model-writtenrun 5216 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Bertrand Drouvot discovered stale index statistics after committed drops and rolled-back creation while working on separate table/index statistics, and proposed explicit creation/drop tracking with standby tests #1.
- Andres Freund explained that index creation bypasses heap_create_with_catalog(), recommended moving statistics registration into the shared heap_create() path, and requested regression coverage including concurrent reindexing; Bertrand incorporated those changes in the next patch #2 #3.
- Kyotaro Horiguchi approved the backend fix and questioned redundant table tests and coverage of other statistics kinds; Bertrand removed the extra table tests, explained existing function and subscription coverage, and added replication-slot cleanup tests #4 #5.
- Kyotaro and Andres challenged the hard-coded replication-slot ID assumption, and Andres requested cache-discard testing and database-OID query cleanup; Bertrand revised the slot tests to compare ID sets, consolidated the OID lookup, and reported successful testing with debug_discard_caches enabled #6 #7 #9.
- Andres committed the backend fix and index regression tests from the final patchset as d811ce6ea3 on master and PostgreSQL 15, leaving out the replication-slot tests because they were independent #9 #10.
- Bertrand agreed that the remaining replication-slot tests could target master only, but their landing was not confirmed in this thread; his final message announced plans to resume the separate table/index statistics split now that the bug was fixed #11 #12.