model-writtenrun 623 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Chao Li reported that GROUP BY ALL ignored the equality semantics selected by ORDER BY USING, demonstrating one group where explicit GROUP BY produced two, and supplied a shared-helper fix with regression tests #1 #2.
- MiĆosz Bieniek reproduced the problem and confirmed that the patch made both queries consistent, while Daniel Gustafsson requested fuller helper documentation and simpler control flow #3 #4.
- Tom Lane argued that the grouping code needed a broader redesign covering grouping sets and SQL92 versus SQL99 behavior, and proposed reverting GROUP BY ALL from PostgreSQL 19 because that refactoring was too risky after beta2 #5.
- Daniel questioned empty-table behavior and possible SelectStmt mutation; Tom explained that grouping a constant correctly differs from GROUP BY (), and Daniel accepted that explanation #6 #7 #9.
- Chao accepted reverting the feature and stopped revising the narrow fix, leaving a broader PostgreSQL 20 rework open #8.
- Daniel supplied a separate revert that retained unrelated documentation and comment cleanups, noting that PostgreSQL 19 also needed its release-note entry removed; Tom allowed time for objections #9 #10.
- Tom committed Daniel's revert with a catalog-version bump and revised commit text as a32733d8f1, then announced that it had been pushed after no objections arrived #11.
- Chao raised the release-note cleanup, and Bruce Momjian later confirmed that he had removed the feature from the notes in his July update #12 #13.
- The thread ended with GROUP BY ALL reverted and the original fix uncommitted; David Christensen pointed to a separate thread containing a refactored PostgreSQL 20 proposal #14.