model-writtenrun 7980 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- The thread began with Tom Lane’s announcement that several array functions and width_bucket now used anycompatible types for greater usability, while functions tied to operator classes were left unchanged because they required broader work #1.
- Andrew Dunstan reported broken cross-version upgrade tests and temporarily dropped two affected aggregates; Tom favored replacing the regression aggregates’ implementation functions, considering genuine user dependence unlikely because array_agg already existed #2 #3.
- Andrew requested compatibility back to PostgreSQL 9.2 and explained that the buildfarm could adjust static source databases for unsupported branches, avoiding changes to their frozen regression suites #4 #5 #6.
- Tom supplied a patch for supported stable branches replacing array_cat_accum with array_larger_accum, suggesting that older source databases could redefine or drop the original aggregate #7.
- Andrew identified the still-unfixed first_el_agg_any case and shared his working buildfarm adjustments; Tom responded with a separate patch introducing a SQL wrapper that preserved the old polymorphic transition signature #8 #9.
- Tom combined both aggregate fixes in 97f73a978f, retaining the development-branch wrapper patch from #9 verbatim and adapting the earlier array_larger_accum change from #7.
- The thread ended with Andrew’s agreement after the commit had landed; the regression-test workaround was committed, while broader compatibility for user-defined aggregates was not addressed #10.