model-writtenrun 7423 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed replacing explicit PIN entries with a fixed OID cutoff in #1, reporting substantially smaller pg_depend storage and lower test CPU time; a dynamically recorded cutoff was rejected because pg_resetwal could not reconstruct it.
- Concern about scarce reserved OIDs led Tom to propose having genbki.pl pass its next available OID to bootstrap in #2, avoiding the extra unused range introduced by the first draft.
- Andres Freund questioned the exceptions for template databases, the public schema, and low-numbered large objects in #3 and #5; Tom retained existing behavior for compatibility and recovery reasons in #4 and #6.
- Tom implemented the set_next_oid bootstrap command and documentation updates in #7; Robert Haas then questioned the documented boundaries and suggested clearer wording in #8.
- Robert and Tom explored reserved-OID exhaustion in #9, #10, and #11, favoring per-catalog reuse of generated OIDs while deferring normal-range allocation for imported collations because its implementation and interactions were unclear.
- After the separate per-catalog allocation change, Tom posted #12, removing set_next_oid and simply starting backend allocation at 10000; he also identified pre-existing bootstrap allocation before catalog indexes exist, leaving that issue outside this patch.
- John Naylor reviewed the final patch in #13 and measured an improvement from 623 ms to 567 ms in #15; Tom agreed the shared-object wrapper could be removed but explained in #14 why initdb still needed SQL to invoke the backend function.
- Tom committed #12 as a49d081235 with the redundant shared-object wrapper removed and minor integration edits, then confirmed completion in #16; the PIN-replacement work was fully landed.