Sep 2025

1 commit
Replace tests of ALTER DATABASE RESET TABLESPACE.
12 Sep 2025Noah Mischmaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surelanded as postedfrom patchset #781 of 1 reviewer saw the final version

Applying #78 to the commit parent reproduces database.sql and database.out exactly, including SET TABLESPACE pg_default and the rolled-back database rename.

  • Tom Lanesaw the final versiontrailerReported the ineffective commands in #77 and explicitly endorsed the replacement patch in #79.

Mar 2025

1 commit
meson: Flush stdout in testwrap
19 Mar 2025Andres Freundmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17.5REL_16_STABLE16.9
LLM analysis, 100% surelanded as postedfrom patchset #741 of 1 reviewer saw the final version

The one-line sys.stdout.flush() addition in src/tools/testwrap exactly matches #74.

  • Noah Mischsaw the final versiontrailerAccepted the flushing fix in #75 before Andres Freund committed it.

Oct 2024

1 commit
Stop reading uninitialized memory in heap_inplace_lock().
24 Oct 2024Noah Mischmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17.1REL_16_STABLE16.5REL_15_STABLE15.9REL_14_STABLE14.14REL_13_STABLE13.17+1 more
LLM analysis, 100% surelanded as postedfrom patchset #720 of 1 reviewer saw the final version

The heap_inplace_lock change and added isolation permutation reproduce #72 exactly when applied to the commit parent.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 94% sureExposed the committed successor-locking code's uninitialized read with Valgrind in #71, leading to the new fix in #72.

Sep 2024

5 commits
Fix use of uninitialized value in previous commit.
25 Sep 2024Noah MischREL_16_STABLE16.5REL_15_STABLE15.9
LLM analysis, 100% surewritten from the discussion, no matching patch

Following the buildfarm failure reported in #70, changes ExecMergeMatched's early exit from goto out to return no_further_action; no indexed attachment contains this correction.

Back-patch "Refactor code in tablecmds.c to check and process tablespace moves"
24 Sep 2024Noah MischREL_13_STABLE13.17REL_12_STABLE12.21
LLM analysis, 99% surewritten from the discussion, no matching patch

This stable-branch prerequisite implements the tablespace-refactoring back-patch explicitly described in #69, avoiding tuple locks held throughout table_relation_copy_data. It appears in no indexed patchset.

Warn if LOCKTAG_TUPLE is held at commit, under debug_assertions.
24 Sep 2024Noah Mischmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17.1REL_16_STABLE16.5REL_15_STABLE15.9REL_14_STABLE14.14REL_13_STABLE13.17+1 more
LLM analysis, 100% surelanded as postedfrom patchset #631 of 1 reviewer saw the final version

The LOCKTAG_TUPLE warning and explanatory comment in lock.c exactly match the corresponding component of #63.

  • Heikki Linnakangassaw the final versionLLM analysis, 100% sureAccepted the warning and supplied its explanatory comment in #54; only that light addition followed.
Fix data loss at inplace update after heap_update().
24 Sep 2024Noah Mischmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17.1REL_16_STABLE16.5REL_15_STABLE15.9REL_14_STABLE14.14REL_13_STABLE13.17+1 more
LLM analysis, 94% surelanded as postedfrom patchset #631 of 5 reviewers saw the final version

Matches the successor-update component of #63 apart from a minimal rebase adaptation in genam.c: the redundant mutable scan-key copy is removed because the commit parent's systable_beginscan already copies keys.

  • Nitin Motianisaw the final versionLLM analysis, 99% sureReviewed the heapam/genam interface, proposed the adopted update-and-unlock name, and approved the revised patch in #64.
  • Heikki Linnakangassaw an earlier versionLLM analysis, 100% sureRequested moving the scan interface to genam.c in #54, after which the wrappers and lock reacquisition behavior changed.
  • Alexander Lakhinsaw an earlier versionLLM analysis, 100% sureFound the ANALYZE failure in #43 and confirmed the later fix series stabilized the database isolation test in #49.
  • Robert Haassaw an earlier versionLLM analysis, 91% sureReported cursory review of the final two patches in #19, before the successor-update API was reorganized.
  • Tom Lanesaw an earlier versionLLM analysis, 88% sureIdentified the wider inplace-update hazards in #8 and proposed the existing tuple-lock infrastructure in #10.
For inplace update durability, make heap_update() callers wait.
24 Sep 2024Noah Mischmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17.1REL_16_STABLE16.5REL_15_STABLE15.9REL_14_STABLE14.14REL_13_STABLE13.17+1 more
LLM analysis, 99% surelanded as postedfrom patchset #631 of 4 reviewers saw the final version

Matches the heap_update-caller locking component of #63, including SearchSysCacheLocked1 and executor locking, modulo rebase context and commit-message updates. The separate optional index-locking component was not included.

