model-writtenrun 7172 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Mar - May 2023 · 8 entries
- Bertrand Drouvot proposed generating wait-event enums, lookup functions, and documentation from one text file to prevent missing documentation and orphaned events, initially covering five event classes in #1.
- Corey Huinker suggested using a template engine in #3, but Michael Paquier opposed adding an external Perl dependency in #4; Michael instead requested enforced sorting, documented input syntax, and clearer generated filenames in #5.
- Bertrand adopted section-based input in #8 after Michael explained that sections would eliminate duplicated class knowledge in the generator in #7.
- Michael proposed documenting Lock and LWLock through the new input while retaining their existing C implementations, and adding BufferPin and Extension to generation in #12; Bertrand implemented that broader coverage in #13.
- Michael supplied a simplified revision in #14, moving the LWLock note outside generated tables, retaining developer guidance in the input rather than generated headers, and recommending a separate BufferPin/Extension prerequisite.
- Michael found missing LWLock entries in #19, and Bertrand restored them in #21; subsequent review exposed case-sensitive sorting problems that were corrected across #23 and #25.
- While reviewing the generated ordering, Michael discovered an existing BufFileTruncate/BufFileWrite documentation error in #22 and fixed it independently in 44e977d7ec.
- Bertrand bundled the prerequisite from thread #47855 with the generator in #25 so CI could apply the series independently; its prerequisite landed unchanged as 2aeaf80e57, and this bundled submission is the latest matching source despite the other thread's claim.
Jul - Sep 2023 · 10 entries
- When PostgreSQL 17 development opened, Michael resumed the remaining generator patch in #27, correcting Perl details and Meson dependencies while identifying broken documentation generation and unnecessary simultaneous generation of code and documentation.
- Bertrand added separate --code and --docs modes and corrected Meson capture behavior in #30; Michael then fixed additional Makefile, Meson, and VPATH problems and committed fa88928470, reporting successful build and cleanup checks in #31.
- After the generator landed, Andres Freund reported an unhelpful parse error and questioned the redundant event-name column in #32, prompting both a diagnostics fix and another format-simplification effort.
- Michael's follow-up series in #36 normalized event names, expanded message-queue enum spellings, and proposed deriving enum symbols from CamelCase names; Bertrand endorsed the simplification and checked the improved diagnostic in #37.
- Alvaro Herrera objected that removing enum spellings would make source searches harder in #42, so Michael separated the undisputed changes from the naming debate in #45.
- The quote-removal portion of #45 landed as 183a60a628, and its diagnostic patch landed unchanged as aea7fe33fb; Michael resubmitted the two disputed patches in #46.
- Andres proposed preserving enum spellings and generating display names instead in #47; after a six-week gap, Michael implemented that reversed conversion in #49, omitting the WAIT_EVENT_ prefix while preserving Lock and LWLock names.
- Bertrand confirmed that the revised format addressed enum searches in #50, while Alvaro expressed only a mild preference for retaining the prefix in #52.
- Bertrand compared pg_wait_events output and found a missing WAL-sender rename in #54; Michael corrected it in #55, and Bertrand supplied the final rebase and restored description formatting in #56.
- The two patches forwarded in #56 landed unchanged individually as 414f6c0fb7 and 59cbf60c0f, with Michael confirming output consistency in #59.
Mar - Apr 2024 · 7 entries
- In March 2024, Noah Misch reopened the thread with an ABI problem: automatically sorting newly backpatched events could renumber existing enum values and make compiled extensions report incorrect events, as explained in #62.
- Michael proposed an append-only Backpatch region whose entries would retain input order in generated enums while documentation remained sorted in #63, and Noah accepted that design in #64.
- Bertrand implemented the region in #65, then corrected its branch guidance and populated empty regions through #68 and #72, incorporating Noah's clarification in #67 and Michael's testing in #71.
- Alvaro separately renewed his objection to generated display names after difficulty locating PgSleep in #66; Michael defended the enum-searchable format in #70, and no reversal of the two-column format followed.
- Michael supplied corrections and dummy test data in #73, clarifying that compatibility regions must remain empty on unreleased branches and that Lock and LWLock need no such regions.
- Bertrand renamed the marker ABI_compatibility and removed test-only data in #74, then clarified the Lock/LWLock comments in #76 after Michael's review in #75.
- The ABI fix landed from #76 with only comment reflow as f98dbdeb51; the final message #77 confirms completion, leaving only the possibility of a better marker name and the unresolved display-name search objection.