 2026 15:26:00 +0000 (09:26 -0600)]
Make use of pg_popcount() in more places.
This replaces some loops over word-length popcount functions with
calls to pg_popcount(). Since pg_popcount() may use a function
pointer for inputs with sizes >= a Bitmapset word, this produces a
small regression for the common one-word case in bms_num_members().
To deal with that, this commit adds an inlined fast-path for that
case. This fast-path could arguably go in pg_popcount() itself
(with an appropriate alignment check), but that is left for future
study.
Suggested-by: John Naylor <johncnaylorls@gmail.com> Reviewed-by: John Naylor <johncnaylorls@gmail.com>
Discussion: https://postgr.es/m/CANWCAZY7R%2Biy%2Br9YM_sySNydHzNqUirx1xk0tB3ej5HO62GdgQ%40mail.gmail.com
Nathan Bossart [Mon, 23 Feb 2026 15:26:00 +0000 (09:26 -0600)]
Remove uses of popcount builtins.
This commit replaces the implementations of pg_popcount{32,64} with
branchless ones in plain C. While these new implementations do not
make use of more sophisticated population count instructions
available on some CPUs, testing indicates they perform well,
especially now that they are inlined. Newer versions of popular
compilers will automatically replace these with special
instructions if possible, anyway. A follow-up commit will replace
various loops over these functions with calls to pg_popcount(),
leaving us little reason to worry about micro-optimizing them
further.
Since this commit removes the only uses of the popcount builtins,
we can also remove the corresponding configuration checks.
Suggested-by: John Naylor <johncnaylorls@gmail.com> Reviewed-by: John Naylor <johncnaylorls@gmail.com>
Discussion: https://postgr.es/m/CANWCAZY7R%2Biy%2Br9YM_sySNydHzNqUirx1xk0tB3ej5HO62GdgQ%40mail.gmail.com
John Naylor [Mon, 23 Feb 2026 12:19:49 +0000 (19:19 +0700)]
Rename pg_crc32c_sse42_choose.c for general purpose
Future commits will consolidate the CPU feature detection functionality
now scattered around in various files, and the CRC "*_choose.c"
files seem to be the natural place for it. For now, just rename in
a separate commit to make it easier to follow the git log. Do the
minimum necessary to keep the build systems functional, and build the
new file pg_cpu_x86.c unconditionally using guards to control the
visibility of its contents, following the model of some more recent
files in src/port.
Limit scope to x86 to reduce the number of moving parts, since the
motivation for doing this now is to clear out some technical debt
before adding AVX2 detection. Arm is left for future work.
Peter Eisentraut [Mon, 23 Feb 2026 09:56:54 +0000 (10:56 +0100)]
Change error message for sequence validate_relation_kind()
We can just say "... is not a sequence" instead of the more
complicated variant from before, which was probably copied from
src/backend/access/table/table.c.
Peter Eisentraut [Mon, 23 Feb 2026 09:37:38 +0000 (10:37 +0100)]
meson: Refactor libpq targets variables
Some of the knowledge of the libpq targets was spread around between
the top-level meson.build and src/interfaces/libpq*. This change
organizes it more like other targets by having a libpq_targets
variable that different subdirectories can add to.
Peter Eisentraut [Mon, 23 Feb 2026 06:37:50 +0000 (07:37 +0100)]
Enable -Wimplicit-fallthrough option for clang
On clang, -Wimplicit-fallthrough requires annotations with attributes,
but on gcc, -Wimplicit-fallthrough is the same as
-Wimplicit-fallthrough=3, which allows annotations with comments. In
order to enforce consistent annotations with attributes on both
compilers, we test first for -Wimplicit-fallthrough=5, which will
succeed on gcc, and if that is not found we test for
-Wimplicit-fallthrough.
