model-writtenrun 8723 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Quan Zongliang reported that changing a column type rebuilt its replica-identity index without restoring pg_index.indisreplident, and proposed saving and replaying the replica-identity SQL command in #1.
- Kyotaro Horiguchi questioned automatic restoration because type changes could invalidate subscriber compatibility in #2; Quan described lost replication after widening a smallint column in #3, and Euler Taveira argued that silently losing the index flag was a bug and recommended restoring it through the internal API in #4.
- Quan rebased the original implementation in #5; Peter Eisentraut subsequently requested an internal API approach and tests in #6, leading to Quan's direct restoration call inside ATRewriteCatalogs and accompanying SQL examples in #8.
- Peter supplied an alternative implementation in #9 that queued a replica-identity command through normal ALTER TABLE processing and added regression tests for both constraint-backed and standalone unique indexes.
- Euler checked branch applicability, advocated backpatching the bug fix, and supplied master and stable-branch adaptations in #10, although those attachments could not be extracted for direct comparison.
- Peter committed 1cc9c2412c, whose inspectable implementation matches #9 with light helper-naming and comment changes, crediting Quan's original patch and Euler's review.
- Peter confirmed completion and backpatching through PostgreSQL 9.5 in #11, leaving no further work proposed in the thread.