test(apps.just): BATS coverage for install-opentabletdriver and cncf

[hivecommons-hive]

Claimed ground

File under test: system_files/shared/usr/share/ublue-os/just/apps.just
Recipes claimed: install-opentabletdriver, cncf — only these two.
New file: tests/test_apps_just.bats (21 cases)
Also touched: Justfile (one added line in the test recipe)

Disjointness check against open PRs

apps.just is partially occupied ground, so this PR deliberately claims only
the recipes no open PR touches:

Open PR	Ground it holds in apps.just	Overlap here
common#978 [scanner] fix: use valid brew tap trust commands	brew tap --trust lines in install-jetbrains-toolbox, install-asus	none
common#941 fix(brew): replace invalid brew tap --trust (+ tests/test_brew_tap_trust.bats)	same brew tap --trust lines	none
common#1048 [quality] test(default.just)	default.just / tests/test_default_just.bats	none
common#1034 [quality] test(shared.just)	shared.just / tests/test_shared_just.bats	none
common#1031 [quality] test(system.just)	system.just / tests/test_system_just.bats	none
common#1063 [architect] setup hook dispatch	ublue-*-setup, hookrunner.sh	none
common#1045 [architect] image-name → upstream routing	bonedigger-report, changelog.just	none

install-jetbrains-toolbox and install-asus are not covered by this PR.
(The earlier scope-boundary test about brew tap was dropped in the
2026-09-21 revision: #941 has merged and #978 is closed, so that carve-out no
longer exists.)

Justfile is also edited by #1048 and #1031, but each adds its own line to the
same test recipe list — a trivial textual conflict at worst, no semantic
overlap.

What this adds

apps.just (76 lines) had zero test coverage: no file under tests/
referenced it, and neither the Justfile test recipe nor
.github/workflows/unit-tests.yml executed anything against it.

tests/test_apps_just.bats extracts each recipe body into a standalone script
and runs it against a sandboxed PATH of mocks (gum, curl, sudo,
flatpak, systemctl, brew, ujust) with /etc/udev/rules.d,
/etc/modprobe.d, /usr/share/ublue-os/homebrew and $HOME redirected into a
per-test temp tree. This follows the existing pattern in
tests/test_update_just.bats. No network, no root, no host mutation.

install-opentabletdriver

• gum confirm exit 0 → install branch; exit 1 → uninstall branch; exit 130
  (Ctrl-C) → neither branch runs and no flatpak call is made
• the jq asset filter picks the .tar.gz release asset and not the .deb
  or .rpm decoys
• the udev rule is copied 70-opentabletdriver.rules → 71-opentabletdriver.rules
• blacklist hid_uclogic / blacklist wacom land in
  blacklist-opentabletdriver.conf
• the mktemp -d extraction directory is removed
• flatpak --system install / flatpak --system remove are issued on the
  correct branches
• the user unit is written to $HOME/.config/systemd/user/ and enabled
• uninstall makes no network calls and does not create the user unit

cncf

• brew bundle --file=.../cncf.Brewfile is invoked with the curated Brewfile path
• ujust --choose is skipped when stdin is not a tty, and the recipe still
  exits 0 (the || true tail)
• a failing brew bundle does not abort the recipe (no set -e in this body)

Bug found while writing these tests

install-opentabletdriver writes the modprobe blacklist to
/etc/modprobe.d/blacklist-opentabletdriver.conf on install, but the uninstall
branch removes /etc/modprobe.d/blacklist-opentabletdriver.rules — a different
filename. Uninstalling leaves hid_uclogic and wacom permanently
blacklisted, so the built-in tablet/Wacom drivers stay disabled after the
user has removed OpenTabletDriver.

Per hold-gated quality scope this PR does not fix the recipe. The test
install-opentabletdriver: uninstall targets the wrong modprobe filename (regression guard) pins the current (buggy) behaviour and carries a comment
saying to flip the assertion when the recipe is corrected. Filed separately as
an issue.

Not included

The matching step in .github/workflows/unit-tests.yml was prepared but had to
be dropped: the hive GitHub App token has no workflows permission, so pushing
a workflow edit is rejected. The suite is wired into the Justfile test
recipe only. A one-line CI step is needed as a follow-up:

      - name: Run bats (apps.just — install-opentabletdriver, cncf)
        run: bats tests/test_apps_just.bats

Verification

$ bats tests/test_apps_just.bats
1..20
ok 1 .. ok 20   # all pass

Revision of 2026-09-21 (rebased onto main, hanthor's review addressed)

• The install-opentabletdriver uninstall path now removes
  blacklist-opentabletdriver.conf, the file the install path writes, instead
  of a .rules file that never existed. The test that documented the old
  behaviour as a "regression guard" is now the correct assertion: it fails on
  the old recipe and passes on the fixed one. This is hanthor's option (a).
• The stale brew tap scope comment and its guard test are gone (see above).
• .github/workflows/unit-tests.yml runs the new suite as its own step.

Fixes #1065

---

Filed by quality agent (hold-gated mode). Human review required — do not merge without review, and do not remove the hold label.

— hive: agent=quality backend=copilot model=claude-opus-5

[hanthor]

New tests/test_apps_just.bats only, scope matches the claim, exercises real control flow with mocked binaries, and correctly pins the existing blacklist uninstall mismatch as a regression guard rather than silently "fixing" it. CI test+E2E green; validate fails only on title-lint (bracket prefix), not a code issue.

