Oct 2020

2 commits
Fix wrong data table horizon computation during backend startup.
29 Oct 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Andres diagnosed premature horizon computation before MyDatabaseId initialization in #147 and announced the fix in #148, without an attached patchset. The procarray.c change includes other backends when MyDatabaseId is InvalidOid.

Centralize horizon determination for temp tables, fixing bug due to skew.
29 Oct 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1Also in: More aggressive vacuuming of temporary tables (not this thread)
LLM analysis, 100% suremodified before commitlight changesfrom patchset #144

Matches #144's core changes and isolation tests, adding only a four-line warning about racy permanent-table tests in src/test/isolation/specs/horizons.spec. This October submission is later than the August patch listed for thread #42867, so this thread supplies the latest matching patchset.

Sep 2020

2 commits
pg_surgery: Try to stabilize regression tests.
18 Sep 2020Robert Haasmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surelanded as postedfrom patchset #1212 of 2 reviewers saw the final version

The changes to contrib/pg_surgery/sql/heap_surgery.sql and its expected output exactly match #121. Robert's patch and commit explicitly credit Ashutosh Sharma and himself, following Ashutosh's submission in #118.

  • Tom Lanesaw the final versionLLM analysis, 98% sureHis substantive reviews in #114 and #116 questioned transaction-wrapped test tables and requested documentation, leading to the test-only revisions ending in #121.
  • Robert Haassaw the final versioncommittercommitted the patch
New contrib module, pg_surgery, with heap surgery functions.
10 Sep 2020Robert Haasmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surelanded as postedfrom patchset #1052 of 8 reviewers saw the final version

All twelve changed files match #105, including heap_surgery.c, extension SQL, regression tests, documentation, and the pgindent typedef. Robert confirms committing this version in #109.

  • Andrey M. Borodinsaw an earlier versionLLM analysis, 99% sureHis review in #36 requested skipped-TID notices and graceful offset validation, incorporated in #46 before later substantial code revisions.
  • M. Beena Emersonsaw an earlier versionLLM analysis, 99% sureShe identified unnecessary dirtying and WAL logging for unchanged pages in #39, which Ashutosh addressed in #46 before later logic changes.
  • Masahiko Sawadasaw the final versionLLM analysis, 97% sureHe reviewed tuple freezing, visibility-map correctness, TID processing, and documentation through #100, questioning its final documentation in #104, after which only a source comment changed.
  • Rajkumar Raghuwanshisaw an earlier versionLLM analysis, 98% sureHe demonstrated corruption recovery using #52 in #58 and reported further successful standby and concurrency testing in #78, before substantial later revisions.
  • Asim Praveensaw an earlier versionLLM analysis, 99% sureHis review in #70 proposed checking the heap-handler OID, adopted in #76 before Robert restored the original AM check in #100.
  • Mark Dilgersaw an earlier versionLLM analysis, 99% sureHis explicit review of #97 in #101 identified critical-section reporting, empty-array handling, and a missing sorted-input comment, with moderate restructuring between that version and the committed patch.
  • Alvaro Herrerasaw an earlier versionLLM analysis, 96% sureHe explained in #79 why leaving visibility-map bits set after surgery could cause further corruption, and Ashutosh accepted the correction in #81 and implemented it in #87.
  • Robert Haassaw the final versioncommittercommitted the patch

Credits across 4 commits

Author
Andres Freund2
Ashutosh SharmaLLM2
Reviewer
Robert Haascommitter2
Committer
Andres Freund2
Robert Haas2

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