model-writtenrun 2311 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jeff Davis opened the thread in #1 after discovering that locale-code unification changed LOWER('I') under the default Turkish single-byte libc collation, potentially invalidating expression indexes after upgrade.
- Tom Lane argued for preserving longstanding behavior and compatibility with identifier lowercasing in #2; Jeff restored the special case without an attached patch in e3fa2b037c and confirmed the fix in #3.
- Tom attributed the UTF-8 exception largely to historical behavior in #4; Peter Eisentraut questioned retaining poorly tested single-byte libc paths in #5, and Tom clarified that identifier lowercasing ignores the name type's collation in #6.
- Jeff broadened the discussion in #7 to Unicode-aware identifier case folding and whether it should follow the default collation or built-in Unicode rules, but the thread reached no decision on those questions.
- Months later, Noah Misch identified a corresponding unrestored regex special case in #8; Jeff supplied the first attached fix in #9, preserving the historical difference between regex and SQL case conversion in Turkish UTF-8 locales.
- Noah's review in #10 found stale descriptions of locale handling and proposed separate default-locale regex strategies; Jeff responded in #11 with separate comment and behavior patches plus an optional strategy refactor, which he considered worse because it duplicated many switch cases.
- Noah accepted the simpler approach in #12 despite preferring the alternative and requested only a clarification that both numbered libc cases enforce ASCII mapping for the default collation.
- The comment component of #11 landed with Noah's wording adjustment as b107744ce7, and its functional-fix component landed unchanged as 2e5353be25; the optional strategy refactor was left out.
- The final message, #12, completed review before the April commits; the supplied commits establish that both targeted regression fixes landed, while the broader identifier-case-folding questions raised in #7 remained unresolved.