model-writtenrun 7164 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Justin Pryzby opened the thread with an order-dependent dump/restore failure involving inherited primary keys, a complaint about restoring new pg_dump output into older servers, and two incorrect version comments in #1.
- Alvaro Herrera explained that loading newer pg_dump output into older servers is unsupported in #2, which Justin accepted while supplying two more upgrade failures involving missing or inherited throwaway constraints in #4.
- After investigating the restore failure, Alvaro rejected suppressing primary-key recursion because it was needed for existing behavior and rejected a special pg_dump dependency as an undesirable workaround; instead, he proposed converting NO INHERIT constraints and allowing their local definition to be dropped in #3.
- Alvaro posted the initial implementation in #5 and identified a remaining case requiring a separate pg_dump adjustment in #6; Justin tested the attachment, confirmed the original failure was fixed, and connected it to Andrew Bille's earlier report in #7.
- Testing the related case exposed missing descendant locks and persistent throwaway constraint names, so Alvaro added hierarchy-wide locking, improved naming, an inheritance safety check, and regression tests in #8.
- Alvaro committed d9f686a72e from #8, further refining pg_dump's naming condition and preventing drops of inherited constraints, and announced that the additional reported restore failures were covered in #9.
- Justin raised upgrade-test coverage and the newly prohibited removal of inherited not-null constraints in #10; Alvaro confirmed the restriction was intentional and agreed it deserved release-note coverage in #11, while Justin explained that older pg_upgrade precheck deficiencies were being pursued separately in #12.
- Robert Haas asked whether release-blocking work remained in #13, and Justin distinguished the remaining PostgreSQL 17 concern from the longstanding pg_upgrade precheck work in #14.
- Justin reported that partitions created with LIKE followed by ATTACH retained explicit not-null constraints absent from partitions created with PARTITION OF in #15.
- Alvaro initially suggested changing LIKE behavior in #16, then established that INCLUDING INDEXES already produced the desired primary-key-only representation and proposed documenting the distinction in #17.
- Justin challenged the differing post-ATTACH results in #18; Alvaro reconsidered whether standards compliance instead required explicit not-null constraints in both creation paths in #19, and Justin clarified his lock-avoidance use case without endorsing a final solution in #20.
- After the last supplied message, 21ac38f498 corrected the version comments reported in #1 alongside error-message consistency fixes from #561677 in thread #46431; Justin's comment report does not establish review of that repair, and the supplied thread leaves the partition-creation consistency question unresolved.