docs/testing.md asks every shell test to start with set -euo pipefail and source base-test.sh through BASH_SOURCE; without it a failed mktemp leaves test_home empty and reset_scenario goes on to rm -rf /data.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
EasyEffects plays into Bluetooth headphones through the same direct port links as into a card, but only alsa_output targets were followed, so with headphones on the keys fell back to the settings file -- which can still name the speakers -- or to easyeffects_sink. Match bluez_output sinks as well, and compare the configured device name literally rather than as a regex.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The layout check skips comments, but the already-fixed check matched `include system-auth` anywhere on a line, so a comment such as `# auth include system-auth` made the migration skip a stack that still listed bare pam_unix, and it was marked complete with the lockout still missing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`sudo tee` truncates the live PAM file before writing it, so a migration interrupted in between left /etc/pam.d/polkit-1 empty; the next run then failed the layout check, exited 0, and was marked complete with polkit authentication broken. sed -i writes a temporary file and renames it into place, as the setup commands already do, so the original stays intact until the rewrite is whole and a failed sed needs no restore.
The test's broken-write stub now empties the file instead of discarding the write, so the restore assertion fails if the restore is removed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex Medium <noreply@openai.com>
No suite failed with the strip restored, so nothing kept the 4.0.3 regression from coming back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Bar widget instances are not handed sourceDir (ModuleSlot.injectProps sets only bar, moduleName and settings), and a plugin could reach its directory through Qt.resolvedUrl, so the comment's rationale was wrong on both counts. Two lines carry what the code cannot say.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fake nodes carried only id and name, so a projection that handed back the node passed every assertion. Giving one an object-valued audio property, as a live PwNode has, makes the deep comparison catch it.
Co-Authored-By: Codex Medium <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
userdel rewrites /etc/subuid and /etc/subgid, and like every shadow-utils database write it leaves the previous contents behind in a dash-suffixed backup. The scrub removed four of the six backups those tools produce, so the retained @factory baseline still named the previous owner in /etc/subuid- and /etc/subgid- along with their subordinate ID range.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
