Mar 2025

7 commits
Fix bitmapheapscan incorrect recheck of NULL tuples
24 Mar 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2361 of 2 reviewers saw the final version

Matches #236: resets recheck before emitting empty tuples in heapam_handler.c and adds the bitmapops regression coverage.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 97% sureReported and bisected the wrong-result regression in the landed integration from #223, prompting the repair in #236.
  • Andres Freundsaw the final versionLLM analysis, 99% sureProduced a deterministic reproducer and tested the adopted recheck reset in #234, before the equivalent fix was attached in #236.
Increase default maintenance_io_concurrency to 16
18 Mar 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1Also in: WIP: WAL prefetch (another approach) (not this thread)
LLM analysis, 100% suremodified before commitlight changesfrom patchset #2292 of 2 reviewers saw the final version

The code and documentation match #229; the commit message expands the rationale for equal concurrency defaults and conservative behavior on low-IOPS systems.

  • Jakub Wartaksaw the final versionLLM analysis, 99% sureRaised the inconsistent defaults in #221 and explicitly supported the attached change in #230.
  • Andres Freundsaw the final versionLLM analysis, 97% sureProposed setting both defaults to the same value in #228, directly informing the subsequent attachment.
Remove table AM callback scan_bitmap_next_block
15 Mar 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 99% suremodified before commitlight changesfrom patchset #2230 of 1 reviewer saw the final version

The callback-removal component of #223 matches the committed diff exactly; the message adds the preceding integration commit reference and removes a CI directive.

  • Tomas Vondrasaw an earlier versiontrailerReviewed the callback-removal series and tested its diagnostic variant, but iterator and empty-tuple control-flow revisions followed #188.
BitmapHeapScan uses the read stream API
15 Mar 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 99% suremodified before commitlight changesfrom patchset #2230 of 1 reviewer saw the final version

The read-stream integration component in #223 has identical code changes; the committed message adds an account of regression testing and analysis.

  • Tomas Vondrasaw an earlier versiontrailerBenchmarked multiple storage and access patterns and reviewed #179, before deferred offset extraction and subsequent integration revisions.
Separate TBM[Shared|Private]Iterator and TBMIterateResult
15 Mar 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2230 of 1 reviewer saw the final version

Exactly matches the iterator/result separation component of #223, including tidbitmap.c, GIN callers, and caller-managed TBMIterateResult storage.

  • Tomas Vondrasaw an earlier versiontrailerReviewed the earlier result-separation design in #188, before its fixed offset arrays and iterator return conventions were revised.
Improve read_stream.c advice for dense streams.
15 Mar 2025Thomas Munromaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1Also in: Some read stream improvements (not this thread)
LLM analysis, 99% suremodified before commitmoderate changesfrom patchset #2230 of 1 reviewer saw the final version

Thomas Munro's advice patch forwarded in #223 is the latest matching attachment, posted later on March 13 than #585955 in thread #51076, so attribution here disagrees with that thread's claim under the latest-match rule. In src/backend/storage/aio/read_stream.c, the commit removes no_wait, gates random-jump advice on pinned_buffers, and moves sequential-region completion handling outside the advice-issued branch.

  • Andres Freundsaw an earlier versiontrailerProfiled and tested #179, demonstrating premature sequential-advice suppression with an inline experiment in #192, before the subsequent advice-control redesign.
Increase default effective_io_concurrency to 16
12 Mar 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1Also in: posix_fadvise v22
LLM analysis, 100% suremodified before commitlight changesfrom patchset #2181 of 1 reviewer saw the final version

Implements the effective_io_concurrency default of 16 from #218, with documentation and commit-message wording changes.

  • Andres Freundsaw the final versiontrailerReviewed the default-change documentation in #219, and only light wording edits followed.

Feb 2025

2 commits
Delay extraction of TIDBitmap per page offsets
24 Feb 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 99% suremodified before commitlight changesfrom patchset #2140 of 1 reviewer saw the final version

Matches the delayed-extraction component of #214, with matchNtuples initialization moved forward from the following attached component and small message edits.

  • Thomas Munrosaw an earlier versiontrailerReviewed deferred extraction in #212, after which the extraction API changed through an intermediate structure to an array-plus-capacity interface.
Add lossy indicator to TBMIterateResult
24 Feb 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2142 of 2 reviewers saw the final version

Exactly matches the standalone lossy-indicator component in #214, replacing negative-ntuples tests with explicit TBMIterateResult lossiness.

  • Thomas Munrosaw the final versionLLM analysis, 97% sureExplicitly proposed the lossy Boolean in #212, which the standalone component directly implements.
  • Tomas Vondrasaw the final versionLLM analysis, 87% sureQuestioned encoding lossiness in ntuples in #160, and Melanie agreed in #161 to address it with a separate lossy Boolean.

Jan 2025

2 commits
Move BitmapTableScan per-scan setup into a helper
30 Jan 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1681 of 1 reviewer saw the final version

Extracts BitmapTableScanSetup exactly as in #168, adding a USE_PREFETCH guard around BitmapHeapNext's remaining pstate local and polishing the message.

