Apr 2024

6 commits
Update mmgr's README to mention BumpContext
16 Apr 2024David Rowleymaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surelanded as postedfrom patchset #561 of 1 reviewer saw the final version

The src/backend/utils/mmgr/README diff exactly matches #56, including an identical stable patch ID. The commit credits Amul Sul and David Rowley as authors, reflecting Amul's original contribution in #54.

  • Daniel Gustafssonsaw the final versionLLM analysis, 98% sureIn #55, Daniel requested repalloc terminology instead of realloc, which was adopted in the documentation revision #56.
Push dedicated BumpBlocks to the tail of the blocks list
16 Apr 2024David Rowleymaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1Also in: Differential code coverage between 16 and HEAD (not this thread)
LLM analysis, 100% suremodified before commitlight changesfrom patchset #57

The commit matches the BumpAllocLarge list-tail change and sysviews.out expectation in #57, adding only BumpFree diagnostic consistency and BumpGetChunkSpace comment cleanup. This supersedes the from_discussion attribution in thread #49455: that thread's proposals and attachments preceded the matching patch posted here in #57, making this thread the latest patch source.

Fix incorrect KeeperBlock macro in bump.c
7 Apr 2024David Rowleymaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1Also in: pgsql: Use bump memory context for tuplesorts (not this thread)
LLM analysis, 100% surewritten from the discussion, no matching patch4 of 4 reviewers saw the final version

Andres identified the missing MAXALIGN in #47, and Tomas supplied and tested the corrected macro inline in #48. The commit applies that correction and replaces the manual keeper-block calculation in BumpContextCreate with KeeperBlock, as explained in #52.

  • Andres Freundsaw the final versionLLM analysis, 100% sureAndres diagnosed the missing alignment in #47, and the commit explicitly thanks him for investigating and testing.
  • Tomas Vondrasaw the final versionLLM analysis, 100% sureTomas reported the 32-bit failure in #45 and confirmed that the inline MAXALIGN correction fixed it in #48.
  • Tom Lanesaw the final versionLLM analysis, 100% sureTom confirmed that the proposed correction also fixed the mamba buildfarm failure in #49.
  • Daniel Gustafssonsaw the final versionLLM analysis, 86% sureDaniel reported another occurrence of the same BumpCheck assertion before the commit in #50, which Andres identified as the same issue in #51.
Use bump memory context for tuplesorts
7 Apr 2024David Rowleymaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #393 of 4 reviewers saw the final version

Commits the complete tuplesort component of the three-patch series #39, leaving its prerequisite and allocator components to separate commits. Its diff and stable patch ID exactly match that attachment.

  • Nathan Bossartsaw an earlier versiontrailerNathan requested explanation of the repeated tuple-accounting logic in #3; the later revision added the helper macro and a BRIN tuple-accounting case.
  • Matthias van de Meentsaw the final versiontrailerMatthias reviewed tuplesort comments and requested the split in #4, then carried forward the final-equivalent tuplesort patch from #20 in #36.
  • Tomas Vondrasaw the final versiontrailerTomas examined the original context-ID choice in #6 and benchmarked the allocator supporting the unchanged tuplesort integration from #20 in #22.
  • John Naylorsaw the final versiontrailerJohn benchmarked the allocator supplied with #20 in #21, supporting the series whose tuplesort component remained unchanged through landing.
Introduce a bump memory allocator
7 Apr 2024David Rowleymaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #393 of 4 reviewers saw the final version

Commits the allocator component of #39, with the ID prerequisite and tuplesort integration committed separately. Differences within this component are comments, copyright, formatting, declaration placement, and adaptation to the finalized prerequisite.

  • Nathan Bossartsaw an earlier versiontrailerNathan reviewed the early combined patch in #3, before the allocation API and hot/cold allocation paths were reworked in #20.
  • Matthias van de Meentsaw the final versiontrailerMatthias questioned allocation sizing and unsupported operations in #5 and #17, then tested and integrated the final-equivalent allocator from #20 in #36.
  • Tomas Vondrasaw the final versiontrailerTomas reviewed ID safety in #6 and reported successful performance comparisons for #20 in #22, after the allocation-path refactoring.
  • John Naylorsaw the final versiontrailerJohn tested several chunk sizes and reset frequencies in #21 and reported successful use of the allocator in #43, explicitly noting that he had not inspected the final series.
Enlarge bit-space for MemoryContextMethodID
7 Apr 2024David Rowleymaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 99% surepart of a posted patchsetmoderate changesfrom patchset #390 of 4 reviewers saw the final version

Commits the ID-expansion component of #39, with additional BOGUS_MCTX entries for the two reserved glibc IDs identified in #40, plus naming and comment edits. The commit credits Matthias van de Meent and David Rowley as authors.

  • Tom Lanesaw an earlier versionLLM analysis, 99% sureTom accepted the expansion approach and requested performance checks in #35, then withdrew his objections in #37 before the four-bit overlapping-field revision.
  • Andres Freundsaw an earlier versionLLM analysis, 95% sureAndres proposed reclaiming low alignment bits from the chunk-to-block offset in #11, an approach developed into the later ID-expansion patches.
  • Tomas Vondrasaw an earlier versionLLM analysis, 94% sureTomas identified the loss of bogus-pointer detection from consuming reserved IDs in #6 and discussed adding an ID bit in #9.
  • Nathan Bossartsaw an earlier versionLLM analysis, 86% sureNathan endorsed reclaiming aligned-offset bits to preserve the reserved glibc patterns in #3, before the expansion implementation was attached.

Credits across 6 commits

Author
David Rowley1 LLM5
Amul Sul1
Matthias van de Meent1
Reviewer
John Naylor2
Matthias van de Meent2
Nathan Bossart2
Tomas Vondra2
Reported by
Amul Sul1
Melanie Plageman1
Tomas Vondra1
Committer
David Rowley6

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