ci: run the suite where a GNOME 50 desktop is, not where it isn't

[blyatiful1]

CI has been red on main since the v2 rebuild started — six runs, every one of them failing for the environment rather than for the code. This is what #4 stood down on, fixed properly.

What was wrong

The ubuntu job could not pass in any configuration. Three experiments, run against a matching ubuntu-24.04 box:

Configuration	Result
As it stands (no gi in the venv)	pytest aborts during collection — 2 errors
venv can see PyGObject	44 failed; 19 of them "your desktop session wasn't running"
…plus a private session bus	14 failed, all of them the runner being a 2024 desktop

• The gi problem is the venv, not the apt line. python3-gi lands in /usr/lib/python3/dist-packages, for the runner's own python3; the venv is built from actions/setup-python, whose --system-site-packages points at the tool-cache tree instead. The 22 "skipped" in that same log line were the Gio-backed tests skipping for the same reason — the tier was proving much less than it looked.
• No session bus means every settings write is skipped with your desktop session wasn't running, which is 19 failures that say nothing about the diff.
• ubuntu-24.04 ships gsettings-desktop-schemas 46. No accent-color key, no org.gnome.shell, so every bundled Look plans zero add-ons: hyperclass declares 6 add-ons and plans 0.

The archlinux container already had the right userspace and the session bus (dbus-run-session). It was only missing packages — its own ten failures were Namespace GnomeDesktop not available, three schema '…' is not installed, and an empty wallpaper catalogue.

What changed

ci.yml — the split moves rather than gets patched:

• tests (Arch container) runs the whole suite, -m "not sandbox", both tiers at once, under broadway + dbus-run-session. Package list gains gsettings-desktop-schemas, gnome-settings-daemon, mutter, gnome-shell (the schemas the pages read), gnome-desktop-4 (the GnomeDesktop-4.0 typelib the thumbnailer needs), gnome-backgrounds (the catalogue the Wallpaper page scans), and desktop-file-utils + appstream so the packaging tests don't skip themselves there.
• lint-and-packaging (ubuntu) keeps only what genuinely needs no desktop: ruff, and the three packaging suites.
• The header note now says why, so the next person doesn't re-split it the old way.

release.yml had the same broken pre-flight — a v2.0.0 tag would have failed before building anything. "Test before shipping" is now that same container job, and the wheel build needs: it.

tests/unit/test_applyrunner.py, tests/unit/test_window.py import gi at module scope under pytestmark = pytest.mark.gtk, but a marker deselects after collection, so -m "not gtk" never got the chance. They now carry the same pytest.importorskip("gi", …) guard the other gi-dependent modules already use.

Verification

Run on an ubuntu-24.04 box with the runner's own apt packages:

• Collection abort reproduced without the guard (Interrupted: 2 errors during collection), clean with it (1340/1417 tests collected).
• A venv built from the distro python with --system-site-packages does see gi — that's the mechanism the old job was missing.
• The three-configuration table above is measured, not inferred: 44 → 14 failures purely from adding dbus-run-session.
• The new ubuntu job passes as written: ruff clean, 31 passed, 1 skipped across the three packaging suites, with no PyGObject and no session.
• python -m build --wheel still produces the wheel.

What I could not run here is the Arch container itself (no container runtime in this environment), so this PR's own CI is the first execution of the container job. If the remaining three widget failures (ActionRow has no set_value, and two reset-button assertions) turn out not to be schema-caused after all, they're real findings and I'll chase them here.

Checklist

• python -m pytest -q passes — for the parts runnable without a GNOME 50 desktop; see Verification.
• python -m gtheme validate all passes — not run: needs the desktop this environment does not have.
• For theme PRs: no themes changed.

---

Generated by Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 90ee588b-999d-4a9e-bfbb-fd26c5108fcf

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}