Peter Eisentraut [Mon, 23 Feb 2026 06:37:50 +0000 (07:37 +0100)]
Fix additional fallthrough warnings from clang
Clang warns if falling through to a case or default label that is
immediately followed by break, but GCC does
not (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91432). (MSVC also
warns about the equivalent code in C++.)
This is in preparation for enabling fallthrough warnings on Clang.
Amit Kapila [Mon, 23 Feb 2026 04:49:05 +0000 (10:19 +0530)]
Avoid including utils/timestamp.h in conflict.h.
conflict.h currently includes utils/timestamp.h despite only requiring
basic timestamp type definitions. This creates unnecessary overhead.
Replace the include with datatype/timestamp.h to provide the necessary
types. This change requires explicitly including utils/timestamp.h in
test_custom_fixed_stats.c, which previously relied on the indirect
inclusion.
Extracted from the larger patch by Andres Freund.
Discussion: https://postgr.es/m/aY-UE-4t7FiYgH3t@alap3.anarazel.de
Michael Paquier [Mon, 23 Feb 2026 04:42:38 +0000 (13:42 +0900)]
doc: Add section "Options" for pg_controldata
Adding this section brings consistency with the pages of other tools,
potentially easing the introduction of new options in the future as
these are now showing in the shape of a list.
Author: Peter Smith <smithpb2250@gmail.com> Reviewed-by: Andreas Karlsson <andreas@proxel.se>
Discussion: https://postgr.es/m/CAHut+PtSF5AW3DHpYA-_muDLms2xBUzHpd545snVj8vFpmsmGg@mail.gmail.com
On common architectures, the PGPROC struct happened to be a multiple
of 64 bytes on PG 18, but it's changed on 'master' since. There was
worry that changing the alignment might hurt performance, due to false
cacheline sharing across elements in the proc array. However, there
was no explicit alignment, so any alignment to cache lines was
accidental. Add explicit alignment to remove worry about false
sharing.
The ordering was pretty random, making it hard to get an overview of
what's in it. Group related fields together, and add comments to act
as separators between the groups.
Michael Paquier [Sun, 22 Feb 2026 06:12:58 +0000 (15:12 +0900)]
doc: Add description of "filename" for pg_walsummary
This command requires an input file (WAL summary file), that has to be
specified without an option name. The shape of the command and how to
use this parameter is implied in its synopsis. However, this page
lacked a description of the parameter. Listing parameters that do
not require an option is a common practice across the docs. See for
example pg_dump, pg_restore, etc.
Author: Peter Smith <smithpb2250@gmail.com>
Discussion: https://postgr.es/m/CAHut+PtbQi8Dw_0upS9dd=Oh9OqfOdAo=0_DOKG=YSRT_a+0Fw@mail.gmail.com
Álvaro Herrera [Sat, 21 Feb 2026 11:22:08 +0000 (12:22 +0100)]
Avoid name collision with NOT NULL constraints
If a CREATE TABLE statement defined a constraint whose name is identical
to the name generated for a NOT NULL constraint, we'd throw an
(unnecessary) unique key violation error on
pg_constraint_conrelid_contypid_conname_index: this can easily be
avoided by choosing a different name for the NOT NULL constraint.
Fix by passing the constraint names already created by
AddRelationNewConstraints() to AddRelationNotNullConstraints(), so that
the latter can avoid name collisions with them.
Bug: #19393
Author: Laurenz Albe <laurenz.albe@cybertec.at> Reported-by: Hüseyin Demir <huseyin.d3r@gmail.com>
Backpatch-through: 18
Discussion: https://postgr.es/m/19393-6a82427485a744cf@postgresql.org
The field was mainly used for the position in a LOCK's wait queue, but
also as the position in a the freelist when the PGPROC entry was not
in use. The reuse saves some memory at the expense of readability,
which seems like a bad tradeoff. If we wanted to make the struct
smaller there's other things we could do, but we're actually just
discussing adding padding to the struct for performance reasons.
