model-writtenrun 8131 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May 2020 · 19 entries
- Bruce Momjian opened the thread with an already-committed first draft of PostgreSQL 13's release notes, inviting corrections while markup and formatting were still unfinished #1.
- Thomas Munro identified an unshipped recovery-prefetch optimization, misleading xid8 naming, and overly broad collation wording; Bruce removed the premature feature claim and corrected the descriptions #4 #12.
- Contributors corrected author credits and omissions, while Justin Pryzby's broad review prompted changes to incremental sorting, partition triggers, WAL thresholds, logging, and vacuumdb descriptions #5 #8 #23 #26 #29 #31.
- Justin Pryzby and Jeff Davis explained that older hash aggregation could exceed work_mem, whereas PostgreSQL 13 could spill to disk; Bruce initially resisted recommending larger work_mem because that would affect other operations #28 #30 #32.
- Fabien COELHO supplied explicit documentation of pgbench's default client-side data generation and removed an empty paragraph; Bruce committed it with a grammatical edit as c3d1fdb598 #47 #54.
- Andrey Borodin reported the missing TOAST decompression improvement, and Bruce added it with contributor credit in the applied patch #55, matching fb544735f1; subsequent replies identified a spelling error #48 #58 #60.
- Amit Langote distinguished publishing partitioned tables, receiving replication into partitioned tables, and publishing via partition roots; Bruce combined two publication entries while retaining a separate subscription entry #51 #62 #63.
- Fabien COELHO advocated recognizing substantial documentation improvements, while Bruce questioned routine presentation changes and Tom Lane initially opposed documenting documentation changes in release notes #61 #64 #70.
- Amit Langote's partition wording patch clarified partition bounds and publication of partition changes, and Bruce applied a lightly revised version #80 #97.
- Alexander Korotkov recommended explaining GiST parameter support through affected data types rather than obscure operator-class names, leading to Bruce's GiST clarification patch #90 #93.
- Michael Paquier's review prompted inclusion of the Windows 2000 support removal, while Bruce excluded routine error-message improvements and documentation-only color-support changes #79 #96.
- Peter Geoghegan argued that B-tree deduplication materially improves low-cardinality workloads, and Bruce clarified duplicate-key storage and the need to REINDEX after pg_upgrade #78 #98 #105 #116.
- Amit Kapila and Justin Pryzby challenged the insert-triggered autovacuum explanation, leading Bruce to describe reduced heap fetches through all-visible pages rather than implying that index-only scans were newly possible #87 #109 #112.
- Alvaro Herrera and Tom Lane supported a separate user-facing Documentation subsection, and Bruce added the glossary and function-table reformatting there; he continued to exclude smaller navigation and formatting improvements #107 #121 #122 #126 #137.
- The WAL-skipping discussion distinguished avoiding per-operation WAL from choosing fsync or WAL at commit, with Noah Misch and Kyotaro Horiguchi explaining the crash-recovery implications; Bruce settled on concise wording mentioning wal_skip_threshold and the previous data-loss bug #81 #139 #143 #144.
- Bruce retained user-visible significance as the criterion for release-note entries and moved the DocBook upgrade into Documentation, while Justin clarified that function-call backtrace support was already listed #145 #147 #150.
- Fujii Masao found a broken sepgsql link while building HTML; Bruce replaced it with an xref, and Fujii confirmed that the correction worked #151 #152 #153.
- Daniel Gustafsson supplied three PostgreSQL spelling corrections, which Bruce applied unchanged as b31d6efe3d #155 #156.
- Daniel Gustafsson identified the bool_plperl entry's unresolved documentation question; Bruce removed it and added a PL/Perl documentation link in d9101e9806 #157 #158.
Jun - Sep 2020 · 10 entries
- Masahiko Sawada proposed relocating the FOREIGN-keyword entry, but Bruce and Alvaro Herrera considered the change too insignificant to list; Bruce instead deleted it and clarified that --include-foreign-data was new, applying #163 unchanged as 1f601b14e3 #159 #160 #161 #162.
- After hash_mem_multiplier became available, Peter Geoghegan requested upgrade guidance and reviewed Bruce's initial patch, arguing for a practical tip beside hash aggregation rather than an incompatibility warning; Bruce expanded the wording but rejected adding GROUP BY terminology #164 #165 #166 #168 #169 #170.
- Bruce committed the revised hash_mem_multiplier guidance unchanged as e7a6cd5dcf, completing that follow-up #170 #173.
- Justin Pryzby returned in September with a broad correction patch and additional wording suggestions, and Amit Langote separately supplied a grammatical correction for partition publication #174 #175.
- Jonathan S. Katz proposed the major-enhancements overview, which Tom Lane landed with light wording edits as 3d92252d7d; Tom deferred the proposed date update until he had audited intervening history #176 #177 #179.
- After intervening editorial work, Justin Pryzby posted an inline remainder of his corrections, adding an example alias; Tom applied it with minor wording changes as 9892564121, incorporating part of the earlier attachment #174 #180 #181.
- Peter Eisentraut spotted the overview's erroneous B-tree restriction on parallel vacuuming, and Masahiko Sawada confirmed its broader applicability; Jonathan supplied a correction that Tom committed with an additional clarification as 001d2c5f15 #182 #183 #184 #185.
- Jonathan S. Katz requested explicit coverage of removed CREATE EXTENSION ... FROM syntax after a PostGIS issue; Tom initially pointed to the existing extension entry, then agreed that the syntax removal should be explicit and moved into Incompatibilities #186 #187 #188 #189 #191.
- Jonathan's revised patch duplicated the removal across sections, so Tom supplied a consolidated compatibility entry with upgrade advice and committed it unchanged as d42c617644 #192 #193 #195.
- All ten candidate commits are accounted for by the thread, but the final message raises a further request to document stronger contrib/amcheck cross-level verification; no response or patch resolving that request appears before the thread ends #197.