Apr 2025

2 commits
Remove GUC_NOT_IN_SAMPLE from enable_self_join_elimination
7 Apr 2025Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 99% suremodified before commitmoderate changesfrom patchset #2292 of 2 reviewers saw the final version

Retains the postgresql.conf.sample addition from #229 and additionally removes GUC_NOT_IN_SAMPLE in guc_tables.c following #230 and #234.

  • Tom Lanesaw the final versiontrailerIn #234 he argued that this documented setting should behave like the other enable flags, directly motivating the committed flag removal.
  • Alexander Korotkovsaw the final versioncommittercommitted the patch
Put enable_self_join_elimination into postgresql.conf.sample
6 Apr 2025Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2291 of 1 reviewer saw the final version

The single postgresql.conf.sample addition exactly matches #229; the subsequent flag inconsistency report #230 was posted after this commit.

  • Alexander Korotkovsaw the final versioncommittercommitted the patch

Feb 2025

2 commits
Get rid of ojrelid local variable in remove_rel_from_query()
27 Feb 2025Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2201 of 1 reviewer saw the final version

The analyzejoins.c diff exactly matches #220, replacing the ojrelid sentinel with explicit branches and changing remove_rel_from_eclass arguments.

  • Andrei Lepikhovsaw the final versiontrailerHe inspected the patch and reported correct behavior in #217, after which only the commit explanation was added.
Implement Self-Join Elimination
17 Feb 2025Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2132 of 19 reviewers saw the final version

The complete implementation and regression diff matches #213; the commit credits Lepikhov and Kuzmenkov as authors, with Korotkov and Rybakina as coauthors.

  • Tom Lanesaw an earlier versiontrailerHe reviewed planning overhead and cache representation in #6 and #36, then requested greater infrastructure reuse in #167 before substantial further revisions.
  • Robert Haassaw an earlier versiontrailerHe examined the initial optimization's use cases and timing in #3 and #26, and later demanded a systematic replacement audit in #163.
  • Andres Freundsaw an earlier versiontrailerHe discussed concrete ORM/view cases and prepared-plan overhead in #4, #13 and #17, before the implementation was repeatedly redesigned.
  • Simon Riggssaw an earlier versiontrailerHe questioned overhead and proposed cost-sensitive application in #5 and #8, well before the later architecture.
  • Jonathan S. Katzno review foundtrailerHis request for ORM examples in #7 does not establish patch inspection, testing or substantive technical review.
  • David Rowleysaw an earlier versiontrailerHe designed the early elimination stage, reviewed correctness and supplied the qual-splitting alternative in #9, #33, #43 and #57, before substantial later rework.
  • Thomas Munrosaw an earlier versiontrailerHe identified EPQ and row-lock issues and reported an isolation-test failure in #12 and #29, before later removal and bookkeeping redesigns.
  • Konstantin Knizhniksaw an earlier versiontrailerHis off-list review is acknowledged in #31, and he supplied OR-clause and other corrections in #50, #51 and #52.
  • Heikki Linnakangassaw an earlier versiontrailerHe challenged the broadened elimination proof and demonstrated wrong results in #60 and #62, causing that extension to be withdrawn.
  • Hywel Carversaw an earlier versiontrailerHe reported successful testing and examined redundant NULL filters in #66 and #73, before subsequent major changes.
  • Laurenz Albesaw an earlier versiontrailerHe explained the planning implications of removing redundant predicates in #70, influencing the scope discussion before later redesigns.
  • Ronan Dunklausaw an earlier versiontrailerHe benchmarked adverse planning cases and reviewed uniqueness bookkeeping and implementation details in #91 and #95.
  • vignesh Cno review foundtrailerThe activity in #76 is a rebase request and CommitFest status change, without a substantive review.
  • Zhihong Yusaw an earlier versiontrailerHe identified cleanup issues, initialization problems and an ineffective search-limit check in #78, #90, #96 and #99.
  • Greg Starksaw an earlier versiontrailerHe examined equivalence-class inference and worst-case planning overhead in #81 and #83, before later rework.
  • Jaime Casanovasaw an earlier versiontrailerHe reported pgbench and representative-query benchmark results in #82, before subsequent implementation changes.
  • Michał Kłeczekno review foundtrailerHis message #104 asks about status and offers support, but contains no patch review or test results.
  • Alena Rybakinasaw the final versiontrailerShe supplied review fixes and clause-deduplication refactoring, then explicitly approved the final patch in #214; that approval is after the commit but establishes Task 3 coverage.
  • Alexander Korotkovsaw the final versiontrailerHe repeatedly reviewed and revised the implementation, posted #213 and reviewed the committed version as its committer.

