model-writtenrun 4228 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier proposed checking that changeDependencyFor() updates exactly one dependency for object schemas, extension schemas and planner support functions in #1.
- Heikki Linnakangas reviewed the checks, suggested consistent error wording and singular dependency comments, and supplied a test exposing an extension-schema failure in #2.
- Alvaro Herrera suggested prohibiting moves of extension members between schemas in #3, while Tom Lane and Michael Paquier argued against that restriction in #4 and #5.
- Michael traced the failure to AlterExtensionNamespace() taking the old schema from the first extension member instead of pg_extension, then submitted a combined fix, expanded regression tests and revised sanity checks in #6, proposing to backpatch only the bug fix.
- Peter Eisentraut asked whether changeDependencyFor() itself should raise errors in #7.
- Andres Freund raised security and dump/restore concerns about moving extension members independently in #8.
- Akshat Jaimini reported that the combined patch looked good and passed all tests on Arch Linux x86_64 in #9.
- Michael retained caller-side checks to preserve flexibility for callers that might change multiple dependencies in #10, then confirmed the dump/restore behavior but deferred broader schema-movement restrictions because of compatibility and usability concerns in #11.
- The extension namespace fix and regression tests from #6 landed unchanged as bd5ddbe866 and were backpatched through PostgreSQL 11; the remaining sanity checks and message cleanup landed on the development branch as 9b286858e3 with minor comment and quoting edits, as Michael confirmed in #12.
- After the commits, Tom argued that restoring extension members according to installation scripts was intended behavior in #13, Alvaro renewed the case for rejecting manual schema moves in #14, and Tom defended retaining that freedom in #15.
- At the thread's end, both patch components had landed, while broader restrictions remained unresolved; Michael noted that schema changes were sufficiently centralized to make such checks practical in #16.