model-writtenrun 6356 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May 2021 · 15 entries
- Bruce Momjian opened the thread with an already committed first draft of the PostgreSQL 14 release notes, asking for clarification and feedback before ordering and marking up the document for beta #1.
- Early review established a focus on user-visible behavior: Julien Rouhaud explained that a data-only dump optimization did not change dump contents, leading Bruce to remove its entry #3 #16, while David Rowley clarified the nested-loop, TID-range, and parallel-scan improvements #5 #28.
- Laurenz Albe clarified when psql ignores edits abandoned without saving, and Bruce adopted a short description covering the affected editor commands #2 #21 #35.
- Matthias van de Meent and Peter Geoghegan explained why reclaiming trailing heap line pointers and admitting wide tuples to nearly empty pages mattered for bloat, persuading Bruce to add improvements he had initially considered too minor #6 #19 #23 #25 #29.
- Justin Pryzby's broad review corrected configuration names, defaults, authentication descriptions, and wording; Bruce moved the SCRAM default change into compatibility notes while keeping ordinary performance tuning changes outside that category #10 #40.
- Amit Kapila helped distinguish logical-decoding APIs from built-in replication capabilities and consolidate streaming and synchronization descriptions, while Amit Langote clarified that execution-time partition pruning was the new UPDATE/DELETE capability #34 #54 #69 #70 #75.
- Joe Conway and Tom Lane resolved the rationale for returning NULL on invalid numeric column references, while Alexander Korotkov refined the description of discarded-token handling in text-search queries #44 #48 #49 #50 #60.
- Peter Geoghegan redirected the bottom-up deletion description toward avoided splits and bloat; Bruce dropped a minor VACUUM VERBOSE reporting item but retained recycling newly deleted B-tree pages during the same vacuum #33 #41 #52 #53 #55.
- Justin's renewed omission review led Bruce to add COPY FREEZE visibility handling, ALTER TYPE subscripting support, and an additional GiST pageinspect source citation in 5eb1b27d20, using the inline additions discussed in #86 #87 #88 #89.
- After that landing, Tom Lane objected to the separate ALTER TYPE entry as redundant, and discussion with Alvaro Herrera and Justin shifted the explanation toward extensible subscripting; Bruce removed the redundant entry and separated general subscripting from JSONB support under data types #90 #92 #94 #96 #97 #99.
- Tom then corrected the subscripting attribution to Dmitry Dolgov and supplied clearer wording, which Bruce adopted #101 #102.
- Further technical review clarified already-MD5-formatted passwords, partition statistics propagation, memory-context logging, and the meaning of pg_locks.waitstart #93 #98 #95 #100 #103 #104.
- Masahiko Sawada, Justin, and Peter distinguished removal of vacuum_cleanup_index_scale_factor in PostgreSQL 14 from its earlier disabling in PostgreSQL 13, convincing Bruce that users needed an upgrade warning despite related backpatched changes #110 #111 #113 #117 #120 #122.
- Discussion of reporting old and new configuration defaults initially produced a vacuum_cost_page_miss description in milliseconds, but Peter identified that these were abstract cost units; Bruce's inline correction #133 landed as 8dcae7f0a3 following #126 #128 #129 #132.
- Stephen Frost explained that the new predefined roles grant read or write privileges on tables, views, and sequences without restricting other grants; Bruce's revised paragraph #135 landed as 7ce7d07e1c following #134.
Jun - Jul 2021 · 12 entries
- When Justin requested explicit coverage of track_wal_io_timing, Bruce kept the release-note focus on the new pg_stat_wal view and left its timing-control details to the linked documentation #136 #137.
- June proofreading from Justin corrected terminology, grammar, and markup and merged reindexdb's tablespace option with the corresponding SQL feature in a2559d4093; Bruce retained separate factorial entries because they explained the replacement function's availability in older releases #138 #139 #142 #143.
- Tomas Vondra corrected the ordering of logical-replication contributor credits, and John Naylor removed wording that could be mistaken for a contrib/bloom improvement; both inline fixes landed together in 86b222b090 #140 #141 #144 #145.
- Masahiko identified missing coverage of Windows files larger than four gigabytes, and Michael Paquier supplied WAL and relation-segment examples; Bruce placed the resulting entry in Source Code and committed #151 as 90855908b7 #146 #148 #149.
- After the Windows entry landed, Masahiko reported an attachment build failure and Tatsuo Ishii resent Bruce's patch in UTF-8; Bruce confirmed that his mail program had changed the attachment encoding while the Git version remained correct #152 #156 #157.
- Peter reported a new control for disabling vacuum's index-cleanup bypass, prompting Bruce's separate update to describe INDEX_CLEANUP's auto default and refresh the release notes' currency date #150 #154.
- Simon Riggs identified omitted concurrent-indexing improvements, with Peter supporting their inclusion; Bruce reconsidered their general interest and posted a vacuum-focused entry #158 #159 #160.
- Justin's next proofreading pass supplied markup, hyperlink, grammar, configuration-reference, and attribution corrections, which Bruce committed as caa0f07d2d and attached in #162 after reviewing #161.
- Simon then explained that concurrent indexing offered two distinct benefits and requested clearer terminology; Bruce revised the vacuum entry to discuss dead rows and minimal-locking operations and added the avoided-wait benefit #164 #165.
- Simon found that combining the benefits still obscured their independence, so Bruce split the vacuum and indexing descriptions in #167 following #166.
- Justin clarified the remaining operator markup, scan wording, and catalog-foreign-key discoverability issues, producing the incremental patch #168, which landed unchanged as a3a681f859 following #163.
- The split concurrent-indexing patch #167 landed as 56d467a07d with small command-tag corrections, and Simon later confirmed that the result satisfied his concerns #169.
Sep 2021 · 2 entries
- In September, Justin revisited the partitioned-table statistics description after the underlying autoanalyze work had been reverted and partially restored; Tom clarified the actual statistics behavior and reported removing internal editorial questions during copy-editing, while rejecting minor dump and psql fixes as insufficiently useful for release-note coverage #170 #171 #172.
- The thread ends with Yugo Nagata requesting coverage of pgbench's changed timing output and pipeline support, leaving those requests and Justin's partitioned-table statistics request without a recorded resolution in the supplied discussion #173 #172.