model-writtenrun 5942 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Robert Haas proposed removing the redundant ReadRecPtr and EndRecPtr globals, identifying both an improper timeline-switch check before the first WAL record and a cosmetic error-reporting issue; he initially wanted a separate, backpatched timeline fix #1.
- Amul Sul approved the cleanup but initially could not reproduce the reported difference with and without the patch #2, prompting Robert to extract the timeline fix and supply a TAP reproducer whose startup timeout and fixed sleep made it unsuitable for committing #3.
- Alvaro Herrera suggested starting postmaster directly and checking readiness #4, but Robert explained that the standby would still be rejecting connections #5; Tom Lane recommended polling the log and warned that custom startup code would miss fixes to the shared test infrastructure #6.
- Robert emphasized the need for better recovery coverage while accepting that this particular test might be too fragile #7; Tom favored extending Cluster.pm and asked whether the history file could be installed before startup #8, while Robert explained that the reproducer needed to exercise a timeline rescan and considered restore_command as another possibility #9.
- Alvaro also questioned whether this narrow test justified startup hacks #10, and Amul subsequently confirmed that the standalone test failed without the fix and passed with it #11.
- Robert submitted the final bug-fix series with a proper commit message and explicitly excluded the test from committing; he abandoned backpatching because older branches also used the wrong timeline identifier, requiring invasive prerequisite changes for a laboratory-only corner case #12.
- The timeline fix landed as e7ea2fa342 from the final submission #12, passing the reader's recovery position explicitly into the timeline check and omitting the reproducer.
- Robert then rebased the remaining cleanup from #1 over the bug fix and committed it as d2ddfa681d; his final message confirms both landings, leaving the test uncommitted and both changes confined to master #13.