model-writtenrun 132 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Zsolt Parragi reported that upgrading from PostgreSQL 18 or earlier could lose the newest multixid when its members ended exactly at the 32-bit offset boundary, causing subsequent reads and vacuum to fail while pg_upgrade --check missed the problem #1.
- The patch restricted zero-offset rejection to offsets read from the next multixid's slot, preserving valid wraparound from the control file, and added a regression scenario for upgrades from pre-19 installations #1.
- shihao zhong reviewed the patch, said it looked right, and supported fixing the bug despite its rarity #2.
- Heikki Linnakangas committed the submitted code and test unchanged as 2323f03a17 and announced completion in #4.
- The thread ended with shihao zhong acknowledging the completed fix and thanking the author and committer, with no further technical changes requested #5.