Jan 2026

1 commit
pg_dump: Fix gathering of sequence information.
9 Jan 2026Nathan Bossartmaster19beta1REL_19_STABLE19beta1REL_18_STABLE18.2Also in: BUG #19365: postgres 18 pg_dump fails whan drop sequence concurrently (not this thread)
LLM analysis, 100% suremodified before commitlight changesfrom patchset #271 of 1 reviewer saw the final version

Matches the backend relation-opening changes and pg_dump NULL tracking and schema-only guard in #27. Beyond base-context differences, the commit only expands the diagnostic to specify SELECT privilege and concurrent drops.

  • Tom Lanesaw the final versiontrailerRead #25, requested NULL results for non-sequence relations in #26, and accepted the revised approach in #28.

Jul 2024

4 commits
Improve performance of dumpSequence().
31 Jul 2024Nathan Bossartmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #170 of 1 reviewer saw the final version

Lands the second component of #17, introducing the sorted SequenceItem cache and bsearch-based metadata lookup. The only substantive textual difference from that component is a rewritten comment about pre-v10 servers.

  • Euler Taveirasaw an earlier versiontrailerTested metadata caching and requested the pre-v10 guard in #3, before the enum conversion and direct SequenceItem use in #16.
Introduce pg_sequence_read_tuple().
31 Jul 2024Nathan Bossartmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #170 of 2 reviewers saw the final version

Lands the third component of #17, including pg_sequence_read_tuple(), its pg_proc entry, and regression tests. Commit-time edits initialize the Datum and null arrays at declaration, remove the redundant memset, and add the promised catalog-version bump.

  • Michael Paquiersaw an earlier versiontrailerSupported returning the actual sequence value and is_called together in #8, before the separate pg_sequence_read_tuple() implementation appeared.
  • Tom Lanesaw an earlier versiontrailerRead the new function and identified its overly permissive ACL_SELECT-or-ACL_USAGE check in #15, which was corrected after his review.
Improve performance of dumpSequenceData().
31 Jul 2024Nathan Bossartmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #170 of 3 reviewers saw the final version

Lands the fourth component of #17, caching last_value and is_called and using them in dumpSequenceData(). Its changed lines match that component; differences are surrounding comment context from the preceding commit.

  • Euler Taveirasaw an earlier versiontrailerBenchmarked the series with and without sequence-data caching and discussed compatibility in #3, before the pg_sequence_last_value() approach was replaced.
  • Michael Paquiersaw an earlier versiontrailerConfirmed the is_called problem and advocated returning both the value and call state in #8, before the new backend function and revised caching query were attached.
  • Tom Lanesaw an earlier versiontrailerReviewed the tuple-reading approach and required SELECT-only privileges in #15, with that behavioral correction incorporated afterward in #16.
Parse sequence type and integer metadata in dumpSequence().
31 Jul 2024Nathan Bossartmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #170 of 1 reviewer saw the final version

Lands the first component of #17, parsing integer metadata and representing sequence types with SeqType. The only added code-text difference is the comment introducing the sequence-type enum.

  • Euler Taveirasaw an earlier versiontrailerQuestioned integer conversion and reviewed string-buffer handling in #3, before strings were replaced with the SeqType enum in #16.

Credits across 5 commits

Author
Nathan BossartLLM5
Reviewer
Euler Taveira3
Tom Lane3
Michael Paquier2
Reported by
Paveł Tyślacki1
Committer
Nathan Bossart5

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