Nathan Bossart [Fri, 20 Feb 2026 18:07:27 +0000 (12:07 -0600)]
Speedup COPY FROM with additional function inlining.
Following the example set by commit 58a359e585, we can squeeze out
a little more performance from COPY FROM (FORMAT {text,csv}) by
inlining CopyReadLineText() and passing the is_csv parameter as a
constant. This allows the compiler to emit specialized code with
fewer branches.
This is preparatory work for a proposed follow-up commit that would
further optimize this code with SIMD instructions.
Author: Nazir Bilal Yavuz <byavuz81@gmail.com> Reviewed-by: Ayoub Kazar <ma_kazar@esi.dz> Tested-by: Manni Wood <manni.wood@enterprisedb.com>
Discussion: https://postgr.es/m/CAOzEurSW8cNr6TPKsjrstnPfhf4QyQqB4tnPXGGe8N4e_v7Jig%40mail.gmail.com
Fix expanding 'bounds' in pg_trgm's calc_word_similarity() function
If the 'bounds' array needs to be expanded, because the input contains
more trigrams than the initial guess, the code didn't return the
reallocated array correctly to the caller. That could lead to a crash
in the rare case that the input string becomes longer when it's
lower-cased. The only known instance of that is when an ICU locale is
used with certain single-byte encodings. This was an oversight in
commit 00896ddaf41f.
Richard Guo [Fri, 20 Feb 2026 08:57:53 +0000 (17:57 +0900)]
Fix computation of varnullingrels when translating appendrel Var
When adjust_appendrel_attrs translates a Var referencing a parent
relation into a Var referencing a child relation, it propagates
varnullingrels from the parent Var to the translated Var. Previously,
the code simply overwrote the translated Var's varnullingrels with
those of the parent.
This was incorrect because the translated Var might already possess
nonempty varnullingrels. This happens, for example, when a LATERAL
subquery within a UNION ALL references a Var from the nullable side of
an outer join. In such cases, the translated Var correctly carries
the outer join's relid in its varnullingrels. Overwriting these bits
with the parent Var's set caused the planner to lose track of the fact
that the Var could be nulled by that outer join.
In the reported case, because the underlying column had a NOT NULL
constraint, the planner incorrectly deduced that the Var could never
be NULL and discarded essential IS NOT NULL filters. This led to
incorrect query results where NULL rows were returned instead of being
filtered out.
To fix, use bms_add_members to merge the parent Var's varnullingrels
into the translated Var's existing set, preserving both sources of
nullability.
Back-patch to v16. Although the reported case does not seem to cause
problems in v16, leaving incorrect varnullingrels in the tree seems
like a trap for the unwary.
Bug: #19412 Reported-by: Sergey Shinderuk <s.shinderuk@postgrespro.ru>
Author: Richard Guo <guofenglinux@gmail.com> Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/19412-1d0318089b86859e@postgresql.org
Backpatch-through: 16
Amit Kapila [Fri, 20 Feb 2026 03:56:33 +0000 (09:26 +0530)]
Avoid including worker_internal.h in pgstat.h.
pgstat.h is a widely included header. Including worker_internal.h there is
unnecessary and creates tight coupling. By refactoring
pgstat_report_subscription_error() to fetch the required
LogicalRepWorkerType internally rather than receiving it as an argument,
we can eliminate the need for the internal header.
Reported-by: Andres Freund <andres@anarazel.de>
Author: Nisha Moond <nisha.moond412@gmail.com> Reviewed-by: vignesh C <vignesh21@gmail.com> Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/aY-UE-4t7FiYgH3t@alap3.anarazel.de
Nathan Bossart [Thu, 19 Feb 2026 22:19:41 +0000 (16:19 -0600)]
Remove SpinLockFree() and S_LOCK_FREE().
S_LOCK_FREE() is used by the test program in s_lock.c, but nobody
has voiced concerns about losing some coverage there.
