model-writtenrun 6990 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane opened this thread with an NLS regression test exercising PRI* formatting macros through Spanish translations, reporting success on Linux but uncertainty about Windows locale naming #1.
- Bryan Green explained that newer Windows gettext versions expect a Windows locale name and can repeatedly perform expensive unsuccessful locale enumeration; Tom deferred that issue to Bryan's separate work #2 #3.
- CI exposed BSD codeset requirements and unsupported j formatting on FreeBSD and macOS, so Tom revised locale selection and added a separate snprintf fix that maps j to long or long long according to intmax_t size, then corrected its Meson header dependency #3 #4.
- The snprintf fix from #4 landed independently as f8715ec866, preserving its implementation while expanding the commit explanation.
- After CI rejected database-derived codeset selection, Tom submitted the remaining test using es_ES.UTF-8, explaining that local testing supported this choice and that the exercised translations contain only ASCII characters #5.
- At the last message, the revised test passed CI and the snprintf fix had landed, but the test remained pending because Windows CI and buildfarm configurations did not exercise NLS; Tom considered proceeding without further Windows verification #6.
- The NLS test subsequently received a Solaris LANGUAGE mapping in thread #52854, whose later submission #608973 matches 8c498479d7 exactly and supersedes attribution to #5; both changes originating here ultimately landed, while Windows NLS verification remained unresolved in this thread.