model-writtenrun 6361 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Alexander Pyhalov reported that ANALYZE cleared partitioned tables' relhasindex flag, causing indexes to disappear from pg_dump and psql descriptions, and supplied a fix with regression tests in #1; he corrected the test's INSERT target in #2.
- Alvaro Herrera found the added inheritance checks unnecessarily complicated and supplied #4, preserving hasindex while resetting nindexes to zero after closing partitioned indexes so existing loops would skip them; he also reported that the tests passed.
- Zhihong Yu questioned whether hasindex was still necessary in #5, while Alexander Pyhalov accepted Herrera's simplification and restored a still-applicable comment about explicit column lists in #6.
- Herrera explained that hasindex remained necessary for vac_update_relstats in #7; Yu proposed encoding index presence with negative nindexes in #8, but Herrera rejected that alternative as harder to read in #9.
- Herrera committed #6 with a further simplification: RelationGetIndexList() determines index presence without opening partitioned indexes, alongside comment and test formatting changes, in d700518d74, backpatched to PostgreSQL 14; his final announcement in #10 confirms the fix landed and credits Pyhalov and Yu.