SpinLockFree() appears to have been unused since it was introduced
by commit 499abb0c0f. There was agreement to remove these in 2020,
but it never happened. Since we still have agreement for removal
in 2026, let's do that now.
Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/aZX2oUcKf7IzHnnK%40nathan
Discussion: https://postgr.es/m/20200608225338.m5zho424w6lpwb2d%40alap3.anarazel.de
Robert Haas [Thu, 19 Feb 2026 18:46:10 +0000 (13:46 -0500)]
Fix add_partial_path interaction with disabled_nodes
Commit e22253467942fdb100087787c3e1e3a8620c54b2 adjusted the logic in
add_path() to keep the path list sorted by disabled_nodes and then by
total_cost, but failed to make the corresponding adjustment to
add_partial_path. As a result, add_partial_path might sort the path list
just by total cost, which could lead to later planner misbehavior.
In principle, this should be back-patched to v18, but we are typically
reluctant to back-patch planner fixes for fear of destabilizing working
installations, and it is unclear to me that this has sufficiently
serious consequences to justify an exception, so for now, no back-patch.
Reviewed-by: Richard Guo <guofenglinux@gmail.com>
Discussion: http://postgr.es/m/CAMbWs4-mO3jMK4t_LgcJ+7Eo=NmGgkxettgRaVbJzZvVZ1koMA@mail.gmail.com
Tom Lane [Thu, 19 Feb 2026 16:08:52 +0000 (11:08 -0500)]
Remove no-longer-useful markers in pg_hba.conf.sample.
The source version of pg_hba.conf.sample contains
@remove-line-for-nolocal@ markers that indicate which lines should
be deleted for an installation that doesn't HAVE_UNIX_SOCKETS.
We no longer support that case, and since commit f55808828 all
that initdb is doing is unconditionally removing the markers.
We might as well remove the markers from the source version and
drop the removal code, which is unintelligible now anyway.
This will not of course save any noticeable number of cycles
in initdb, but it might save some confusion for future
developers looking at pg_hba.conf.sample. It also reduces the
number of distinct cases that replace_token() has to support,
possibly allowing some tightening of that function.
This commit allows setting wal_receiver_timeout per subscription
using the CREATE SUBSCRIPTION and ALTER SUBSCRIPTION commands.
The value is stored in the subwalrcvtimeout column of the pg_subscription
catalog.
When set, this value overrides the global wal_receiver_timeout for
the subscription's apply worker. The default is -1, which means the
global setting (from the server configuration, command line, role,
or database) remains in effect.
This feature is useful for configuring different timeout values for
each subscription, especially when connecting to multiple publisher
servers, to improve failure detection.
Bump catalog version.
Author: Fujii Masao <masao.fujii@gmail.com> Reviewed-by: Japin Li <japinli@hotmail.com> Reviewed-by: Chao Li <li.evan.chao@gmail.com>
Discussion: https://postgr.es/m/a1414b64-bf58-43a6-8494-9704975a41e9@oss.nttdata.com
Fujii Masao [Thu, 19 Feb 2026 15:52:43 +0000 (00:52 +0900)]
Make GUC wal_receiver_timeout user-settable.
When multiple subscribers connect to different publisher servers,
it can be useful to set different wal_receiver_timeout values for
each connection to better detect failures. However, previously
this wasn't possible, which limited flexibility in managing subscriptions.
This commit changes wal_receiver_timeout to be user-settable,
allowing different values to be assigned using ALTER ROLE SET for
each subscription owner. This effectively enables per-subscription
configuration.
Author: Fujii Masao <masao.fujii@gmail.com> Reviewed-by: Japin Li <japinli@hotmail.com> Reviewed-by: Chao Li <li.evan.chao@gmail.com>
Discussion: https://postgr.es/m/a1414b64-bf58-43a6-8494-9704975a41e9@oss.nttdata.com
Fujii Masao [Thu, 19 Feb 2026 14:55:12 +0000 (23:55 +0900)]
Log checkpoint request flags in checkpoint completion messages.
