model-writtenrun 9295 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Lucas Viecelli proposed warning during CREATE PUBLICATION when wal_level is not logical, so users discover the required configuration change before attempting subscriptions or missing a restart window #1.
- David Fetter proposed rejecting publication creation and incompatible wal_level settings #2, but Tom Lane and Robert Haas explained that this would break dump/restore workflows and that global configuration cannot reliably inspect per-database publication catalogs #3 #6 #7.
- Andres Freund showed that logical replication already checks wal_level #8, after which Tom Lane endorsed the original creation-time warning without stronger enforcement #9.
- When the July Commitfest began, Thomas Munro requested a rebase and suggested separating the corrective instruction into a hint #11; Lucas incorporated that feedback and then supplied the correctly rebased attachment #12 #13.
- Tom Lane identified wording issues and regression expectations that would fail under other wal_level settings #15; Thomas Munro polished the warning and suppressed it around successful CREATE PUBLICATION test commands, reporting that the revised patch passed installcheck with any wal_level #16.
- Lucas accepted the adjustments #17 and Tom Lane agreed #18; Thomas Munro committed the exact final attachment from #16 as b31fbe852c and closed the thread by confirming it was pushed #19.