Nov - Dec 2021

2 commits
Move into separate file all the SQL queries used in pg_upgrade tests
2 Dec 2021Michael Paquiermaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.2REL_13_STABLE13.6+3 more
LLM analysis, 100% suremodified before commitlight changesfrom patchset #660 of 3 reviewers saw the final version

Matches the test.sh extraction and upgrade_adapt.sql implementation in #66, except for comment edits and removal of the putenv cleanup command. The commit credits Justin Pryzby and Michael Paquier as co-authors.

  • Tom Lanesaw an earlier versionLLM analysis, 98% sureIn #43, proposed sharing a psql cleanup script between test.sh and the buildfarm, which became this change after substantial implementation work.
  • Jacob Championsaw an earlier versionLLM analysis, 99% sureCompared the proposed script with buildfarm cleanup in #47, identifying version-range differences, missing object recreation, and the more resilient OID cleanup.
  • Andrew Dunstansaw an earlier versionLLM analysis, 87% sureEndorsed the shared cleanup-script design in #60, before the later consolidation and version-specific corrections.
Add table to regression tests for binary-compatibility checks in pg_upgrade
18 Nov 2021Michael Paquiermaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.2REL_13_STABLE13.6+1 more
LLM analysis, 100% surelanded as postedfrom patchset #593 of 4 reviewers saw the final version

All three regression-test file diffs match #59; only abbreviated blob identifiers differ. Michael Paquier forwarded the revised patch, while the commit explicitly credits Justin Pryzby.

  • Jacob Championsaw the final versiontrailerIdentified the missing BRIN summary-type exclusions in #45, after which this test received only small exclusions, naming, formatting, and comment changes.
  • Peter Eisentrautsaw an earlier versiontrailerRequested explanatory documentation and commit messages in #35, before the completeness query and covered type families were substantially revised.
  • Tom Lanesaw the final versiontrailerChallenged the completeness query in #43 and subsequently reported the landed version's ACL-quoting failure in #63.
  • Michael Paquiersaw the final versiontrailerReviewed the coverage query and exclusion comments in #55 and #58, prepared #59, and committed that version.

Oct 2021

1 commit
Fix tests of pg_upgrade across different major versions
13 Oct 2021Michael Paquiermaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.1REL_13_STABLE13.5+1 more
LLM analysis, 99% surelanded as postedfrom patchset #550 of 4 reviewers saw the final version

The entire src/bin/pg_upgrade/test.sh diff matches #55, apart from blob-identifier abbreviation. The commit credits Justin Pryzby, Andrew Dunstan and Michael Paquier, including Andrew's buildfarm-derived OID cleanup.

  • Peter Eisentrautsaw an earlier versionLLM analysis, 90% sureRequested documentation and commit messages for the test.sh repairs in #35, before their later extraction and rewrite.
  • Bruce Momjiansaw an earlier versionLLM analysis, 85% sureRequested comments explaining each adjustment in #36 and explicitly examined the revised patch and commit message in #38.
  • Tom Lanesaw an earlier versionLLM analysis, 99% sureObjected in #43 to removing WAL-size and permission-test coverage, which the final repair preserved while reworking other logic.
  • Jacob Championsaw an earlier versionLLM analysis, 97% sureRecommended the buildfarm's catalog-driven OID cleanup in #47, which replaced the hard-coded ALTER TABLE commands in the final repair.

Oct 2019

4 commits
Add missing include to pg_upgrade/version.c
16 Oct 2019Tomas VondraREL9_6_STABLE9.6.16REL9_5_STABLE9.5.20REL9_4_STABLE9.4.25
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Adds catalog/pg_class.h to src/bin/pg_upgrade/version.c without a matching attachment. Tom Lane diagnosed the missing RELKIND definitions in #28, and Tomas Vondra confirmed the fix in #30.

  • Tom Lanesaw the final versionLLM analysis, 100% sureReported the old-branch runtime failure in #27 and identified the missing header in #28, directly supplying the fix.
Improve the check for pg_catalog.unknown data type in pg_upgrade
16 Oct 2019Tomas Vondramaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+3 more
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #231 of 1 reviewer saw the final version

Takes the unknown-check patch from #23, retaining its recursive CTE and removal of standalone composite types from the final scan. Adds explanatory comments and corrects the copied type-name comment.

  • Tom Lanesaw the final versionLLM analysis, 86% sureApproved the shared recursive-check logic and explicitly requested its application to the unknown-type check in #19, with dropped-column handling clarified in #21.
Improve the check for pg_catalog.line data type in pg_upgrade
16 Oct 2019Tomas Vondramaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+6 more
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #23

Takes the line-check patch from the two-patch submission #23, preserving its recursive CTE and relation-kind filtering. Changes are limited to explanatory comments and commit-message wording.

Check for tables with sql_identifier during pg_upgrade
14 Oct 2019Tomas Vondramaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+1 more
LLM analysis, 100% suremodified before commitlight changesfrom patchset #181 of 1 reviewer saw the final version

Uses the recursive sql_identifier check from #18, adding NOT a.attisdropped inside the composite-type recursion and improving the check.c comment. The added predicate is defensive because dropped attributes already have zero atttypid, as demonstrated in #21.

  • Tom Lanesaw the final versiontrailerIdentified missing domain coverage in #14, tested recursive-query planning in #16, and approved the revision with a dropped-column correction in #19.

Credits across 7 commits

Author
Justin Pryzby3
Tomas Vondra3
Michael Paquier2
Andrew Dunstan1
Reviewer
Tom Lane2
Jacob Champion1
Michael Paquier1
Peter Eisentraut1
Reported by
Hans Buschmann1
Committer
Tomas Vondra4
Michael Paquier3