Checkpoint completion log messages include more detail than checkpoint
start messages, but previously omitted the checkpoint request flags,
which were only logged at checkpoint start. As a result, users had to
correlate completion messages with earlier start messages to see
the full context.
This commit includes the checkpoint request flags in the checkpoint
completion log message as well. This duplicates some information,
but makes the completion message self-contained and easier to interpret.
Author: Soumya S Murali <soumyamurali.work@gmail.com> Reviewed-by: Michael Banck <mbanck@gmx.net> Reviewed-by: Yuan Li <carol.li2025@outlook.com> Reviewed-by: Fujii Masao <masao.fujii@gmail.com>
Discussion: https://postgr.es/m/CAMtXxw9tPwV=NBv5S9GZXMSKPeKv5f9hRhSjZ8__oLsoS5jcuA@mail.gmail.com
Peter Eisentraut [Thu, 19 Feb 2026 07:41:03 +0000 (08:41 +0100)]
Use fallthrough attribute instead of comment
Instead of using comments to mark fallthrough switch cases, use the
fallthrough attribute. This will (in the future, not here) allow
supporting other compilers besides gcc. The commenting convention is
only supported by gcc, the attribute is supported by clang, and in the
fullness of time the C23 standard attribute would allow supporting
other compilers as well.
Right now, we package the attribute into a macro called
pg_fallthrough. This commit defines that macro and replaces the
existing comments with that macro invocation.
We also raise the level of the gcc -Wimplicit-fallthrough= option from
3 to 5 to enforce the use of the attribute.
Peter Eisentraut [Thu, 19 Feb 2026 07:41:03 +0000 (08:41 +0100)]
Remove useless fallthrough annotation
A fallthrough attribute after the last case is a constraint violation
in C23, and clang warns about it (not about this comment, but if we
changed it to an attribute). Remove it. (There was apparently never
anything after this to fall through to, even in the first commit da07a1e8565.)
Michael Paquier [Thu, 19 Feb 2026 06:59:20 +0000 (15:59 +0900)]
Sanitize some WAL-logging buffer handling in GIN and GiST code
As transam's README documents, the general order of actions recommended
when WAL-logging a buffer is to unlock and unpin buffers after leaving a
critical section. This pattern was not being followed by some code
paths of GIN and GiST, adjusted in this commit, where buffers were
either unlocked or unpinned inside a critical section. Based on my
analysis of each code path updated here, there is no reason to not
follow the recommended unlocking/unpin pattern done outside of a
critical section.
These inconsistencies are rather old, coming mainly from ecaa4708e5dd
and ff301d6e690b. The guidelines in the README predate these commits,
being introduced in 6d61cdec0761.
Tom Lane [Wed, 18 Feb 2026 19:14:44 +0000 (14:14 -0500)]
Simplify creation of built-in functions with default arguments.
Up to now, to create such a function, one had to make a pg_proc.dat
entry and then overwrite it with a CREATE OR REPLACE command in
system_functions.sql. That's error-prone (cf. bug #19409) and
results in leaving dead rows in the initial contents of pg_proc.
Manual maintenance of pg_node_tree strings seems entirely impractical,
and parsing expressions during bootstrap would be extremely difficult
as well. But Andres Freund observed that all the current use-cases
are simple constants, and building a Const node is well within the
capabilities of bootstrap mode. So this patch invents a special case:
if bootstrap mode is asked to ingest a non-null value for
pg_proc.proargdefaults (which would otherwise fail in
pg_node_tree_in), it parses the value as an array literal and then
feeds the element strings to the input functions for the corresponding
parameter types. Then we can build a suitable pg_node_tree string
with just a few more lines of code.
This allows removing all the system_functions.sql entries that are
just there to set up default arguments, replacing them with
proargdefaults fields in pg_proc.dat entries. The old technique
remains available in case someone needs a non-constant default.
