GUI tests that poll the file model for a fixed 2 s fail under CPU load #24

Symptom

With every CPU core busy, tests/test_gui.py fails in 7 of 10 runs on the branch of #22 (6badf80), always at test_arrow_keys_move_the_preview (tests/test_gui.py:438, assert 0 >= 2: the explorer lists no .tif files). It passes when the machine is idle.

Cause

The test waits for QFileSystemModel to fill the folder with a fixed budget:

for _ in range(20):  # QFileSystemModel populates asynchronously
    pump(100)
    if model.rowCount(tree.rootIndex()) >= len(three_tifs):
        break

That is 2 s. The model fills the folder on a worker thread, and with the cores saturated the thread does not get to run in that time. test_arrow_keys_do_nothing_without_a_preview_tab (line 465) has the same loop, and several tests use a bare pump(200) or pump(400) before reading the tree.

#22 added a wait_until fixture for exactly this and applied it to the filter tests only.

Reproduce

On the branch of #22, with 12 busy-loop processes running (one per logical CPU):

QT_QPA_PLATFORM=offscreen uv run pytest -q -x tests/test_gui.py

3 passed and 7 failed in 10 runs. The filter tests (earlier in the file) passed in all 10, and in 20 more runs of -k "filter or escape or clearing".

Expected

The tests wait on the condition (wait_until(lambda: model.rowCount(root) >= n, describe=...)) instead of a fixed time, so a loaded machine slows them down without failing them. Because -x stopped each failing run at this test, later tests may have the same problem; the count is a lower bound.