May 2024

1 commit
Revert: Remove useless self-joins
6 May 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1reverts 6 commits here
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 2 reviewers saw the final version

No revert attachment exists; #160 requests withdrawal because basic bugs continued late in the release cycle, and #162 accepts that course.

  • Tom Lanesaw an earlier versionLLM analysis, 99% sureHe requested the implemented revert in #160 before its commit; coverage is earlier against the thread's final patchset because substantial redevelopment followed.
  • Robert Haassaw an earlier versionLLM analysis, 99% sureHe supported reverting and explained the inadequate relid audit in #163 before the commit, preceding the later redesigned feature.

Feb 2024

2 commits
Replace lateral references to removed rels in subqueries
23 Feb 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1442 of 2 reviewers saw the final version

Implements #144's sublevels_up-aware replacement; differences are comment wording, SQL layout and test naming, consistent with the minor edits reported in #147.

  • Andrei Lepikhovsaw the final versiontrailerHe reviewed Richard Guo's recursive replacement fix in #146, after which only light edits followed.
  • Alexander Korotkovsaw the final versiontrailerHe reviewed and pushed the patch with minor edits in #147 and is the commit's committer.
Replace relids in lateral subquery parse tree during SJE
20 Feb 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surelanded as postedfrom patchset #1430 of 1 reviewer saw the final version

The analyzejoins.c and regression diffs match #143; the sender is a credited coauthor alongside Alexander Korotkov.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 99% sureHe tested the initial target-list fix in #142, before #143 broadened the implementation to traverse the subquery parse tree.

Jan 2024

3 commits
Forbid SJE with result relation
9 Jan 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

No attached fix matches; #136 reports incorrect UPDATE RETURNING results, and #137 explains why target relations must be excluded because of EPQ and snapshot semantics.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 99% sureHis concrete UPDATE RETURNING counterexample in #136 prompted this unattached correction, before substantial changes leading to the thread's final patchset.
Fix misuse of RelOptInfo.unique_for_rels cache by SJE
8 Jan 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

The unattached fix answers #133 by adding UniqueRelInfo.self_join and requiring stricter outerrelids matching; #135 confirms the correction.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 99% sureHe supplied the wrong-result reproducer in #133 that exposed invalid uniqueness-cache reuse, before the later reworked feature and final patchset.
Replace the relid in some missing fields during SJE
2 Jan 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1312 of 2 reviewers saw the final version

The analyzejoins.c additions exactly match #131; the commit adds an UPDATE regression and adjusts expected plans in join.sql/join.out.

  • Alexander Lakhinsaw the final versionLLM analysis, 100% sureAfter reporting the failure in #130, he tested the attached fix successfully in #132; only regression coverage was added afterward.
  • Alexander Korotkovsaw the final versioncommittercommitted the patch

Oct 2023

1 commit
Remove useless self-joins
25 Oct 2023Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surelanded as postedfrom patchset #1291 of 19 reviewers saw the final version

The complete implementation and regression diff matches #129; although Korotkov sent it, the commit credits Lepikhov and Kuzmenkov as authors.