The initial contents of pg_proc are demonstrably the same after
this patch, except that (1) json_strip_nulls and jsonb_strip_nulls
now have the correct provolatile setting, as per bug #19409;
(2) pg_terminate_backend, make_interval, and drandom_normal
now have defaults that don't include a type coercion, which is
how they should have been all along.
In passing, remove some unused entries from bootstrap.c's TypInfo[]
array. I had to add some new ones because we'll now need an entry for
each default-possessing system function parameter, but we shouldn't
carry more than we need there; it's just a maintenance gotcha.
Bug: #19409 Reported-by: Lucio Chiessi <lucio.chiessi@trustly.com>
Author: Tom Lane <tgl@sss.pgh.pa.us>
Author: Andrew Dunstan <andrew@dunslane.net> Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/183292bb-4891-4c96-a3ca-e78b5e0e1358@dunslane.net
Discussion: https://postgr.es/m/19409-e16cd2605e59a4af@postgresql.org
Use standard die() handler for SIGTERM in bgworkers
The previous default bgworker_die() signal would exit with elog(FATAL)
directly from the signal handler. That could cause deadlocks or
crashes if the signal handler runs while we're e.g holding a spinlock
or in the middle of a memory allocation.
All the built-in background workers overrode that to use the normal
die() handler and CHECK_FOR_INTERRUPTS(). Let's make that the default
for all background workers. Some extensions relying on the old
behavior might need to adapt, but the new default is much safer and is
the right thing to do for most background workers.
Michael Paquier [Wed, 18 Feb 2026 07:07:13 +0000 (16:07 +0900)]
Force creation of stamp file after libpq library check in meson builds
Previously, if --stamp_file was specified, libpq_check.pl would create a
new stamp file only if none could be found. If there was already a
stamp file, the script would do nothing, leaving the previous stamp file
in place. This logic could cause unnecessary rebuilds because meson
relies on the timestamp of the output files to determine if a rebuild
should happen. In this case, a stamp file generated during an older
check would be kept, but we need a stamp file from the latest moment
where the libpq check has been run, so as correct rebuild decisions can
be taken.
This commit changes libpq_check.pl so as a fresh stamp file is created
each time libpq_check.pl is run, when --stamp_file is specified.
Reported-by: Andres Freund <andres@anarazel.de>
Author: Nazir Bilal Yavuz <byavuz81@gmail.com> Reviewed-by: VASUKI M <vasukim1992002@gmail.com>
Discussion: https://postgr.es/m/CAN55FZ22rrN6gCn7urtmTR=_5z7ArZLUJu-TsMChdXwmRTaquA@mail.gmail.com
Michael Paquier [Wed, 18 Feb 2026 00:58:38 +0000 (09:58 +0900)]
Switch SysCacheIdentifier to a typedef enum
The main purpose of this change is to allow an ABI checker to understand
when the list of SysCacheIdentifier changes, by switching all the
routine declarations that relied on a signed integer for a syscache ID
to this new type. This is going to be useful in the long-term for
versions newer than v19 so as we will be able to check when the list of
values in SysCacheIdentifier is updated in a non-ABI compliant fashion.
Most of the changes of this commit are due to the new definition of
SyscacheCallbackFunction, where a SysCacheIdentifier is now required for
the syscache ID. It is a mechanical change, still slightly invasive.
There are more areas in the tree that could be improved with an ABI
checker in mind; this takes care of only one area.
Reported-by: Tom Lane <tgl@sss.pgh.pa.us>
Author: Andreas Karlsson <andreas@proxel.se> Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/289125.1770913057@sss.pgh.pa.us
Michael Paquier [Wed, 18 Feb 2026 00:25:52 +0000 (09:25 +0900)]
Add concept of invalid value to SysCacheIdentifier
This commit tweaks the generation of the syscache IDs for the enum
SysCacheIdentifier to now include an invalid value, with -1 assigned as
value. The concept of an invalid syscache ID exists when handling
lookups of a ObjectAddress, based on their set of properties in
ObjectPropertyType. -1 is used for the case where an object type has no
option for a syscache lookup.
