model-writtenrun 4299 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Geoghegan proposed separating new-page allocation from ordinary nbtree buffer lookup so that heap-relation arguments introduced for logical decoding on standbys would remain only where needed #1.
- Michael Paquier questioned passing a separately computed catalog flag into the single-caller WAL helper #2, while Alvaro Herrera raised concerns about applying a substantial cleanup after PostgreSQL 16 beta1 despite its potential to simplify future backpatches #3.
- Peter folded the WAL helper into _bt_allocbuf, renamed _bt_newroot to _bt_newlevel, and standardized heap-relation parameter spelling in the next revision; removing P_NEW throughout the buffer manager remained outside this patch #4.
- Peter argued for PostgreSQL 16 to avoid branch divergence and explained why even VACUUM can need a heap relation when completing an interrupted split #5; he separately questioned TestForOldSnapshot's use of index relations, an issue left unresolved by this cleanup #6.
- Andres Freund supported separating allocation but questioned removing heap-relation arguments #7, and Peter explained that the goal was clearer interfaces and easier backpatching rather than performance #8.
- Michael continued to favor postponement after beta1 #11, but Heikki Linnakangas supported PostgreSQL 16, requested explicit assertions, and suggested always passing the heap relation to search and root lookup #12.
- Peter strengthened assertions and documentation while retaining optional heap relations for read-only searches, explaining that requiring them would preserve unnecessary arguments in other interfaces #14; Alvaro subsequently recorded the cleanup as a PostgreSQL 16 open item #15.
- Andres and Alvaro objected to unrelated parameter-case changes enlarging a late-cycle patch #16 #18, so Peter removed those changes and deferred naming consistency work #19.
- Andres requested retaining the table argument for cluster sorting #20, which Peter initially accepted #21, while their disagreement over broader argument removal and release timing remained unresolved #22 #23.
- Peter committed #19 with only comment refinements as d088ba5a5a and closed the open item; he ultimately removed the tuplesort_begin_cluster heap argument because he could not justify choosing NewHeap over OldHeap, leaving any follow-up to Andres #24.