model-writtenrun 7048 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andy Fan reported that initdb with track_commit_timestamp=on fails a normal-transaction-ID assertion during bootstrap, proposing either bypassing special transaction IDs or removing the assertion #1.
- Michael Paquier argued for retaining the assertion and checking bootstrap mode instead; Andy implemented that revision, while Bertrand Drouvot questioned frozen-XID handling and Michael recommended retaining that sanity check and adding regression coverage #2 #3 #4 #5.
- Fujii Masao connected the report to Hayato Kuroda's earlier work, and Hayato endorsed the current fix; Andy then adopted Hayato's more efficient activation-time guard and added Michael's proposed initdb test #6 #8 #9.
- Fujii approved the activation-based patch and reported a separate bootstrap assertion with transaction_timeout; Michael committed the timestamp fix from #9 with light comment and test cleanup as 5a6c39b6df, backpatching through PostgreSQL 13 and leaving the timeout issue for separate work #10 #11.
- Hayato proposed a general flag to suppress unsafe configuration settings during bootstrap; Fujii initially preferred individual overrides, but Tom Lane supported a shared mechanism and suggested eventually replacing the committed timestamp-specific guard #12 #13 #14.
- Fujii identified that configuration processing preceded bootstrap-mode initialization, prompting Tom's tested proof of concept to move that initialization earlier and ignore transaction_timeout during bootstrap; very short timeouts still legitimately interrupted post-bootstrap initialization #15 #16.
- Michael questioned initialization ordering and favored limiting the broader change to the development branch, while Tom considered allowing dynamic defaults and argued that comparable bugs justified backpatching #17 #18.
- Fujii found that moving IgnoreSystemIndexes earlier allowed ignore_system_indexes=off to break initdb; Tom responded with a general GUC_NOT_IN_BOOTSTRAP flag covering that setting, transaction_timeout, and track_commit_timestamp, without yet reverting the committed timestamp fix #19 #20 #21.
- Andy tested the generalized patch after locally reverting the timestamp fix and verified that requested settings persisted in postgresql.conf; Fujii approved and requested TAP coverage, while Michael accepted the concept but retained his backpatching concern #22 #23 #24.
- Andy and Hayato proposed inline tests, but Michael recommended separate tests for each feature and warned that timeout thresholds could fail on slow test configurations; the thread ends with the timestamp fix landed, the generalized patch still under review, and Hayato asking whether transaction_timeout coverage should be omitted #25 #26 #27 #28.