  • Nazir Bilal Yavuzsaw the final versiontrailerRequested comment-format improvements and explicitly approved the revised helper in #169.
Add and use BitmapHeapScanDescData struct
16 Jan 2025Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 99% suremodified before commitlight changesfrom patchset #1611 of 1 reviewer saw the final version

Uses the inherited bitmap scan descriptor from #161; code differences preserve upstream uint32 field types during rebase, while the message corrects structure names.

  • Tomas Vondrasaw the final versiontrailerEndorsed the inheritance design and requested cast and scan-type checks in #160, followed by light revisions.

Dec 2024

3 commits
Fix bitmap table scan crash on iterator release
19 Dec 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Richard supplied an inline iterator-cleanup guard and reproducer in #163, extending the fix to endscan in #164; Melanie committed guards, an assertion, and regression coverage.

  • Melanie Plagemansaw the final versioncommittercommitted the patch
Bitmap Table Scans use unified TBMIterator
18 Dec 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1611 of 1 reviewer saw the final version

Uses the unified iterator component of #161, adjusting the exhaustion accessor for the newly named union and filling in the predecessor commit reference.

  • Tomas Vondrasaw the final versiontrailerReviewed unified-iterator use in #146 and the revised series in #160, followed only by light changes.
Add common interface for TBMIterators
18 Dec 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1611 of 2 reviewers saw the final version

Matches the unified-iterator abstraction in #161, with a named union for portability, corresponding member accesses, and comment corrections.

  • Tomas Vondrasaw the final versiontrailerReviewed the iterator abstraction through #160, after which changes were limited to portability, naming, and comments.
  • Heikki Linnakangassaw an earlier versiontrailerProposed the common interface and reviewed it in #152, before the exhausted flag and pointer-based representation were replaced.

Oct 2024

2 commits
Make table_scan_bitmap_next_block() async-friendly
25 Oct 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1611 of 4 reviewers saw the final version

Matches the async-control-flow component of #161 except for removing an unnecessary tidbitmap.h include and editing the commit explanation.

  • Tomas Vondrasaw the final versiontrailerReviewed successive control-flow versions and requested the final naming and explanation changes in #160.
  • Andres Freundsaw an earlier versiontrailerReviewed initial iterator control and diagnosed parallel prefetch races in #5 and #58, before later descriptor and callback restructuring.
  • Heikki Linnakangassaw an earlier versiontrailerReviewed callback semantics and parameter documentation in #152, before the revised scan-state union and combined callback changes.
  • Mark Dilgersaw an earlier versiontrailerTested the original series and assessed migration of a proprietary table AM in #2 and #4, before substantial refactoring.
Move EXPLAIN counter increment to heapam_scan_bitmap_next_block
25 Oct 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1612 of 2 reviewers saw the final version

The counter-relocation diff exactly matches its component in #161; the committed subject and message more directly explain the table-AM responsibility.

  • Tomas Vondrasaw the final versiontrailerReviewed the counter changes and requested clearer lossiness handling and formatting in #160.
  • Heikki Linnakangassaw the final versiontrailerApproved the counter patch in #152, with only field/type adaptation and explanatory changes afterward.

May 2024

1 commit
BitmapHeapScan: Remove incorrect assert and reset field
16 May 2024Melanie Plagemanmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surelanded as postedfrom patchset #1413 of 3 reviewers saw the final version

Exactly matches #141, removing the invalid endscan assertion, resetting pending empty tuples on rescan, and adding the join regression test.

  • Tomas Vondrasaw the final versiontrailerReproduced the assertion, tested the fix in #131, and refined test settings in #135 and #137.
  • Michael Paquiersaw the final versiontrailerRequested an explanation of the reset and SQL casing fixes in #133, followed by explanatory and test cleanup.
  • Alvaro Herrerasaw the final versiontrailerIdentified an unnecessary planner setting in #139 and accepted committing the revised fix in #143.

Apr 2024

5 commits
Change BitmapAdjustPrefetchIterator to accept BlockNumber
6 Apr 2024Tomas Vondramaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surelanded as postedfrom patchset #1063 of 3 reviewers saw the final version

Exactly matches the BlockNumber parameter component of #106 in nodeBitmapHeapscan.c.

  • Tomas Vondrasaw the final versiontrailerReviewed the preparatory patches and committed this version, announcing the batch in #107.
  • Andres Freundsaw the final versiontrailerDiscussed the BlockNumber change's table-AM implications in #5, with no subsequent substantive change to this component.
  • Heikki Linnakangassaw the final versiontrailerReviewed the preparatory series in #37 and #66, after which this parameter cleanup remained materially unchanged.
BitmapHeapScan: Use correct recheck flag for skip_fetch
6 Apr 2024Tomas Vondramaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1063 of 3 reviewers saw the final version

The prefetch-recheck diff matches #106; the commit message expands the explanation of changing recheck flags and limited effects with built-in opclasses.

