Faster CI, part 2: Windows build without waiting, VM test only for code the guest builds #42
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "ci-faster-2"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Follow-up to #41. Run 152 took 10:00 against 10:44 before; the jobs mostly waited for a runner slot.
windows-build without
needs. The installer and portable zip job (about 5 min) waited for windows-sysroot and windows-gtk only to share their caches. Forgejo queues a job withneedsbehind every job that was ready at the start, so it started about five minutes late. Withoutneedsit is sixth of the nine jobs ready at the start. On a cache miss (newmsys2.lock, GTK recipe or tools) build-gui.sh fetches the packages and builds GTK itself; that run builds GTK twice. windows-gui-test keeps itsneeds(short job, would be a third GTK build).VM test trigger. vm.yml's pull request paths now leave out what Linux never compiles into ogc, the helper or the btrfs tests: unit-test files (
tests.rs,tests/,*_tests.rs,tests_*.rsand four named files, all#[cfg(test)]) and modules declared#[cfg(windows)]. The list was checked against rustc's dep-info for exactly tests/vm/build.sh's build: 121 files excluded, none of them compiled; every compiled file still triggers (src/platform/windows/scheduler.rsis built for Linux and taken back in). Inlinemod tests {}changes still trigger. Nightly and manual runs are unchanged.Docs: ARCHITECTURE.md CI section and README updated.
Expected: CI about 7 min; a PR that touches only Windows code or tests no longer waits 18 min for the VM test.