model-writtenrun 6173 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Vignesh C proposed invalidating relcache when creating or dropping FOR ALL TABLES publications in #1, crediting Amit Kapila with identifying how stale publication information could incorrectly permit publisher updates or deletes without replica identity.
- Zhijie Hou found that the original regression test passed without the fix in #2, prompting Vignesh C to populate the table and exercise UPDATE before creating the publication in #3.
- Kyotaro Horiguchi confirmed the revised test and implementation worked in #4, and Vignesh C incorporated his error-call formatting suggestion in #5.
- Amit Kapila supplied minor formatting and test improvements in #6 and proposed backpatching through PostgreSQL 10 despite the absence of user reports; Zhijie Hou endorsed the patch and backpatching in #7.
- After finding that the patch did not apply cleanly to the back branches, Zhijie Hou supplied HEAD and PostgreSQL 10–14 variants and reported successful regression tests in #8.
- Amit Kapila committed the final submission as 8bd5342740, preserving the last readable patch's implementation and tests modulo whitespace and rebasing, crediting Vignesh C and Hou Zhijie, and backpatching through PostgreSQL 10; he announced the push in #9.
- Tom Lane requested closing the CommitFest entry in #10, and Amit Kapila confirmed closure in #11, leaving the fix landed with no outstanding work stated.