model-writtenrun 8729 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Aug - Sep 2019 · 4 entries
- Juan José Santamaría Flecha began with a TODO-list feature allowing to_date() and to_timestamp() to recognize localized month and weekday names under the TM modifier, attaching implementation, documentation and Turkish tests in #1.
- Juan separated ASCII and localized matching in #2, explaining in #3 that the localized arrays lacked null terminators and therefore required explicit lengths.
- Alvaro Herrera challenged the first-byte comparison and hard-coded array dimensions in #4, leading Juan to introduce whole-string normalization and comparison in #5 and refine allocation handling in #6.
- After the September rebase in #8 and a several-month gap, Tomas Vondra reviewed documentation and naming in #9, while Arthur Zakirov identified allocation and array-handling problems in #11; Juan incorporated these reviews in #10 and #12.
Jan - Mar 2020 · 13 entries
- Tom Lane's review in #13 exposed unsafe multibyte truncation, questionable normalization-length calculations and unnecessary array-length plumbing, prompting a separate prerequisite cleanup in #14.
- Arthur proposed displaying the entire unmatched remainder in the incremental patch #15, but Alvaro recommended stopping at whitespace in #17 and supplied locale examples in #19; Tom accepted that compromise in #20.
- Tom committed both prerequisite patches from #14 as 4c70098ffa, adding whitespace-based error truncation, switching to ASCII downcasing and updating regression expectations, with backports justified by potentially malformed multibyte error messages.
- Tom then took over revision work in #21, rebasing Juan's feature over the cleanup, simplifying matching and null-terminating the locale arrays while raising unresolved concerns about normalization, allocation sizes and match lengths.
- Juan removed internal normalization and its test in #22, proposing that users normalize explicitly; Peter Eisentraut supported leaving normalization outside this patch in #25, while Tom deferred cache and debug-code changes in #24.
- Peter's review in #28 exposed Greek-sigma case-folding problems, and tests in #31 and #32 also highlighted grammatical variants of month names; Tom investigated the libc and ICU differences in #33.
- Mark Dilger raised alternate weekday names in #34, and the discussion explored library limitations and language collisions in #38, #42 and #43, but Juan reaffirmed the narrower goal of reversing to_char() output in #39.
- Peter accepted round-trip conversion as the feature's scope in #45, leaving general alternate-name support outside the patch, and Juan implemented uppercase-then-lowercase comparison in #47 to handle non-injective case mappings.
- After another month, Tom added input-collation handling and an exact-match fast path in #48, then extended parse_datetime()'s public signature in #49 so the JSON caller explicitly selects the default collation.
- Tom landed #49 as d677550493, preserving its localized-name implementation and adding light JSON documentation, error-message and test edits described in #51.
- James Coleman reported failing locale regression tests with diagnostic output in #52; investigation showed that his system had the UTF-8 Turkish locale but lacked the bare tr_TR locale required by the test, and installing it made make check pass in #59.
- Tom corrected his initial explanation in #61, explaining that the bare Turkish locale also exercises conversion from the locale's encoding to the database encoding, and defended retaining that coverage in #63.
- Both implementation changes remained committed at the thread's end, with no follow-up code fix attached; James left the locale-packaging mystery and possible test-coverage improvements unresolved in #64.