model-writtenrun 4665 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrew Dunstan introduced his Meson buildfarm-client work, with successful Linux runs, dynamically discovered test suites, and separate suite execution; Windows testing exposed a locale failure and unwanted touch and cp requirements #1.
- Andres Freund advocated parallel test execution with ordered failure reporting, while Andrew prioritized a working client and manageable load; Andrew also proposed removing unused touch discovery and considered Windows copy for Unicode updates #3 #5 #6.
- The Windows locale investigation ruled out environment stripping, identified a difference in existing buildfarm locale coverage, and prompted Juan José SantamarÃa Flecha to suggest testing de-DE instead of de_DE, without an explicit resolution of that diagnosis in the thread #7 #8 #9.
- Andrew subsequently reported a working Windows client including TAP tests, but upgrading Meson to 1.0.1 appeared to break it; Andres explained the TAP warning fix and asked whether an already-fixed PostgreSQL compatibility issue explained the failure, which Andrew said occurred on current master #10 #11 #12 #13 #14.
- Testing an existing buildfarm configuration, Andres found both a TestICU client error and missing test-extension files during installcheck; Andrew fixed the client error, while Andres proposed an explicit target to install test modules into the normal installation #15 #16.
- Andres supplied the installation patch, adding install-test-files, moving destination-path handling into the Python helper, supporting DESTDIR, and replacing directory-specific arguments with repeatable --install groups #17.
- For the separate Windows utility problem, Andres rejected relying on Windows copy and instead made Unicode updates conditional on cp availability while removing unused touch discovery; that attachment landed unchanged as 401874ab02, with other required Unix utilities left for possible later work #18 #19.
- Andrew's first installation-target test failed because he set DESTDIR to the existing installation prefix; Nazir Bilal Yavuz explained the duplicated path, and Andrew confirmed that removing that setting worked #20 #21 #22.
- Andres committed the installation attachment unchanged as 0d237aeeba, confirming Nazir's explanation and prioritizing unblocking buildfarm integration while leaving room for later refinement #17 #23.
- Andrew then achieved a clean Meson 1.0.1/MSVC run after switching to the Ninja backend, which was necessary for the new installation target, and planned deployment to buildfarm animals before a client release, with log-artifact polishing still needed #25 #26.
- Further integration exposed missing header-check targets, questions about PDF generation, and regression-database naming differences affecting cross-version upgrades; Andres had unfinished header-check work and explained the documentation targets, while Tom Lane suggested changing Make behavior and allowing upgrade support to wait #27 #28 #29.
- Andrew expected a small future AdjustUpgrade.pm change to handle database-name differences once a new branch made cross-version adjustments necessary #30.
- Andrew reported that the OSSP UUID header check failed on Fedora because it assumed an ossp/ subdirectory; Andres subsequently reported successful use of the Meson client on several animals, with status-page flags still needing attention #31 #32.
- Andres fixed the UUID check directly from discussion in 4e633016ab, changing the include to uuid.h because pkg-config already supplies the correct directory, and announced the fix #33.
- Andrew fixed the status page while noting that the failures page still needed work #34.
- The thread ended with Andrew diagnosing Windows PL/Python builds selecting the MSI-packaged Meson Python instead of his intended installation; installing Meson and Ninja through pip into an existing Python installation provided a workaround, with all three candidate changes landed and remaining buildfarm integration work outside the two attachments #35 #36 #37 #38.