This has been found as independently useful while discussing a switch of
SysCacheIdentifier to a typedef, as we already have places that want to
know about the concept of an invalid value when dealing with
ObjectAddresses.
Reviewed-by: Andreas Karlsson <andreas@proxel.se>
Discussion: https://postgr.es/m/aZQRnmp9nVjtxAHS@paquier.xyz
Michael Paquier [Tue, 17 Feb 2026 23:47:58 +0000 (08:47 +0900)]
Fix one-off issue with cache ID in objectaddress.c
get_catalog_object_by_oid_extended() has been doing a syscache lookup
when given a cache ID strictly higher than 0, which is wrong because the
first valid value of SysCacheIdentifier is 0.
This issue had no consequences, as the first value assigned in the
enum SysCacheIdentifier is AGGFNOID, which is not used in the object
type properties listed in objectaddress.c. Even if an ID of 0 was
hypotherically given, the code would still work with a less efficient
heap-or-index scan.
Álvaro Herrera [Tue, 17 Feb 2026 15:38:24 +0000 (16:38 +0100)]
Fix memory leak in new GUC check_hook
Commit 38e0190ced71 forgot to pfree() an allocation (freed in other
places of the same function) in only one of several spots in
check_log_min_messages(). Per Coverity. Add that.
While at it, avoid open-coding guc_strdup(). The new coding does a
strlen() that wasn't there before, but I doubt it's measurable.
Previously, SIGINT was treated the same as SIGTERM in walwriter and
walsummarizer. That decision goes back to when the walwriter process
was introduced (commit ad4295728e04), and was later copied to
walsummarizer. It was a pretty arbitrary decision back then, and we
haven't adopted that convention in all the other processes that have
been introduced later.
Summary of how other processes respond to SIGINT:
- Autovacuum launcher: Cancel the current iteration of launching
- bgworker: Ignore (unless connected to a database)
- checkpointer: Request shutdown checkpoint
- bgwriter: Ignore
- pgarch: Ignore
- startup process: Ignore
- walreceiver: Ignore
- IO worker: die()
IO workers are a notable exception in that they exit on SIGINT, and
there's a documented reason for that: IO workers ignore SIGTERM, so
SIGINT provides a way to manually kill them. (They do respond to
SIGUSR2, though, like all the other processes that we don't want to
exit immediately on SIGTERM on operating system shutdown.)
To make this a little more consistent, ignore SIGINT in walwriter and
walsummarizer. They have no "query" to cancel, and they react to
SIGTERM just fine.
Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://www.postgresql.org/message-id/818bafaf-1e77-4c78-8037-d7120878d87c@iki.fi
Peter Eisentraut [Tue, 17 Feb 2026 09:06:39 +0000 (10:06 +0100)]
Test most StaticAssert macros in C++ extensions
Most of the StaticAssert macros already worked in C++ with Clang and
GCC:(the only compilers we're currently testing C++ extension support
for). This adds a regression test for them in our test C++ extension,
so we can safely change their implementation without accidentally
breaking C++.
The only macros that StaticAssert macros that don't work yet are the
StaticAssertVariableIsOfType and StaticAssertVariableIsOfTypeMacro.
These will be added in a follow-on commit.
Peter Eisentraut [Tue, 17 Feb 2026 09:06:32 +0000 (10:06 +0100)]
Test List macros in C++ extensions
All of these macros already work in C++ with Clang and GCC (the only
compilers we're currently testing C++ extension support for). This
adds a regression test for them in our test C++ extension, so we can
safely change their implementation without accidentally breaking C++.
Some of the List macros didn't work in C++ in the past (see commit d5ca15ee5), and this would have caught that.
Thomas Munro [Tue, 17 Feb 2026 00:53:32 +0000 (13:53 +1300)]
Fix test_valid_server_encoding helper function.
