model-writtenrun 9424 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier reported spurious pg_dump[all] column-number diagnostics against older servers and supplied a patch adding the missing NULL AS amname output to eight getTables() queries in #1; Dmitry Dolgov successfully tested it against PostgreSQL 9.4 in #2.
- Tom Lane endorsed the fix and questioned why cross-version upgrade tests had missed the problem in #3; Andres Freund proposed checking for a missing amname column instead of adding dummy columns to every older query in #4.
- Tom Lane opposed silently accepting missing columns because that could conceal query mistakes, and proposed constructing version-dependent queries incrementally while retaining explicit NULL outputs in #5 and #7; Andres defended the smaller missing-column check in #6.
- Tom Lane identified that PQgetisnull treats an invalid column as NULL in #7, and Andrew Dunstan confirmed that the diagnostics appeared in buildfarm logs but neither failed pg_dumpall nor prevented the upgrade tests from succeeding in #8.
- Michael Paquier rejected masking missing outputs, raised readability and backpatching concerns about incremental query construction, and suggested an optional assertion mechanism for exposing such errors before asking to commit the original fix in #9.
- Andres Freund maintained his objection to duplicated query changes but accepted that the patch was better than leaving the bug unfixed in #10; Michael committed #1 unchanged as 54487d1560 and deferred broader redesign in #11.
- The immediate fix had landed by the thread's end, but the design disagreement remained open: Andres cited existing missing-column checks and disputed the consistency argument in #12, with no further patch submitted.