model-writtenrun 8527 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Masahiko Sawada reported that a no-op synchronization cycle could persist a slot without a consistent snapshot, causing an assertion failure or data loss after promotion, and supplied a standalone TAP reproducer in #1.
- Zhijie Hou proposed preserving the previous no-consistent-snapshot skip reason and integrated a simplified reproducer into the existing recovery test in #2; Bertrand Drouvot identified another possible previous skip reason in #3, prompting Hou to check whether the slot remained temporary in #4.
- Joao Rito tested #4, reporting successful repeated recovery tests and regression checks while confirming that the new test failed on unpatched code in #6.
- Amit Kapila proposed retaining the previous skip reason in #7 and #9, while Bertrand explained that doing so could expose an obsolete reason after the original obstacle disappeared in #8 and #11.
- Sawada supplied an alternative in #10 that restored an explicit consistency output parameter and separated persistence eligibility from skip statistics; Hou and Amit opposed maintaining duplicate state in #13 and #14, while Sawada and Bertrand emphasized fragile caller assumptions in #16 and #17.
- Hou's compromise in #18 retained the skip-reason interface, tracked consistency throughout the function, centralized the failure check and logging, and asserted the temporary-slot invariant; it also split the TAP test into a separate patch.
- Bertrand accepted the assertion-based design in #20 but questioned repeated LOG messages in #21; Hou retained LOG for consistency with other skip reasons and because synchronization backs off in #22, and Sawada endorsed the clearer current-call logic in #23.
- Amit renamed the flag to consistent_snapshot_available and revised comments in #24, posting only the code fix; ba11c48390 landed that version with comment reflow and a rewritten commit explanation, with a backport to PostgreSQL 19.
- Amit confirmed the push in #25; the code fix was complete, but the separately proposed TAP test from #18 was absent from the committed change, with no stated reason or subsequent disposition in this thread.