model-writtenrun 5887 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Gilles Darold proposed simpler regular-expression functions for counting, locating, and extracting matches, motivated by portability and avoiding complex queries, in #1, and supplied the initial implementation in #2.
- Gilles expanded the implementation in #3 to include regexp_like() and start-position/Nth-occurrence options for regexp_replace(), then targeted the next development cycle because PostgreSQL 14 was already too far along in #4.
- Erik Rijkers reported successful compilation and world checks and supplied documentation and catalog-description fixes in #5, which Gilles incorporated in #6.
- Chapman Flack explained that PostgreSQL's regex semantics should retain the de facto regexp_foo names rather than claim SQL-standard XQuery behavior in #7; Tom Lane agreed in #8, and Gilles corrected the description and resent the unchanged patch in #9 and #10.
- When review resumed in July, Tom warned that the new four-argument regexp_replace() overload could confuse argument resolution in #11; Gilles suggested a separate regexp_substitute() name in #12, but Tom reconsidered after showing that incorrect resolution would produce an identifiable error rather than silent misbehavior in #13.
- Tom reviewed Oracle and DB2 differences in #14, favoring strict end-option validation, avoiding additional ambiguous overloads, and rejecting the global flag except where meaningful for regexp_replace(); Gilles added a numeric-flags diagnostic and prepared-statement test in #15.
- Tom substantially rewrote the patch in #17, removing unsafe parentheses around user patterns, extending shared matching and replacement helpers, clarifying argument names, and simplifying documentation and regression tests.
- Gilles defended returning no match for nonexistent subexpressions in #16, Tom found supporting precedent in existing regexp_replace() behavior in #18, and Gilles endorsed the rewrite and retained behavior in #19.
- Gilles proposed and implemented a start argument for regexp_like() in #20 and #21, with support from Erik in #22, but Tom opposed the additional API ambiguity in #23, and Gilles withdrew that version in favor of the preceding patch in #24.
- Tom committed #17 as 6424337073, adding only wording and comment refinements plus a catalog-version bump, and announced the landing in #25.
- Peter Eisentraut proposed a separate documentation follow-up in #26 explaining the relationship between the new functions and SQL-standard regex functions, including an equivalence table and feature-list guidance; Gilles endorsed it in #27.
- Peter committed that documentation unchanged as 222b697ec0 and confirmed completion in #28, leaving both phases landed and the proposed regexp_like() start argument withdrawn.