  • Nitin Motianisaw the final versionLLM analysis, 100% sureProposed the ResultRelInfo assertion in #62, reviewed its implementation in #64, and concluded his review in #66.
  • Heikki Linnakangassaw an earlier versionLLM analysis, 99% sureReviewed the locking rules in #54 and proposed clearer rules in #56 before subsequent assertion and interface changes.
  • Robert Haassaw an earlier versionLLM analysis, 91% sureExplicitly reported cursory review of the final two locking patches in #19, before later executor and interface revisions.
  • Tom Lanesaw an earlier versionLLM analysis, 88% sureProposed reusing existing tuple locks instead of new locktags in #10, a design suggestion adopted by the later series.

Jul 2024

2 commits
Fix new assertion for MERGE view_name ... DO NOTHING.
13 Jul 2024Noah Mischmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17beta3
LLM analysis, 100% surelanded as postedfrom patchset #470 of 1 reviewer saw the final version

Exactly matches the MERGE view DO NOTHING assertion correction and regression tests attached in #47.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 97% sureDemonstrated that the earlier committed executor assertion fails for MERGE on a view with only DO NOTHING actions in #46.
Don't lose partitioned table reltuples=0 after relhassubclass=f.
13 Jul 2024Noah Mischmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13
LLM analysis, 100% surelanded as postedfrom patchset #470 of 1 reviewer saw the final version

Applying the ANALYZE component of #47 reproduces analyze.c and both vacuum regression files exactly.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 97% sureReported the partitioned-table ANALYZE failure against #42 in #43, prompting the separately committed command-counter correction.

Jun 2024

11 commits
Remove configuration-dependent output from new inplace-inval test.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13REL_13_STABLE13.16+1 more
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Removes the configuration-dependent final read from one inplace-inval permutation, implementing option one proposed in #40 and endorsed in #41; no matching attachment was posted.

  • Tom Lanesaw the final versionLLM analysis, 100% sureSelected the implemented test-neutralization strategy in #41 and is credited for that strategy review in the commit message.
Cope with inplace update making catcache stale during TOAST fetch.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13REL_13_STABLE13.16+1 moreAlso in: Re: Recovering from detoast-related catcache invalidations (not this thread)
LLM analysis, 100% suremodified before commitlight changesfrom patchset #331 of 1 reviewer saw the final version

The catcache component of #33 matches the commit's changes in catalog.c, catcache.c and catalog.h, except for an added catalog/catalog.h include in catcache.c. This disagrees with the from_discussion attribution in thread #49022: this thread supplies a matching June attachment, while that thread's pre-commit attachments date from January and its remaining attachments postdate the commit.

  • Robert Haassaw the final versionLLM analysis, 100% sureRequested clearer TOAST-staleness rationale and code-linked comments in #19, which Noah adopted in #23; subsequent changes were comments, commit text and an include, and the commit explicitly credits his review.
Add wait event type "InjectionPoint", a custom type like "Extension".
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3
LLM analysis, 100% surelanded as postedfrom patchset #391 of 2 reviewers saw the final version

Applying the wait-event component of #39 reproduces all committed files exactly, including the full-information hash key from #37.

  • Michael Paquiersaw the final versionLLM analysis, 100% sureReviewed the new class in #34 and endorsed the full-information hash-key correction in #38, followed only by light test adjustments.
  • Robert Haassaw an earlier versionLLM analysis, 97% sureProposed the dedicated InjectionPoint class in #19 and #21, which Noah implemented after the initial naming-based design.
Create waitfuncs.c for pg_isolation_test_session_is_blocked().
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3
LLM analysis, 100% surelanded as postedfrom patchset #331 of 1 reviewer saw the final version

Exactly matches the waitfuncs.c extraction in #33, including the build-file changes and removal from lockfuncs.c.

