model-writtenrun 6406 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jeff Davis questioned the undocumented requirement to run IDENTIFY_SYSTEM before START_REPLICATION and proposed either documenting and consistently warning about it or removing the error in #1; Fujii Masao supported removal if safe and reported that the error occurred on standbys in #2.
- Andres Freund explained that GetStandbyFlushRecPtr() already refreshed the standby timeline and that determining the primary timeline was straightforward in #3; Jeff supplied the original discussion link in #4.
- Jeff revived the issue after five months in #5, and Andres reported encountering it during other review work and proposed explicitly computing the current timeline in #6.
- Jeff supplied a simpler patch in #7, removing the error and calling RecoveryInProgress() before choosing the replication path; he explained how existing code then initialized the timeline for primary, promoted, and standby servers, while noting that he had not yet tested it.
- Andres accepted Jeff's reasoning in #8, treating his concern about timeline determination through global state as an existing issue outside this patch.
- Jeff committed #7 unchanged as 8e7811e952, completing the submitted work; the commit explicitly declined backpatching because the prerequisite was expected behavior and unlikely to cause major problems.