model-writtenrun 6374 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jun - Jul 2021 · 16 entries
- Jonathan S. Katz opened the thread with an availability-query example that needed to decompose a multirange into individual ranges, and confirmed that procedural-language users otherwise had to unpack the values themselves #1 #4.
- Tom Lane supported a polymorphic unnest(anymultirange) returning anyrange but questioned adding it during the PostgreSQL 14 beta; Alvaro Herrera regarded the missing unpacking facility as an omission, while Jonathan preferred unnest if only one conversion mechanism could be added #2 #5 #6.
- Alexander Korotkov supplied the first unnest implementation and deferred subscripting because proper expanded-object support required too much work for PostgreSQL 14 #7.
- Alexander expanded the patch with array casts, documentation, and tests; Justin Pryzby's testing then exposed a crash when casting an empty multirange, which Alexander fixed by initializing the array builder and adding empty-input regression cases #11 #12 #14.
- Jonathan successfully tested SQL, PL/pgSQL, and custom range types, and his documentation wording was incorporated; Zhihong Yu's terminology and naming suggestions were discussed, with the existing set-returning terminology and surrounding naming conventions retained #15 #16 #19 #20 #23 #24.
- Alexander landed the combined patch from #17 as 29854ee8d1, adding the catalog-version bump, and announced the push in #30.
- Tom reported widespread buildfarm failures involving pg_dump and pg_upgrade and requested a revert; Alvaro proposed preserving only unnest, while Jonathan agreed that casts were optional for his original use case #31 #33 #34 #36.
- Alexander reverted the entire initial landing with 817bb0a7d1 and confirmed the rollback in #38; he later reproduced the check-world failure and acknowledged that his earlier successful test report was mistaken #41.
- Alexander explained that existing polymorphic rules produced an array of scalar subtypes rather than ranges, then posted a revised combined patch that made pg_dump skip automatically generated casts; a follow-up clarified comments and passed commitfest testing #40 #42 #43 #44.
- Tom still opposed the per-type array-cast functions and wanted improved polymorphism considered for PostgreSQL 15, so Alexander posted an unnest-only patch preserving the original implementation #45 #49.
- The reduced patch waited over concerns about forcing another initdb after beta2, although Noah Misch argued that a feature important enough to bypass feature freeze should also justify its own catalog-version bump #47 #48.
- Alexander separately found two incorrect multirange operator descriptions and a mismatched join-selectivity function name, posting an independent catalog cleanup #50.
- Alvaro recommended applying the cosmetic catalog cleanup without a version bump, Justin supplied precedents and checked the selectivity hunk, and Alexander explained that the old and new selectivity functions had identical implementations #51 #53 #57 #58 #59.
- The catalog cleanup from #50 landed unchanged as 768ea9bcf9, without a catalog-version bump, and Alexander confirmed the push in #60.
- After almost a month without unnest, Alvaro argued that multiranges remained incomplete without it; Tom accepted the unnest-only addition, and Alexander and Jonathan supported proceeding while leaving array casts for later work #61 #62 #63 #64.
- Alexander landed the unnest-only patch from #49 as 9e3c217bd9, unchanged apart from rebasing and whitespace, and reported successful check-world runs on master and PostgreSQL 14; he also noted separately correcting a forgotten master catalog-version bump #65.
Sep 2021 · 1 entry
- The thread ended with unnest and the catalog cleanup committed, array casts deferred, and Magnus Hagander pointing out that the commitfest entry still incorrectly showed work needing review and should be closed or split for any remaining work #66.