timescale.info#217 · Fix the release's Rocky Linux build, and desktop tests hanging after they pass
- timeline-app
- pull request
- packages
The v0.1.0 release failed on **Desktop / Rocky Linux 10 / RHEL 10 RPM**: ``` Error: Command failed: git rev-parse HEAD fatal: detected dubious ownership in repository at '/__w/timeline-app/timeline-app' ``` That job builds inside a container, where the checkout belongs to a different user than the one running git. `actions/checkout` marks the directory safe only in a temporary global config, so git refuses later commands. `scripts/desktop/build.mjs` checks that the checkout matches the release tag, and that check only runs when `RELEASE_TAG` is set, so PR CI never exercised it. The git calls now pass `-c safe.directory=*`, as `scripts/build-info.mjs` already does. Checked in a Fedora container run as root against this checkout. Rocky 10 itself won't run on the dev machine because its CPU lacks x86-64-v3. Plain `git rev-parse HEAD` reproduces the error; the new form returns HEAD and `v0.1.0^{commit}`. ### Desktop tests hanging after they pass On the first CI run of this PR, both "UI interactions within budget" jobs failed. In each, `test/desktop-bulk.mjs` printed `PASS` and then hung until its 5-minute step limit. Killing tauri-driver left something holding its output pipes open (WebKitWebDriver or the app), which kept Node running. `startTauriDriver` in `bench/ui/webdriver.mjs` now starts the driver in its own process group, and `stop()` kills the whole group and closes the pipes. The bulk test and the UI benchmark both use it. I couldn't reproduce the hang locally. With this change, local bulk-test and desktop-benchmark runs exit and leave no driver or app processes behind. A separate mid-test flake, which also happens on main, is filed as #218.