model-writtenrun 3325 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nathan Bossart proposed caching pg_class metadata in a sorted array in #1, reporting that a many-table pg_dump --binary-upgrade test improved from roughly two minutes to eighteen seconds while acknowledging increased memory consumption.
- Corey Huinker estimated modest memory overhead and favored caching over a prepared statement that would retain round trips in #2, while Michael Paquier supported the caching approach without detailed inspection or testing in #3.
- Daniel Gustafsson suggested deriving index status from relkind and explained that dumping all OIDs together would require backend changes in #4; Nathan accepted the prerequisite cleanup and retained per-relation output in #6, then supplied the two-patch series in #7.
- Daniel reported that both patches looked good and requested explanatory comments in #8; Nathan added those comments in #9 and noted that landing would have to wait a couple of months.
- Nathan aligned the implementation with existing pg_dump conventions in #10, moving cache population into a dedicated startup collector while retaining binary-search lookups, and rebased the series in #11.
- The cleanup from #11 landed unchanged apart from rebasing as 6e1c4a03a9, and the performance patch landed with field-name and SQL-alias cleanup as 2329cad1b9.
- Nathan confirmed completion in #12; both patches had landed, with no remaining work identified in the thread.