model-writtenrun 6391 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Geoghegan proposed maintaining an opt-in list of pgindent commits for git blame, reporting better results than whitespace-based filtering, and supplied an initial standalone list in #1 and #3.
- Bruce Momjian suggested automatic generation in #4, but Peter argued for manual curation to avoid incorrect entries in #5, and Bruce agreed to document maintenance in the pgindent README in #6.
- Tom Lane requested readable dates and commit subjects in #7 and proposed a hash-first format in #11; discussion expanded the scope to perltidy and reformat-dat-files in #13, with Tom recommending a separate tracking-file commit after the formatting commits were pushed in #16.
- After a pause, Peter returned with an ISO-date list that also excluded a BufferGetPage change and its revert in #18; Tom identified missing formatting commits and challenged the subjective inclusion of change-and-revert pairs in #19.
- Peter added Tom's missing entries and removed the BufferGetPage pair, settling on 68 formatting-related commits in #20; Tom accepted the conventional top-level .git-blame-ignore-revs location in #21.
- Peter submitted the complete patch with maintenance documentation in #22, Bruce endorsed it in #23, and Tom accepted it after checking the relative-path configuration in #24 and #26.
- Peter committed #22 as 8e638845ff with only introductory comment wording changes and announced the push in #27.
- Tom noted that the push should have waited until beta2 was tagged in #28, and Peter acknowledged overlooking the release-process window in #29.
- After the commit, Alvaro Herrera reported that upgrading from Git 2.20 to 2.30 enabled the feature and improved blame output in #30; Peter explained the unavoidable remaining attribution of formatting-created lines in #31, leaving the work landed with no further patch pending.