[earakely-scale]

Summary

The computer probe's fixture could lose to an app's startup window on X11, and the probe then failed with P4 a click in frame space hits its cell.

The fixture relied on full screen to be on top. xfwm4 won't put a full-screen window above a window an app has put above. One app keeps a licence dialog above for about two seconds at startup. If the fixture appeared within those two seconds:

1. The dialog stayed over the middle of the fixture and took P4's first clicks.
2. A later click raised the fixture, and the remaining checks passed.

This is a race, so it fails some boots and not others. A stacking capture during a natural failure shows the dialog ABOVE and taking the click. The same cause fits an app that opens maximized at boot.

On X11 the fixture is now:

• maximized and -topmost;
• raised again, with the focus, when its <Visibility> says another window covers it, at most once every 0.5 s, so it can't fight a window that raises itself.

macOS and Windows keep full screen. The probe aims at the boxes the fixture reports, so the smaller, maximized area changes nothing for the checks.

Test plan

• New unit test, with Tk stood in for: on X11 the fixture is maximized and above, and takes the top back once covered. It fails on main and passes here. All 37 conformance tests pass.
• New integration test, tst/integration/computer/covered_test.py: an app whose boot window stays on top over the middle of the screen, run with agent-env run --no-emulate. It must reach ready with no failed check.
  • On main it fails with P4's frame, norm, double and crop clicks.
  • Here it passes, both on an arm64 Linux host and on local Docker on macOS.
• Real app on an arm64 Linux host:
  • Natural boots: 1 failure in 9 on main, 5 of 5 here.
  • With a window held on top: 2 of 2 failed on main, 3 of 3 pass here.
• After review: the widget that last had the focus gets it back when the fixture takes the top back, so keys the probe types into a text box still reach it; a new unit test failed first. The integration test now also requires no unverified checks and at least one passed; it passes on local Docker on macOS (53 s).
• Unit tier: the same 5 unrelated failures as main (cua driver, a GitHub clone test that needs credentials, a Windows disk recipe).

🤖 Generated with Claude Code

Confidence Score: 5/5

The PR appears safe to merge.

Summary

The PR keeps the X11 fixture above app windows and adds a covered-window regression test.

• The X11 probe fixture retakes the top when an app window covers it.

_{Reviews (7) · Last reviewed commit: "fix(computer): give the focus back to th..." · Reviewed by Greptile}