model-writtenrun 7360 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund proposed ten startup cleanups in #1 after shared-memory statistics work exposed confusing bootstrap and auxiliary-process coupling, platform-dependent initialization order, and interleaved single-user code.
- Tom Lane warned that replacing BackendIds everywhere could leave sparse sinval arrays and hurt performance in #2; Andres agreed in #3 that moving backend-status and process-signal indexing to pgprocnos could proceed independently, leaving that broader work outside this series.
- Kyotaro Horiguchi examined checker-mode separation, favored a separate argv entry for the auxiliary-process type, and discussed single-user initialization placement in #4; Robert Haas considered the initial cleanups mechanical and uncontroversial while reserving judgment on the remaining work in #5.
- The initial submission #1 supplied the unchanged --forkaux rename in 50017f7772, the bootstrap-entry separation with comment polish in 5aa4a9d207, and the block unindent with comment reflow in 27f7903466.
- The same submission #1 supplied the auxiliary-process file extraction with header polish in 0a692109dc, the AuxProcType cleanup with naming and macro cleanup in f8dd4ecb0b, and the signal-initialization relocation in 07bf378509; checker mode retained shared bootstrap implementation to avoid excessive duplication.
- Andres announced the initial landings and thanked both reviewers in #6, then explained in #7 that a standalone wrapper calling PostgresMain avoids fabricated argc/argv and that standalone PgStartTime initialization was later than postmaster's.
- The initialization-order patch from #1 landed unchanged in b406478b87, making BaseInit consistently follow PGPROC initialization and eliminating the separate buffer-manager initialization stage.
- Using that common initialization point, ee3f8d3d3a moved pgstat initialization into BaseInit and added lifecycle assertions to address autovacuum reporting statistics before initialization; its patch origin is #474367 in the shared-memory statistics thread thread #38970, while this thread supplied prerequisite startup restructuring.
- The subsequent fb2c5028e6, credited to Andres and Kyotaro, extracted the pgstat shutdown-hook change from #474367 in thread #38970, moving shutdown to before_shmem_exit so future shared-memory statistics could still access DSM; this thread and thread #44631 supplied prerequisite ordering discussion.
- Andres returned with the three remaining patches in #8, separating the startup-time move, polishing the initialization-order change, and naming the standalone wrapper PostgresSingleUserMain; moving the disconnection checks had not produced a clear improvement.
- Kyotaro explicitly approved all three remaining patches in #9 while questioning redundant processing-mode assignment; Andres explained in #10 that catalog and cache initialization still occurs below PostgresMain and left broader process-state simplification for later.
- After the final email, the remainder from #8 landed as the PgStartTime move with comment polish in 2c7615f77b, the unchanged InitProcess relocation in 3d7c752a2f, and the single-user extraction with comment edits in 7c83a3bf51; further extraction from InitPostgres and broader identifier cleanup remained future work.