model-writtenrun 9271 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane opened the thread with a reproduction showing that DROP DOMAIN CASCADE could remove a partition-key column and leave its table unusable, and proposed either explicit dependencies or a lower-level deletion check #1.
- Amit Langote reproduced the failure with a composite type and initially preferred the deletion check because it would also protect existing tables, while Alvaro Herrera suggested representing dependencies on a table's partitioned status #2 #3.
- Tom rejected the lower-level check because temporary-object cleanup and deletion ordering could make it fail during legitimate drops, choosing internal dependencies with protection limited to newly created tables on released branches #4.
- Robert Haas and Tom considered preserving data by removing partitioning, with Tom clarifying that partitions could be detached without moving data, but left that feature aside because its desired semantics were uncertain #5 #6 #7 #8.
- Tomas Vondra suggested a repair function for existing tables, but Tom explained that reconstructing dependencies would require inspecting partition expressions for column references and did not volunteer to implement it #9 #10.
- Tom posted the sole patchset, making partition-key columns internally dependent on their table, guarding dependency traversal against recursion, improving error messages, and testing direct and expression-based partition keys #11.
- Tom landed that patch as a0555ddab9 with light cleanup, a catalog-version bump, and an expanded pg_dump comment; backport testing also exposed circular-dependency warnings in older pg_dump versions, requiring an existing suppression to be backported #12.
- Because noisy test output nearly concealed those warnings, Tom proposed backporting the older pg_upgrade noise-reduction change from thread #213535, and Alvaro welcomed the action without providing a substantive patch review #12 #13.
- Manuel Rigger thanked Tom after the partition fix had landed, without reporting patch testing or code review #14.
- The older pg_upgrade change was backported in 8d21512dc3, 69c3d51911, 75348a7332, 7ac7bf50cc, and 19f9a5aed9; Tom closed the thread confirming consistent quiet test output across branches, with no attached patch left pending #15.