model-writtenrun 5151 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut opened the thread by questioning why a publication could not include both a schema and an explicitly listed table within it, particularly because that restriction prevented otherwise valid table moves between schemas #1.
- Amit Kapila explained that the restriction avoided ambiguity when dropping schema membership, but Tom Lane advocated independent schema and table membership and questioned concurrency safety; Amit confirmed a locking gap in the existing checks #2 #3 #4.
- After Zhijie Hou clarified that schema publication already tracked schemas rather than individual table entries, Robert Haas argued that dropping schema membership should retain explicitly published tables, and Amit accepted restoring that behavior #6 #7 #9.
- A parallel syntax debate emerged because Robert Haas and Mark Dilger found ALL TABLES IN SCHEMA misleading beside GRANT, while Mark explained the need to distinguish future object types and Hou defended consistency with FOR ALL TABLES #10 #8 #12 #13.
- Vignesh C supplied the first patch removing the overlap prohibition and the corresponding ALTER TABLE SET SCHEMA restriction, with documentation and regression coverage #14.
- Hou found that the patch allowed column-list behavior inconsistent with existing replication rules; Peter Eisentraut proposed forbidding any schema membership in publications containing column lists, avoiding a renewed restriction on table moves #17 #18.
- Hou incorporated that restriction and made schema membership override table row filters, following the existing OR-combination semantics, then posted the combined implementation based on Vignesh's patch #19 #20.
- Peter Smith rebased the patch, Amit supplied review comments and an attached comment fix, and Hou incorporated these plus off-list feedback from Peter and Vignesh; Amit then posted a polished version and Hou supplied a PostgreSQL 15 rebase #21 #23 #24 #25 #26.
- Alvaro Herrera challenged the breadth of the column-list restriction and the superuser requirement, while Jonathan S. Katz identified the overlap problem as a PostgreSQL 15 release issue and considered privilege relaxation suitable for a future release #27 #30.
- Alvaro proposed removing only ALL from the schema-publication syntax, preserving TABLES so future support for other object types would remain unambiguous #34.
- The release management team requested resolution of the overlap restriction before release but initially opposed syntax changes; Mark subsequently proposed FUTURE wording and Hou suggested an include_future option, neither of which appeared in the submitted patches #37 #38 #40.
- Alvaro submitted the independent TABLES IN SCHEMA syntax patch, and Jonathan and Amit endorsed it as a reasonable simplification that preserved future extension options #41 #42 #43.
- Amit announced his intention to commit the overlap fix before RC1, retaining the conservative column-list restriction while leaving possible relaxation to future releases #44.
- The thread's last messages reported successful syntax-patch CI and no objections; both changes were awaiting their recorded commits at that point #45 #46.
- Alvaro committed the syntax patch from #41 with only formatting and rebase differences as 790bf615dd, including a PostgreSQL 15 backport.
- Amit then committed the final development-branch overlap patch from #25, adapted to the new syntax, as 13a185f54b and backported it to PostgreSQL 15; broader column-list combinations and relaxation of superuser requirements remained future work.