  • Tom Lanesaw an earlier versiontrailerHe discussed overhead and reviewed declarations, tests and uniqueness-cache representation in #6 and #36, before substantial redesigns.
  • Robert Haassaw an earlier versiontrailerHe examined view-expansion use cases and optimization timing in #3 and #26, before the implementation changed substantially.
  • Andres Freundsaw an earlier versiontrailerHe discussed ORM/view cases and amortized planning cost in #4, #13 and #17, before later architecture changes.
  • Simon Riggssaw an earlier versiontrailerHe challenged overhead and proposed cost-sensitive application in #5 and #8, before the subsequent redesigns.
  • Jonathan S. Katzno review foundtrailerHis request for ORM examples in #7 does not establish a substantive patch review.
  • David Rowleysaw an earlier versiontrailerHe supplied the early-stage design and alternative uniqueness proof and repeatedly reviewed correctness in #9, #33, #43 and #57.
  • Thomas Munrosaw an earlier versiontrailerHis EPQ concerns and failing isolation test in #12 and #29 prompted row-mark safeguards, before later rework.
  • Konstantin Knizhniksaw an earlier versiontrailerHe provided the off-list review acknowledged in #31 and supplied OR-clause, bookkeeping and cleanup fixes in #50, #51 and #52.
  • Heikki Linnakangassaw an earlier versiontrailerHe challenged the expanded proof and demonstrated wrong results in #60 and #62, causing the unsafe extension to be removed.
  • Hywel Carversaw an earlier versiontrailerHe reported testing and reviewed newly introduced NULL filters in #66 and #73, before later revisions.
  • Laurenz Albesaw an earlier versiontrailerHis explanation of redundant predicates' planning role in #70 informed the scope discussion, before subsequent implementation changes.
  • Ronan Dunklausaw an earlier versiontrailerHe benchmarked worst-case planning and reviewed extra-clause bookkeeping and code details in #91 and #95.
  • vignesh Cno review foundtrailerHis message #76 requests rebasing and updates CommitFest status without reviewing the implementation.
  • Zhihong Yusaw an earlier versiontrailerHe supplied concrete cleanup, initialization and search-limit corrections in #78, #90, #96 and #99.
  • Greg Starksaw an earlier versiontrailerHe examined equivalence-class inference and adverse planning overhead in #81 and #83, before subsequent rework.
  • Jaime Casanovasaw an earlier versiontrailerHe reported extensive pgbench and representative-query measurements in #82, before later implementation changes.
  • Michał Kłeczekno review foundtrailerHis status inquiry and offer of support in #104 provide no evidence of review or testing.
  • Alena Rybakinasaw an earlier versiontrailerShe supplied assertions and quick-exit checks in #115 and discussed degenerate cases in #119, before iterative elimination and search-limit removal.
  • Alexander Korotkovsaw the final versiontrailerHe reviewed and polished the patch through #129 and necessarily reviewed the final implementation as committer.

Feb 2019

1 commit
Fix mark-and-restore-skipping test case to not be a self-join.
21 Feb 2019Tom Lanemaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 99% surewritten from the discussion, no matching patch

No preceding attachment contains this separate test repair; #36 says Tom Lane already pushed it, replacing the mark-and-restore test's self-join with a join between distinct tables.

Credits across 12 commits

Author
Andrei Lepikhov4
Alexander Korotkov1 LLM2
Alexander Kuzmenkov2
Tender Wang2
Richard Guo1
Reviewer
Alexander Korotkovcommitter6
Tom Lane3
Alena Rybakina2
Andrei Lepikhov2
Andres Freund2
David Rowley2
Greg Stark2
Heikki Linnakangas2
Hywel Carver2
Jaime Casanova2
Jonathan S. Katz2
Konstantin Knizhnik2
Laurenz Albe2
Michał Kłeczek2
Robert Haas2
Ronan Dunklau2
Simon Riggs2
Thomas Munro2
vignesh C2
Zhihong Yu2
Reported by
Alexander Lakhin4
Tom Lane2
Co author
Alena Rybakina1
Alexander Korotkov1
Committer
Alexander Korotkov11
Tom Lane1

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