Commit c67bef3f325 introduced this test helper function for use by
src/test/regress/sql/encoding.sql, but its logic was incorrect. It
confused an encoding ID for a boolean so it gave the wrong results for
some inputs, and also forgot the usual return macro. The mistake didn't
affect values actually used in the test, so there is no change in
behavior.
Also drop it and another missed function at the end of the test, for
consistency.
Noah Misch [Tue, 17 Feb 2026 02:04:58 +0000 (18:04 -0800)]
Suppress new "may be used uninitialized" warning.
Various buildfarm members, having compilers like gcc 8.5 and 6.3, fail
to deduce that text_substring() variable "E" is initialized if
slice_size!=-1. This suppression approach quiets gcc 8.5; I did not
reproduce the warning elsewhere. Back-patch to v14, like commit 9f4fd119b2cbb9a41ec0c19a8d6ec9b59b92c125.
Reported-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/1157953.1771266105@sss.pgh.pa.us
Backpatch-through: 14
Michael Paquier [Mon, 16 Feb 2026 23:41:26 +0000 (08:41 +0900)]
hstore: Fix NULL pointer dereference with receive function
The receive function of hstore was not able to handle correctly
duplicate key values when a new duplicate links to a NULL value, where a
pfree() could be attempted on a NULL pointer, crashing due to a pointer
dereference.
This problem would happen for a COPY BINARY, when stacking values like
that:
aa => 5
aa => null
The second key/value pair is discarded and pfree() calls are attempted
on its key and its value, leading to a pointer dereference for the value
part as the value is NULL. The first key/value pair takes priority when
a duplicate is found.
Nathan Bossart [Mon, 16 Feb 2026 21:13:06 +0000 (15:13 -0600)]
pg_upgrade: Use COPY for LO metadata for upgrades from < v12.
Before v12, pg_largeobject_metadata was defined WITH OIDS, so
unlike newer versions, the "oid" column was a hidden system column
that pg_dump's getTableAttrs() will not pick up. Thus, for commit 161a3e8b68, we did not bother trying to use COPY for
pg_largeobject_metadata for upgrades from older versions. This
commit removes that restriction by adjusting the query in
getTableAttrs() to pick up the "oid" system column and by teaching
dumpTableData_copy() to use COPY (SELECT ...) for this catalog,
since system columns cannot be used in COPY's column list.
Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/aYzuAz_ITUpd9ZvH%40nathan
Tom Lane [Mon, 16 Feb 2026 20:20:15 +0000 (15:20 -0500)]
Ensure that all three build methods install the same set of files.
syscache_info.h was installed into $installdir/include/server/catalog
if you use a non-VPATH autoconf build, but not if you use a VPATH
build or meson. That happened because the makefiles blindly install
src/include/catalog/*.h, and in a non-VPATH build the generated
header files would be swept up in that. While it's hard to conjure
a reason to need syscache_info.h outside of backend build, it's
also hard to get the makefiles to skip syscache_info.h, so let's
go the other way and install it in the other two cases too.
Another problem, new in v19, was that meson builds install a copy of
src/include/catalog/README, while autoconf builds do not. The issue
here is that that file is new and wasn't added to meson.build's
exclusion list.
While it's clearly a bug if different build methods don't install
the same set of files, I doubt anyone would thank us for changing
the behavior in released branches. Hence, fix in master only.
Author: Tom Lane <tgl@sss.pgh.pa.us> Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/946828.1771185367@sss.pgh.pa.us
doc: Add note to ssl_group config on X25519 and FIPS
The X25519 curve is not allowed when OpenSSL is configured for
FIPS mode, so add a note to the documentation that the default
setting must be altered for such setups.
Author: Daniel Gustafsson <daniel@yesql.se> Reported-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/3521653.1770666093@sss.pgh.pa.us
projects / thirdparty / postgresql.git / log
