model-writtenrun 9626 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Geoghegan proposed temporary development OIDs and an automatic replacement tool to prevent unrelated patches from acquiring OID conflicts in #1; Tom Lane suggested remapping a numeric range while skipping occupied destinations in #2.
- Peter's alternative of visibly symbolic temporary identifiers was dropped because it would complicate genbki.pl, while Robert Haas emphasized that any numeric convention needed documentation in #3, #4, #5, and #6.
- John Naylor volunteered and posted an initial implementation inside reformat_dat_file.pl, covering ordinary and array-type OIDs and documenting a development range starting at 9000 in #7 and #8.
- Peter Eisentraut opposed adding final renumbering and rebuilding steps to every affected commit, while Peter Geoghegan and Tom Lane argued that repeated conflicts discouraged review and disproportionately affected long-lived submissions in #10, #12, and #20.
- Tom identified remaining numeric catalog cross-references that could break under automatic renumbering in #13, and explained that ignoring collisions in CI would not solve conflicts within the same catalog in #19.
- Robert proposed automatically assigning new OIDs during builds and making them permanent once per release cycle in #21; Tom rejected enlarging the normal OID boundary because it would require dump/reload and questioned the implementation burden of automatic assignment in #22.
- The process dispute led to Tom's compromise: accept patches using randomly chosen high OIDs and renumber them together after feature freeze, with a release-checklist reminder, in #26; Peter and Robert supported this approach in #29 and #30.
- Tom extracted John's option-parsing and Makefile cleanup from #8 into 27aaf6eff4, then requested a separately named script and an explicit input OID interval in #34, also moving the target from PostgreSQL 13 to 12.
- John supplied the standalone script in #35, using Data::Dumper to avoid duplicating formatting logic; Tom then rewrote it in #36 to preserve formatting through textual substitutions, compute deterministic mappings before writing files, and renumber OIDs defined in headers as well as data files.
- Tom reported that check-world passed after moving OIDs above 4000, but moving lower OIDs exposed remaining numeric cross-references in #36; he addressed those with a separate symbolic-reference cleanup in #37.
- Testing the cleanup let Tom renumber all catalog OIDs after additional experimental adjustments, but exposed hard-wired OIDs in regression tests, initdb.c, and information_schema.sql; he considered the information-schema functions the most concerning remaining case in #37.
- The renumbering script from #36 landed as a6417078c4 with additional documentation, establishing the 8000–9999 development range and the release-cycle renumbering procedure.
- The symbolic-reference cleanup from #37 landed as 3aa0395d4e with expanded documentation and an additional conversion from the PGHEAPAM special case to ordinary access-method lookup, which Tom explained in #39.
- Tom declared the work essentially complete in #39; possible removal of PGNSP and PGUID special cases and remaining hard-wired OIDs in initdb.c and information_schema.sql were left as optional follow-up work.