model-writtenrun 6211 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Zhijie Hou reported that dropping and re-adding a publication could skip required table synchronization or resynchronize unrelated tables, and proposed operation-specific refresh logic in #1.
- Masahiko Sawada reproduced the bug and suggested refreshing the resulting publication list in #2, but Hou initially preserved selective refresh to respect the documented behavior and added tests in #3.
- Sawada demonstrated that the patch mishandled tables removed from a publication and tables shared by multiple publications in #6, leading Hou to compare DROP against the remaining publications and add dedicated tests in #8.
- Further review exposed another mismatch with selective-refresh semantics in #9, and Sawada explained in #12 that the subscriber lacks the publication-to-table history needed to implement selective DROP reliably.
- Amit Kapila proposed refreshing all publications for both operations in #11 and favored consistency in #15; Japin Li agreed in #16, and Hou replaced the earlier implementation with full refresh and copy_data support for DROP in #17.
- At Kapila's request in #18, Hou tested a separately reported drop-and-re-add synchronization failure and confirmed that the patch fixed it in #19.
- Sawada raised usability and query-overhead concerns about refreshing existing publications in #20, while Kapila defended the consistent behavior in #21 and Peter Eisentraut endorsed refreshing everything for both ADD and DROP in #22, leaving selective refresh outside this fix.
- Kapila requested a separate test file in #23, and Sawada identified outdated copy_data documentation in #25; Hou addressed both, expanded the tests and added psql completion in #26, after which Kapila supplied cosmetic test improvements in #28.
- Hou supplied a PostgreSQL 14 adaptation alongside unchanged HEAD content in #29, corrected a backport comment after Sawada's review in #30, and resent the final attachments in #31, which Sawada approved in #32.
- Kapila committed the latest HEAD patch unchanged as 1046a69b30, with a PostgreSQL 14 backport, and announced the push in #33; Sawada marked the open item resolved in #34, leaving no patch pending in this thread.