  • Robert Haassaw the final versionLLM analysis, 99% sureThe commit explicitly credits Robert Haas as reviewer, although #19 contains no comments specific to the unchanged function extraction.
Add an injection_points isolation test suite.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3
LLM analysis, 100% surelanded as postedfrom patchset #330 of 2 reviewers saw the final version

Exactly matches the injection_points isolation-suite component of #33, including detection through the InjectionPoint wait-event type.

  • Robert Haassaw an earlier versionLLM analysis, 100% sureObjected to recognizing injection waits by name in #19, prompting the subsequent switch to a dedicated wait-event class.
  • Michael Paquiersaw an earlier versionLLM analysis, 100% sureReviewed the blocking-detection interface in #14 and #24 before its implementation changed from name-prefix matching to class detection.
Lock owned sequences during ALTER TABLE SET { LOGGED | UNLOGGED }.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8
LLM analysis, 100% surelanded as postedfrom patchset #331 of 1 reviewer saw the final version

The SequenceChangePersistence lock addition exactly matches its separated component in #33.

  • Robert Haassaw the final versionLLM analysis, 98% sureRequested concrete consequences for the missing-lock fixes in #19, after which this unchanged sequence-lock addition gained its own explanation.
Improve test coverage for changes to inplace-updated catalogs.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13REL_13_STABLE13.16+1 more
LLM analysis, 100% suremodified before commitlight changesfrom patchset #391 of 1 reviewer saw the final version

Matches the tests component of #39, with database.sql and its expected output changed to explicit LC_COLLATE/LC_CTYPE settings and a large datacl value instead of datcollversion for TOAST coverage.

  • Robert Haassaw the final versionLLM analysis, 90% sureThe commit credits his review of the earlier series discussed in #19; subsequent changes to this component consist of test additions and adjustments.
Expand comments and add an assertion in nodeModifyTable.c.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13REL_13_STABLE13.16+1 more
LLM analysis, 100% surelanded as postedfrom patchset #331 of 1 reviewer saw the final version

Exactly matches the nodeModifyTable.c comments-and-assertion component of #33.

  • Robert Haassaw the final versionLLM analysis, 100% sureRequested an explanation distinguishing automatically updatable views from INSTEAD OF-trigger views in #19, followed only by comment changes.
Lock before setting relhassubclass on RELKIND_PARTITIONED_INDEX.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13REL_13_STABLE13.16+1 more
LLM analysis, 100% surelanded as postedfrom patchset #330 of 1 reviewer saw the final version

Exactly matches the partitioned-index locking component of #33, including SetRelationHasSubclass assertions and LockHeldByMe's stronger-lock support.

  • Robert Haassaw an earlier versionLLM analysis, 100% sureRequested enforcement of SetRelationHasSubclass's locking contract in #19, leading to new assertions and supporting lock-check changes.
AccessExclusiveLock new relations just after assigning the OID.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13REL_13_STABLE13.16+1 more
LLM analysis, 100% surelanded as postedfrom patchset #330 of 1 reviewer saw the final version

The heap.c change exactly matches #33, taking AccessExclusiveLock immediately after allocating the relation OID.

  • Robert Haassaw an earlier versionLLM analysis, 100% sureQuestioned the late heap_create lock in #19, after which Noah moved acquisition to the earlier OID-allocation layer.
Make TAP todo_start effects the same under Meson and prove_check.
28 Jun 2024Noah Mischmaster17beta3REL_19_STABLE17beta3REL_18_STABLE17beta3REL_17_STABLE17beta3REL_16_STABLE16.4
LLM analysis, 100% surelanded as postedfrom patchset #331 of 1 reviewer saw the final version

Applying the testwrap component of #33 reproduces the committed file exactly, rewriting successful TODO markers before forwarding output to Meson.

  • Robert Haassaw the final versionLLM analysis, 99% sureThe commit explicitly credits Robert Haas as reviewer, although his series review in #19 contains no testwrap-specific comments.

Credits across 21 commits

Author
Noah MischLLM17
Andres FreundLLM1
Reviewer
Noah Misch1
Tom Lane1
Reported by
Tom Lane1
Committer
Noah Misch20
Andres Freund1

N LLM means N of the credits come from LLM analysis, not from commit trailers.