  • Tomas Vondrasaw the final versiontrailerInvestigated changing recheck flags in #102 and committed the corrected prefetch test.
  • Andres Freundsaw the final versiontrailerReviewed the original prefetch correction in #5, characterizing it as an optimization, and its implementation remained unchanged.
  • Tom Lanesaw the final versiontrailerConfirmed in #104 that using the current page's recheck flag had been an oversight and supported the correction.
BitmapHeapScan: Push skip_fetch optimization into table AM
6 Apr 2024Tomas Vondramaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 99% suremodified before commitlight changesfrom patchset #1051 of 4 reviewers saw the final version

The code matches #105, including the empty-tuple assertions, with a rewritten explanation; the later reset assignments in #106 were not adopted in this commit.

  • Andres Freundsaw an earlier versiontrailerReviewed skip-fetch ownership, state names, and flags in #5, before the callback interface and empty-tuple assertions evolved.
  • Heikki Linnakangassaw an earlier versiontrailerReviewed the optimization and proposed the need-tuples flag in #66, before subsequent interface and assertion changes.
  • Tomas Vondrasaw the final versiontrailerProposed the explicit Boolean interface in #100, reviewed its revisions, and committed the patch.
  • Mark Dilgersaw an earlier versiontrailerTested the initial series and evaluated its table-AM changes in #2 and #4, before later interface revisions.
BitmapHeapScan: postpone setting can_skip_fetch
6 Apr 2024Tomas Vondramaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surelanded as postedfrom patchset #1063 of 3 reviewers saw the final version

Exactly matches the can_skip_fetch timing component in #106, moving its assignment from initialization into BitmapHeapNext.

  • Tomas Vondrasaw the final versiontrailerReviewed the preparatory prefix and committed this component, as reported in #107.
  • Andres Freundsaw the final versiontrailerReviewed the original preparatory series in #5, and this assignment relocation underwent no material redesign.
  • Heikki Linnakangassaw the final versiontrailerReviewed the independent preliminary changes in #37 and #66, with only light subsequent edits to this component.
BitmapHeapScan: begin scan after bitmap creation
6 Apr 2024Tomas Vondramaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1052 of 3 reviewers saw the final version

The scan-initialization diff exactly matches #105, the latest submission before this commit; the opening sentence of the commit message was rewritten.

  • Tomas Vondrasaw the final versiontrailerReviewed the initialization change through #103 and committed the final version.
  • Andres Freundsaw an earlier versiontrailerChallenged the two-snapshot design in #5, before snapshot handling was substantially simplified.
  • Heikki Linnakangassaw the final versiontrailerReviewed initialization and resolved redundant snapshot handling in #37, then reviewed the simplified series in #66.

Mar 2024

2 commits
Fix EXPLAIN Bitmap heap scan to count pages with no visible tuples
18 Mar 2024Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1REL_16_STABLE16.3REL_15_STABLE15.7REL_14_STABLE14.12REL_13_STABLE13.15+1 moreAlso in: Why does BitmapPrefetch() skip fetch based on current block recheck flag (not this thread)
LLM analysis, 99% suremodified before commitmoderate changesfrom patchset #501 of 3 reviewers saw the final version

Implements the EXPLAIN component of #50 with identical partition_prune expected-output changes, but directly moves counter increments before skip-fetch instead of introducing a valid temporary.

  • Andres Freundsaw an earlier versionLLM analysis, 94% sureChallenged the original accounting in #5, and #8 explicitly credits his off-list discussion for counting pages without visible tuples.
  • Tomas Vondrasaw an earlier versionLLM analysis, 94% sureTested the preliminary series containing this fix in #59, before the committer simplified its counter-placement logic.
  • Heikki Linnakangassaw the final versioncommittercommitted the patch
Remove redundant snapshot copying from parallel leader to workers
14 Mar 2024Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surelanded as postedfrom patchset #432 of 2 reviewers saw the final version

Exactly matches #43, removing redundant bitmap-worker snapshot serialization and adding active-snapshot assertions.

  • Dilip Kumarsaw the final versiontrailerExamined the original snapshot rationale and the revised assumptions in #38 and #44.
  • Robert Haassaw the final versiontrailerReviewed snapshot equivalence and cursor implications in #39 and #45.

Credits across 24 commits

Author
Melanie Plageman8 LLM20
Heikki LinnakangasLLM1
Richard Guo1
Thomas MunroLLM1
Reviewer
Tomas Vondra14
Andres Freund8
Heikki Linnakangascommitter8
Mark Dilger2
Alvaro Herrera1
Dilip Kumar1
Melanie Plagemancommitter1
Michael Paquier1
Nazir Bilal Yavuz1
Robert Haas1
Thomas Munro1
Tom Lane1
Tested by
Andres Freund2
Melanie Plageman1
Nazir Bilal Yavuz1
Thomas Munro1
Tomas Vondra1
Reported by
Melanie Plageman2
Alexander Lakhin1
Andres Freund1
Richard Guo1
Tomas Vondra1
Committer
Melanie Plageman16
Tomas Vondra5
Heikki Linnakangas2
Thomas Munro1

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