---

Generated by Claude Code

[hanthor]

Tests verified green and non-vacuous; two things to fix before merge

I merged this into current main, resolved the Justfile conflict, and ran it:

$ bats tests/test_apps_just.bats
1..21
ok 1 install-opentabletdriver recipe body is extractable and non-empty
...
ok 21 cncf: a failing brew bundle does not stop the recipe (no set -e)

Full suite after merge is identical to the main baseline — same three pre-existing failure groups (test_motd_integration 1/2/4, test_shared_just 2/3, test_ujust 7-10/12), no new failures. These 21 are pure addition.

I spot-checked the one test that looked like it could pass vacuously — "install removes the temp extraction directory" guards its assertion behind if [ -n "${tmpdir}" ], which is the shape that usually hides a no-op. It is genuinely wired up. Deleting the code under test makes it fail:

$ sed -i 's|^      rm -rf "${OTD_TMPDIR}"$|      : rm-removed-for-mutation-test|' .../apps.just
$ bats tests/test_apps_just.bats -f "removes the temp extraction directory"
not ok 1 install-opentabletdriver: install removes the temp extraction directory
# (in test file tests/test_apps_just.bats, line 233)
#   `[ ! -d "${tmpdir}" ]' failed

The bug you found is real, confirmed on main:

$ grep -n "modprobe.d" system_files/shared/usr/share/ublue-os/just/apps.just
35:      echo -ne "blacklist hid_uclogic\nblacklist wacom\n" | sudo tee /etc/modprobe.d/blacklist-opentabletdriver.conf
47:      sudo rm -f /etc/modprobe.d/blacklist-opentabletdriver.rules /etc/udev/rules.d/71-opentabletdriver.rules

Writes .conf, removes .rules. Uninstall leaves hid_uclogic and wacom blacklisted forever.

1. Please don't pin the bug as the contract

Test 16, install-opentabletdriver: uninstall targets the wrong modprobe filename (regression guard), asserts that the stale blacklist file survives uninstall. That encodes a user-facing breakage (built-in tablet/Wacom drivers stay disabled after removal) as the expected behaviour, and it means the eventual one-character fix arrives as a red CI run that a future reader has to decide is "expected red". The comment says to flip it, but comments don't run.

Two better options: either (a) drop test 16 and land the one-line fix (.rules → .conf) in this PR — the risk is a sudo rm -f of a file this recipe itself owns, and you already have the coverage to prove it; or (b) keep the test but invert it to the correct assertion and mark it skip "blocked on <issue>", so it flips to green when the fix lands instead of flipping to red.

2. The scope-boundary rationale is stale

The header comment and PR body say install-jetbrains-toolbox / install-asus are excluded because they hold brew tap --trust lines contested by open PRs #978/#941. #941 already merged:

$ git log --oneline -3 -- tests/test_brew_tap_trust.bats
2281f4f fix(brew): replace invalid `brew tap --trust` with `brew tap` + `brew trust` (#941)

$ sed -n '10,14p' system_files/shared/usr/share/ublue-os/just/apps.just
install-jetbrains-toolbox:
    #!/usr/bin/env bash
    brew tap ublue-os/tap 2>/dev/null || true
    brew trust ublue-os/tap 2>/dev/null || true

main has no --trust left in system_files/, and tests/test_brew_tap_trust.bats passes on main. So the carve-out no longer protects anything, and test 3 (apps.just recipes covered here do not call brew tap) is now guarding a boundary that doesn't exist. Worth refreshing the comment and, if you want, extending coverage to those two recipes in a follow-up.

3. Mechanics

• mergeable_state is dirty, but only trivially — the Justfile conflict is your added bats tests/test_apps_just.bats line against main's test_system_just / test_brew_tap_trust lines. All three belong; keep all three.
• The missing .github/workflows/unit-tests.yml step matters more than the PR body implies: without it these 21 tests run only under just test locally and never in CI. Please get that one-line step in (a maintainer can push it to this branch) rather than merging a suite CI does not execute.

---

Generated by Claude Code

[Danathar]

Rebased onto current main (the Justfile conflict was the three added bats lines; all three are kept) and pushed a follow-up commit for the three points in @hanthor's review, head is now 5a035e8.

1. Took option (a). apps.just uninstall now removes /etc/modprobe.d/blacklist-opentabletdriver.conf, the file install writes, instead of the .rules name (fixes [quality] install-opentabletdriver uninstall removes the wrong modprobe file — wacom/hid_uclogic stay blacklisted forever #1065). The old "regression guard" test is now the correct assertion: seed the .conf, run the uninstall branch, assert it is gone. I checked it fails on the old recipe and passes on the fixed one.
2. Refreshed the scope comment at the top of tests/test_apps_just.bats (the brew tap --trust carve-out is gone since fix(brew): replace invalid brew tap --trust with brew tap + brew trust #941 merged) and dropped the test that guarded that boundary. 20 cases now.
3. Added the Run bats (apps.just ...) step to .github/workflows/unit-tests.yml, right after the bonedigger step, so CI runs the suite.

Local run at 5a035e8:

$ bats tests/test_apps_just.bats
1..20   # all ok

[Danathar]

What I checked at 5a035e8: read the whole diff (new tests/test_apps_just.bats, the Justfile and unit-tests.yml wiring, and the one-line apps.just uninstall fix). The suite extracts the two recipe bodies and runs them against mocked gum, curl, sudo, flatpak, systemctl, brew and ujust with /etc and $HOME redirected into a per-test tree, so nothing touches the host. Ran bats tests/test_apps_just.bats locally, 20 ok; reverting the apps.just fix makes the uninstall test fail, so it is not vacuous. tests/test_brew_tap_trust.bats and tests/test_native_recipes.bats still pass. CI at this head is all green, including the test job that now runs the new step and the composed E2E suite. I pushed the rebase and the follow-up commit that addresses hanthor's three points myself, so a second maintainer look at 5a035e8 is fair; second approval needed from castrojo or hanthor.