model-writtenrun 1747 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jeff Davis proposed using the database default locale for tsearch parsing in #1, avoiding an unnecessary libc dependency for builtin and ICU databases while preserving libc behavior and the C-locale treatment of non-ASCII characters; he explained that making tsvector and tsquery collatable would act too late, after parsing.
- Jeff landed the first two preparatory patches from #1 as 8efe982fe2 and af164f31b9, extending the internal renaming and revising the exported API names and fallback checks; the duplicate regex helpers remained because removing them could affect performance.
- Jeff submitted the four remaining patches in #2, adapted to the committed APIs, and highlighted the parser's consolidation onto pg_wchar; he judged parsing changes more likely than index inconsistencies because existing parsed values retain their interpretation.
- Peter Eisentraut reviewed the remainder favorably in #3, questioned whether the C-locale non-ASCII exception still depended correctly on encoding, and suggested making char2wchar() local to pg_locale_libc.c.
- The hexadecimal classifier and database-locale accessor from #2 landed unchanged as 67a8b49e96 and e533524b23.
- In #4, Jeff explained why the replacement condition preserved the old encoding-dependent behavior, resent the two remaining patches unchanged, and added Peter's char2wchar() cleanup.
- Peter accepted the final revision in #5, leaving no expressed review objection at the thread's last message.
- After that final reply, the three patches from #4 landed unchanged as e113f9c102, 844385d12e, and ff53907c35, completing the tsearch locale conversion and its cleanups.