model-writtenrun 8119 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Vik Fearing proposed using anycompatible for LEAD/LAG defaults after a mixed numeric/integer call failed and caused a feature matrix to misidentify PostgreSQL's support, supplying catalog, documentation, and regression changes in #1.
- Tom Lane opposed squeezing the change into PostgreSQL 13 after beta and requested a broader investigation of polymorphic built-ins and overload ambiguity in #2; Vik moved the proposal to PostgreSQL 14 in #3.
- The scope expanded from LEAD/LAG and array_replace to functions combining array and element arguments in #4 and #5; Pavel Stehule warned about expensive implicit array casts in #6, while Tom argued that automatic coercion improved usability and rejected adding another polymorphism family in #7.
- Pavel's review found that the patch could change the result type, whereas SQL specifies coercing the default to the value's type, and raised this compatibility concern in #9 and #11.
- Tom accepted common-supertype behavior as less surprising than silently losing information, found no apparent overload conflicts, and advocated broader conversions in #12; Pavel continued to favor standard semantics but offered a documented anycompatible solution as a practical compromise in #13 and #14.
- Tom's experiments showed that converting operator-class members would require cascading changes throughout their families, so #15 limited the extension to array functions and operators without those connections and paired it with a lightly revised version of Vik's patch.
- In #15, Tom also identified compatibility risks for user-defined aggregates, rejected relaxing exact support-function lookup because of security implications, and raised an unresolved question about accepting declared anyarray values as anycompatiblearray inputs.
- Pavel tested both submitted patches successfully through compilation, documentation building, and make check-world, and requested additional array regression coverage in #16.
- Tom committed the LEAD/LAG component of #15 as 5c292e6b90 with a catalog-version bump and cosmetic catalog changes, then committed the array component as 9e38c2bb50 with completed documentation, two new regression cases, formatting, and another catalog-version bump.
- Tom announced both patches pushed in #17, and Pavel acknowledged completion in #18; the submitted series was fully landed, while broader operator-class conversions and the declared-anyarray question remained outside the committed work.