model-writtenrun 7143 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Dec 2023 - Apr 2024 · 24 entries
- Jeff Davis proposed checked-in Unicode property and case-mapping tables in #1, aiming to combine platform-independent character semantics with fast bytewise sorting; the independent category-table compaction landed unchanged as 719b342d36.
- Jeremy Schneider raised Unicode-update and expression-index stability concerns in #2, while Jeff Davis clarified in #3 that the proposed tables would change with PostgreSQL major releases rather than external-library upgrades.
- Daniel Verite identified additional language-specific case tailoring in #4, and Jeff Davis acknowledged in #6 that ICU would retain advantages for tailoring and sophisticated word boundaries.
- Robert Haas questioned the consequences of bytewise defaults in #13, Daniel Verite explained the value of bytewise indexes with linguistic sorting selected explicitly in #11, and Jeff Davis clarified in #14 that changing initdb's default was outside this proposal.
- Jeff Davis integrated the Unicode infrastructure with a builtin C/C.UTF-8 provider in #22, then improved conversion performance, tests, and documentation in #23.
- Jeremy Schneider's review in #24 preceded the cleanup commit cf64d4e99f, which used #23 with a small assertion-build fix; his separate successful-codepoint counter in #28 was incorporated into the later series in #32.
- Testing by Jeremy Schneider in #25 and Daniel Verite in #29 prompted clearer builtin-locale documentation, the PG_C_UTF8 object name, and libc-compatible initcap behavior in #32.
- The catalog rename initially raised old-psql compatibility concerns, but Daniel Verite, Robert Haas, and Michael Paquier supported prioritizing the new catalog design in #34, #35, and #37, and Jeff Davis retained the rename in #36.
- Daniel Verite's cross-version ICU tests in #38 showed that already-assigned characters can change properties, leading Jeff Davis in #39 to distinguish validation of generated tables from investigation of Unicode-version differences.
- Peter Eisentraut's detailed review in #45 led to generator cleanup, more precise terminology, and separation of pure catalog renaming from provider behavior in #48.
- Peter Eisentraut argued for unconditional and language-insensitive SpecialCasing rules in #47; Jeff Davis added full mappings, Final_Sigma handling, and a proposed UCS_BASIC change in #49, then introduced callback-based titlecasing and withdrew the UCS_BASIC behavior change in favor of PG_UNICODE_FAST in #50.
- The standard-collations documentation component of #50 landed with wording edits as 875e46a0a2, and the remaining series in #51 added optimized ASCII property lookup and a separate UTF-8 inlining patch.
- The property tables from #51 landed with additional update-unicode documentation as ad49994538, while 5c40364dd6 extracted only simple case mapping from the broader case-conversion patch.
- Jeff Davis reorganized the uncommitted work into independently reviewable stages in #52, and its catalog rename landed with documentation and test cleanup as f696c0cd5f.
- Peter Eisentraut's review in #53 led to #54, removing the POSIX spelling and unnecessary pg_utf_mblen duplication; its C-only builtin provider landed as 2d819a08a1 with additional validation and command-line fixes enumerated in #55.
- Further review by Peter Eisentraut in #56 produced the repair component of #58, committed with light documentation and test adjustments as 846311051e.
- Tom Lane identified unreachable locale-validation code in #60, prompting 61f352ece9, and then reported a missing-return warning in #62, prompting the control-flow repair 60769c62dc.
- The C.UTF-8 component of #58 landed with conversion-buffer and documentation changes as f69319f2f1, followed by its unchanged UTF-8 inlining component as 9acae56ce0.
- Peter Eisentraut requested version tracking in #57 and a simple '1' version string in #67; #68 included that change and a cache refactoring, but Jeff Davis deferred the refactoring as too invasive for PostgreSQL 17 in #80.
- Alexander Lakhin demonstrated an out-of-bounds read during lower() in #71, and Jeff Davis fixed convert_case bounds handling and added focused unit tests in 503c0ad976, confirming the repair in #72.
- The versioning component of #68 landed with light edits as 46a44dc372, and its unicode_strtitle refactoring landed with allocation and iterator-reset corrections as 46e5441fa5.
- Peter Eisentraut objected to initcap changing '123abc' into '123Abc' in #69 and favored a separate function in #74; Jeff Davis deferred full mappings and PG_UNICODE_FAST in #81, leaving naming, UCS_BASIC, regex interactions, casefold, and richer titlecasing for further work.
- Thomas Munro reported UTF-8 test failures under a LATIN1 environment in #84; e2a2357671 forced C environment locales in both initdb and createdb tests, extending the initdb-only patch attached retrospectively in #85.
- The full-mapping and PG_UNICODE_FAST work deferred in #81 subsequently continued in thread #50736, whose January 2025 patchset #579938 supplied 286a365b9c and d3d0983169; those landings belong to the later thread, with revised titlecasing, expanded tests, and UCS_BASIC left unchanged.
Jun - Jul 2024 · 3 entries
- Noah Misch reopened the stability question in #86 and demonstrated potential expression-index and partial-index inconsistencies after a Unicode update in #98, arguing for preserved behavior or an explicit upgrade safeguard.
- Tom Lane argued in #99 that IMMUTABLE need not prohibit all major-release changes, while Jeff Davis moved the broader Unicode-update policy discussion to a separate thread in #107; Laurenz Albe emphasized caution about index-affecting changes in #118.
- Peter Eisentraut, Joe Conway, and Jeremy Schneider supported retaining the PostgreSQL 17 feature in #122, #123, and #124, and Noah Misch closed the open item as a non-bug in the thread's final message, #125, without a revert.