model-writtenrun 7468 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed replacing opr_sanity's approximate PL/pgSQL binary_coercible() implementation with a C wrapper around IsBinaryCoercible(), reporting a reduction from 29m9s to 3m19s under cache-clobber testing; his patch split create_function_0 from create_function_1 to make the wrapper available before the sanity tests without introducing troublesome shell types #1.
- Julien Rouhaud questioned whether the unused explicitly_binary_coercible() helper was safe to remove, while Andrew Dunstan tested moving network-type tests to the next parallel group to make room for create_function_0 #2 #3.
- Tom defended removing the old helper because its permissive checks supported obsolete type tricks and any future equivalent should use C; Julien accepted that reasoning #4 #5.
- Andres Freund's profiling shifted discussion toward time spent scanning catalog-cache and hash-table buckets, leading to suggestions about shrinking tables and compiler optimization; Tom noted that the relevant buildfarm animals already used optimization #6 #7 #8 #9.
- Tom posted a separate patch that skips work for empty catalog caches and nonpartitioned hash tables, reporting improvements on a slow privileges.sql query but favoring only the catalog-cache change because relying on counters for correctness seemed riskier in shared hash tables #10.
- Andres suggested smaller initial cache sizes and discussed excluding shared-memory tables from the optimization; Thomas Munro raised a local-lock list alternative, and Andres clarified that the LockReleaseAll() table was already local, leaving these broader optimizations unresolved #11 #12 #13.
- Tom committed the regression-test patch from #1 as 6303a57309, retaining its C implementation while moving more setup into create_function_0 and tidying test schedules; his final announcement confirmed that moving tstypes made room in the parallel group, while the separate empty-table optimization had no recorded landing in this thread #14.