model-writtenrun 8096 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Saeed Hubaishan reported that to_date parsed year -1 as 2 BC in #1, then demonstrated its disagreement with make_date in #3.
- David G. Johnston initially favored documenting the existing behavior and investigating compatibility before changing it in #4, while Dar Alathar-Yemen supplied an inline formatting.c correction in #6.
- Laurenz Albe demonstrated that Oracle rejects year zero and accepts negative BC years with SYYYY in #9, providing concrete compatibility evidence.
- David submitted documentation and regression tests for the existing behavior in #10, also identifying make_timestamp's rejection of negative years as a separate inconsistency.
- Bruce Momjian advocated correcting the parsing behavior for PostgreSQL 14 in #11 and questioned documenting behavior scheduled to change in #13, while David still favored documenting existing releases in #12.
- Peter Eisentraut independently implemented negative-year support for make_timestamp in #14, and Asif Rehman reported successful regression and feature testing in #15.
- Tom Lane argued that the parsing error warranted a back-patched bug fix in #16, but committed Peter's constructor patch only to the development branch as a094c8ff53, adding documentation and explaining that distinction in #17.
- Tom's parsing patch in #18 adapted Dar Alathar-Yemen's inline fix so that a negative year combined with an explicit BC marker produces AD, added tests and documentation, and retained year zero as 1 BC for compatibility.
- The parsing patch #18 landed unchanged as 489c9c3407 with backports to all supported branches; Tom explained in #22 and #24 that correcting parsing justified back-patching whereas extending the constructor looked like a feature.
- After the parsing commit, Robert Haas requested reverting its stable-branch changes in #26 and cited earlier compatibility discussions in #28, arguing that applications could depend on the longstanding behavior.
- Tom defended the correction using Oracle compatibility and the low likelihood of applications relying on the off-by-one result in #29 and #31, while Robert maintained his objection in #33 and David preferred changing only PostgreSQL 14 in #34.
- At the thread's end, both fixes remained committed, the original documentation-only proposal had been superseded, and no revert was recorded; Tom continued defending the parsing backports in #35, with that policy disagreement unresolved and stricter year-zero handling left as a possible follow-up from #18.