Before, CI ran the test suite on Linux only. Now it also runs the tests of how Obeya starts processes on a Windows runner, so the Windows-only fault fixed in “Windows: a quick process no longer waits for a slow one started next to it” can't come back unnoticed.
On Windows, a process Obeya started next to a slow one (a fetch, a push, a hanging gh call) inherited the slow process's pipes. Its output wasn't finished until the slow one ended. The fix makes Windows start one process at a time. The contributor measured this on Windows 11 (four 8 s processes, 16 quick calls among them):
| Job | Runner | Runs |
|---|---|---|
| test | Linux | the whole suite, type check (unchanged) |
| windows-processes (new) | Windows | spawn.test.ts and pr-watcher.test.ts, including “gh slow to answer” and the Windows-only test of the fix |
| release | Linux | now also waits for the Windows job |
Timeouts aren't raised: bun run test scales them on any Windows machine, as it already did. Only these two files run on Windows because a few other tests there still fail. Getting the whole suite green on Windows is the proposed card “CI runs the whole test suite on Windows too”.
| macOS, full suite and type check | passes |
| Windows runner | first run once this lands on main (CI runs on pushes to main only) |