model-writtenrun 4215 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Thomas Munro proposed removing check_strxfrm_bug() because the affected Solaris and illumos versions were obsolete, while warning that broader strxfrm correctness problems still prevented enabling TRUST_STRXFRM #1.
- Thomas also proposed eventually requiring locale_t, initially blocked by old Solaris and OpenBSD build-farm systems; after OpenBSD upgrades, only wrasse's Solaris version remained an obstacle #2 #3.
- Nathan Bossart and Tom Lane supported removing the obsolete check, and Tom proposed dropping TRUST_STRXFRM as well; Peter Geoghegan and Thomas supported that broader cleanup because ICU was the practical direction for reliable collation #4 #5 #6 #8.
- Michael Paquier raised release-management considerations and mentioned Jeff Davis's alternative GUC proposal, while Jonathan S. Katz considered the cleanup acceptable for the PostgreSQL 16 open-items process #7 #9.
- Joe Conway questioned the remaining strxfrm use in selectivity estimation, and Tom Lane explained that approximate results were sufficient there but that treating transformed output as ASCII-like text was questionable #10 #11.
- Thomas committed the original check-removal patch #1 without substantive changes as 7d3d72b55e, then invited Jeff Davis to handle the broader libc strxfrm cleanup; Jeff supported removal but left open whether an abstraction should continue hiding provider details #12 #13.
- Discussion resumed around threaded servers: Thomas identified locale_t and thread-local locale selection as prerequisites, and Heikki Linnakangas highlighted an existing conflict between libperl's uselocale() calls and PostgreSQL's setlocale() usage #15 #16.
- New Solaris-capable build-farm hardware reopened the locale_t cleanup, leading Thomas to submit #18; Noah Misch supported dropping the old Solaris requirement because it removed thirty-eight conditionals, and Tristan Partin approved the patch #17 #19 #21.
- Thomas revised the Windows build definitions and comments in #22, then reported a late MinGW pg_upgrade failure; Peter Eisentraut challenged placing HAVE_* definitions in win32_port.h and pointed to Solution.pm's existing conventions #23 #24.
- The next revision kept Windows build systems consistent by leaving the conversion HAVE_* macros undefined and moving fallback implementations into pg_locale.c; Thomas reported passing CI and declined backpatching the previously disabled configure/MinGW libc-provider support #25.
- Tristan reviewed the revised helpers and requested a Meson line break; Thomas accepted the formatting change but retained the WIN32 guard, then landed #25 with light edits as 8d9a9f034e #26 #27 #28.
- After that landing, Peter argued that HAVE_* should describe usable functions, including replacements, and proposed restoring the Windows definitions; Thomas defended keeping the three build systems consistent, then proposed confining the Windows functions entirely to pg_locale.c because their UTF-8 limitations made global aliases hazardous #29 #30 #31.
- Peter endorsed the containment approach, and Thomas committed #31 unchanged as 4e9fa6d56b; the final message confirmed all attached changes had landed and left removal of TRUST_STRXFRM for future work #32 #33.