model-writtenrun 1842 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May - Jun 2025 · 35 entries
- Bruce Momjian opened the thread with a committed PostgreSQL 18 release-note draft containing 209 items and invited corrections through beta and final release in #1.
- Amit Langote asked to separate partition planning from execution improvements in #2, while Bruce explained his preference for user-facing descriptions and substantial performance changes, without a separate partitioning section, in #5.
- Jacob Champion pointed out that the old SSL configuration name remained supported in #3, and Bruce combined that correction with broad editorial reordering in aa82ebdc29.
- Jelte Fennema-Nio requested protocol 3.2 coverage and a missing tracing reference in #7; Bruce attached #8 and committed it unchanged as 9f8fcadb20, while declining a separate pgbench limit entry.
- Jelte's review in #9 moved the protocol explanation from libpq to the server cancellation-key entry through #11 and 04b269da56.
- After jian he identified a missing GROUP BY optimization in #10, Bruce adapted wording from older release notes in #12, committed unchanged as 24987c6f06.
- Jelte's further review in #13 produced the client typo, libpq terminology, and temporal-constraint link fixes in a675149e87, but the suggested relocation of search_path reporting was not settled.
- Alexander Borisov requested recognition of Unicode case-conversion performance improvements in #14; Jelte and Jonathan S. Katz argued that performance gains matter to users in #21 and #26, while Bruce preferred grouping the work with an existing item rather than adding a separate entry.
- Jelte requested additional protocol references in #22, leading to 8c9eec540d; Bruce rejected the requested timing-safe comparison reference as too remote from the entry in #28.
- Alvaro Herrera's constraint review in #23 led to reordered entries and an explicit NOT NULL restriction in #29, committed unchanged as d83981c24b, while Bruce declined merging the entries into one.
- Atsushi Torikoshi's detailed COPY review in #24 produced #30 and b3754dcc9f, correcting COPY FROM terminology, REJECT_LIMIT availability, logging scope, and capitalization.
- David Rowley supported separating partition planning and locking in #31, and Bruce implemented the split in #36 and cf847d6340, although both descriptions still needed correction.
- David supplied missing hashing coverage and measured memory savings in #32, which Bruce incorporated through #37 into 092e72a930.
- Bertrand Drouvot proposed consistent reset-function coverage in #33, and Bruce chose separate I/O and WAL explanations in #38, committed as 0de2e1c8b5.
- Jacob's OAuth terminology review in #39 became #40 and c0e6aace02, clarifying the authentication method and token-validation libraries.
- David and Amit corrected the partition descriptions in #44 and #45; Bruce applied #48 as 45750c6cfe and separately moved locking to General Performance in 575f6003ed.
- Jian he's review of jsonb null and dump options in #46 was applied through #51 as ada78f9bef.
- David showed that the partition planning improvement also applies without joins in #52, prompting #53 and b560ce7884; he approved the resulting wording afterward in #54.
- Richard Guo requested mention of longstanding GROUPING SETS correctness fixes in #55, leading to #57 and 3bd5271729.
- Yugo Nagata corrected the scope of pgbench reporting in #56, and #58 landed as 69aca072eb, also completing missing links for the GROUPING SETS entry.
- David identified omitted pg_log_backend_memory_contexts() coverage in #59, producing #60 and f8d49aa130; his follow-up review in #61 produced the attribution and parentheses fixes in #62 and 9fef27a83b.
- Jian he's clearer RETURNING explanation in #65 was adapted in #67 and committed unchanged from that attachment as 9d710a1ac0.
- Richard questioned grouping correctness fixes under an optimization entry and supplied more precise descriptions in #66, #69, and #70, but Bruce chose to retain the existing presentation pending beta feedback in #72.
- Steven Niu's crc32c parentheses report in #71 became 89372d0aaa, and David's placement correction in #74 moved the hashing entry to General Performance in 99ddf8615c.
- Discussion of removed catalog and monitoring columns began in #76; after feedback from Laurenz Albe and Alvaro, Bruce explained in #80 that user-facing renames and removals normally belong in incompatibilities, with exceptions where replacement features are clearer together.
- Nazir Bilal Yavuz questioned file_copy_method attribution in #81, and Bruce's explanation of Co-authored-by conventions in #82 led to a broader discussion of commit-message guidance and recording corrections, without a corresponding candidate release-note change.
- Peter Geoghegan explained the distinction between skip scan and loose index scan in #85, leading to #86 and 7ddfac79f2; his follow-up in #95 prompted the equality-reference wording in #97 and 44ce4e1593.
- Vignesh corrected the CREATE SUBSCRIPTION attribution in #88, and Bruce applied #89 as 883339c170.
- Michael Paquier supplied the injection-point runtime-argument entry in #90; Bruce preserved it in the broader update #96, committed as 1ca583f6c0 with Michael credited as an author.
- Peter Smith's markup report in #100 became 08b8aa1748, and Nathan Bossart's inline MD5 removal warning in #101 became 3e782ca322 with Nathan credited as author.
- Dean Rasheed clarified numeric arithmetic and added missing references in #105, which landed unchanged as c861092b0e.
- Yugo separated ANALYZE reporting from EXPLAIN in #106, then changed the destination to Monitoring in #107; the latter version landed as bf6034d00d.
- Melanie Plageman identified distinct connection-duration logging and missing asynchronous-I/O examples in #112, producing #113 and a1de1b0833; Bruce declined an additional insert-triggered vacuum entry because he could not explain it usefully as a release-note feature.
- Melanie's follow-up in #116 made the I/O examples explicitly non-exhaustive through #117 and 03c53a7314, while the question of listing individual read-stream changes continued in #131 and #132.
- Noah Misch requested an AFTER-trigger role-change incompatibility entry in #118; discussion with David G. Johnston and Tom Lane clarified event queuing, execution timing, and SECURITY DEFINER behavior, leading to Bruce's inline wording in #129 and 73e26cbeb5.
Aug - Sep 2025 · 6 entries
- Amit Kapila requested coverage of changed missing-publication behavior in #133, but Bruce questioned whether the uncommon bug fix warranted an entry and requested another opinion in #134; no resolution is shown.
- Jelte renewed the Unicode performance request in #135, but Bruce declined to revisit the earlier decision in #137.
- Yugo supplied the omitted temporal foreign-key and PERIOD coverage in #136, and Bruce applied it unchanged as dad1e259df, acknowledging the omission in #138.
- Peter reported that the skip-scan wording had caused a published misunderstanding about inequality support in #139; detailed discussion in #140 and #144 led Bruce to the broader explanation in #145.
- Daniel Gustafsson proposed delaying committed release notes until RC1 in #141, but Bruce defended early publication as useful both to users and to finding mistakes in #142 and #143.
- The thread ends with Bruce announcing the final skip-scan correction in #146, corresponding to b4879afdf5; the attachment body is unavailable, all represented patch work has landed, and the declined or unresolved inclusion questions remain as described.