model-writtenrun 5876 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Wang wei reported that dropping a replica identity index prevented published deletes despite an existing primary key, contradicting the source comment; Ashutosh Bapat confirmed the discrepancy and missing documentation in #1 and #2.
- Michael Paquier confirmed that the existing behavior corresponds to REPLICA_IDENTITY_NOTHING and proposed documenting it on the ALTER TABLE page while presenting the replica identity options as a variable list in #3.
- Alvaro Herrera proposed forbidding removal of replica identity indexes in #7, and Euler Taveira supported that restriction to prevent replication failures in #9; Michael advocated correcting existing branches' comments and documentation first in #8.
- Wang wei supplied the comment and documentation patch in #10, and Michael polished its wording and markup in #11; Euler endorsed the revised patch in #12, and Wang accepted the edits in #14.
- Euler also submitted an independent implementation rejecting DROP INDEX for replica identity indexes, with regression tests, in #12; Michael supported pursuing such a behavior change on the development branch in #13.
- Michael committed the documentation and comment patch from #11 unchanged as fc95d35b94, backpatched it through PostgreSQL 10, and announced completion of that part in #15.
- The DROP INDEX restriction remained unresolved: Zhijie Hou identified indirect index removal through dropping columns or primary-key constraints in #16, and Michael explained in #17 that enforcement in index_drop() could cover those dependency paths, while contextual error messages still needed consideration.