model-writtenrun 5701 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Holly Roberts reported that PostgreSQL 14 rejected a column type change when a clustered index supported a key referenced by a foreign key, although the operation worked in earlier releases #1.
- Japin Li traced the regression to get_constraint_index() returning an index for foreign-key constraints and proposed tolerating repeated clustered-index requests by comparing index names #2 #3 #4.
- Michael Paquier demonstrated the corresponding replica-identity failure, requested regression coverage, and questioned whether duplicate rebuild requests should instead be prevented; Japin extended the patch to cover replica identities and added tests #5 #7.
- Tom Lane revised the deduplication implementation to retain index OIDs and defer name lookups, then Japin renamed the test tables to avoid concurrent regression-test conflicts #9 #10.
- Before committing, Tom concluded that deduplication masked a change in get_constraint_index() semantics and demonstrated another failure when altering the referencing table; he favored restoring the function's original behavior to protect callers, including extensions #11 #12.
- Tom replaced the earlier approach with a constraint-type check in get_constraint_index(), preserving its faster catalog lookup while excluding foreign-key references, and expanded the tests to alter columns on both sides of the relationship #13.
- The final attachment landed unchanged as 641f3dffcd, with credit to Tom Lane and Japin Li and a backpatch to PostgreSQL 14.
- Japin thanked Tom after the commit, and Michael subsequently endorsed the semantic fix and corrected his commit reference; the thread ended with the fix committed and no outstanding patch revisions #14 #15 #16.