Tatsh / sbclaude / 32930334449

100%

push

github

project: require a Linux host, and stop building for the others

It was never going to work anywhere else. The host copy of claude is
bind-mounted into the container and executed there, so the host has to
supply an ELF build of it, and a macOS host has a Mach-O one that cannot
run in a Linux container at all. The identity mirroring reads
os.getuid(). Every device and socket the run flags pass through
(/dev/nvidia*, /dev/dri, /dev/kvm, /dev/bus/usb, /tmp/.X11-unix,
/run/user/<uid>, /var/run/usbmuxd) is a Linux one. Windows has neither
os.getuid nor syslog, and syslog is what finally broke the build: the
PyInstaller job died on "No module named 'syslog'" while testing the
binary it had just produced.

supported_platforms is now ['linux'], and that is what deletes
pyinstaller.yml. Its matrix held macOS and Windows and nothing else, so
there was nothing left to keep. The same setting drops it from the
required-workflow gates in publish.yml and release.yml, which would
otherwise leave every release waiting on a workflow that no longer runs.
Linux binaries are untouched: appimage.yml still builds x86_64 and
aarch64 and has its own release-assets job.

The CLI now refuses to start on any other host, and says why, rather
than failing later with an exec format error from docker. -h/--help and
--version still work, because click handles eager options before the
group callback; a test pins that, since moving the guard would otherwise
break it silently. Another pins that the refusal covers every
subcommand, not just run.

The classifier is hand-placed rather than generated. Wiswa produced the
rest, but this machine has no user defaults file, so the run also
blanked the author in pyproject.toml and CITATION.cff, the funding
entries, the social badges, and the snap donation link. All of that was
reverted, and package.json deliberately still records the earlier run
rather than claiming a clean one. `classifiers+:` in .wiswa.jsonnet
concatenates after the std.set() t... (continued)

226 of 226 branches covered (100.0%)

Branch coverage included in aggregate %.

5 of 5 new or added lines in 1 file covered. (100.0%)

868 of 868 relevant lines covered (100.0%)

5.0 hits per line