model-writtenrun 4701 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Heikki Linnakangas opened the thread with #1, explaining that pg_rewind could misidentify a newly promoted source's timeline before its checkpoint completed and proposing to consult minRecoveryPointTLI instead of relying solely on checkpoint metadata.
- Kyotaro Horiguchi endorsed the fix and probabilistic test coverage but questioned making sourceHistory file-local in #2; Heikki responded with #3, passing histories explicitly and freeing sourceHistory in the caller while retaining removal of the forced-checkpoint test workarounds.
- Ibrar Ahmed supplied a rebase in #4, and Aleksander Alekseev reported successful testing and approval in #5, suggesting an optional explanatory test comment that was not added.
- After a long delay, Heikki revived the patch in #8, updating its base and renaming function parameters to avoid shadowing the global targetHistory variable without changing behavior.
- Heikki committed #8 as 009eeee746 with an identical code diff but a revised backpatch policy, then confirmed completion in #9: the fix landed only on the development branch because the issue had historically been tolerated and adapting it to PostgreSQL 13 and older did not justify the risk and effort.