model-writtenrun 9107 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- David Steele opened the thread with testing showing that recovery_target_timeline works with standby.signal and supplied a documentation exception in #1.
- Fujii Masao pointed out that other recovery targets also work in standby mode in #2, and David's detailed tests in #3 exposed a more serious PostgreSQL 12 regression: targets could affect recovery without either signal file.
- Peter Eisentraut considered moving the timeline setting's documentation in #4, but David explained its practical use for rebuilding standbys on a specific timeline in #5; Peter committed the original documentation patch unchanged as 775578a445, announcing it in #6.
- David emphasized that standby targeting was established behavior but signal-free targeted recovery was a regression in #7; Fujii proposed preserving older behavior for PostgreSQL 12 in #8 and supplied a reproducible crash-recovery failure plus a patch rejecting configured targets in #10.
- Tom Lane objected that rejecting retained settings could prevent genuine crash recovery in #13, while David opposed fatal errors for persistent settings and preferred conditional loading in #14; the rejection approach was abandoned in favor of ignoring settings during crash recovery.
- Peter supplied inline guards for recoveryStopsBefore() and recoveryStopsAfter() in #15; Fujii changed their condition from InArchiveRecovery to ArchiveRecoveryRequested to cover backup_label without a signal file, attached the revised patch, and reported successful testing in #16.
- Peter separately corrected the documentation to state that all recovery targets apply in standby mode through e04a53a607, announced in #17, superseding the narrower timeline-only exception.
- After Peter's approval in #18, Fujii committed #16 unchanged as 7acf8a876b, crediting Peter as author, and submitted a separate patch for restore_command and recovery_end_command in #19.
- The discussion then considered whether recovery settings should be suppressed centrally: Michael Paquier favored avoiding their effects in #21, Fujii explored SetConfigOption() and signal-aware GUC handling in #22, and Robert Haas argued for checks in the consuming code in #23.
- Stephen Frost argued that displaying inactive settings was misleading in #24 and #29, while Robert defended SHOW as reporting configured values in #31; Fujii declined to delay the PostgreSQL 12 fix for that broader redesign in #27, and the disagreement remained unresolved in #32.
- Peter reviewed the command guards and identified recoveryApplyDelay() as another case to address in #33; Fujii committed the command patch unchanged as 20961ceaf0 and confirmed completion in #34.
- At the final message, the documentation and both crash-recovery patches had landed, the fatal-error proposal had been replaced, and Fujii directed the remaining apply-delay issue to thread #41286 in #34; that thread's later patch #434843 supplied the subsequent fix 9b95a36be8.