Fix test_window_state_assignment testbed on macOS-arm64

[proneon267]

Fixes #3016

The testbed test was failing since the test assertion was being done even before the window state change was completed. The failure only affected macOS-arm64, since a longer delay before assertion is required on macOS-arm64 for fullscreen transition. Since, we cannot differentiate between macOS-x86_64 and macOS-arm64, so on the cocoa backend the delay for fullscreen is 1s now.

I apologise for causing any inconveniences.

PR Checklist:

• All new features have been tested
• All new features have been documented
• I have read the CONTRIBUTING.md file
• I will abide by the code of conduct

[proneon267]

@mhsmith could you please confirm if this PR fixes the test failure on your local machine?

[mhsmith]

No, it doesn't, and even increasing the 1 to 2 or 5 doesn't make any difference.

See attached screen recording. My dock is hidden on the left edge of the screen, but you can see when a window minimizes to it.

Screen.Recording.2024-12-04.at.21.55.31.mov

[freakboy3742]

FWIW: I don't see this on my machine - I only see the issue in CI. I'm on macOS 14.7; CI indicates its running on 14.7.1.

[freakboy3742]

I'm unclear how this change would impact the failure mode that has been reported. The test failure is on L812; the preceding wait_for_window call explicitly sets state_switch_not_from_normal=True. So... shouldn't this be a change to the 1.75, not the line modified here?

[proneon267]

My running theory about this error is that the window cleanup state change isn't getting enough delay in case of fullscreen when it finishes running the first test case.

toga/testbed/tests/window/test_window.py

Lines 764 to 769 in df65161

	# Complex state changes on cocoa:
	(WindowState.MINIMIZED, WindowState.FULLSCREEN),
	(WindowState.FULLSCREEN, WindowState.MINIMIZED),
	# Complex state changes on gtk:
	(WindowState.FULLSCREEN, WindowState.PRESENTATION),
	(WindowState.PRESENTATION, WindowState.NORMAL),

toga/testbed/tests/conftest.py

Lines 87 to 114 in df65161

	async def window_cleanup(app, app_probe, main_window, main_window_probe):
	# Ensure that at the end of every test, all windows that aren't the
	# main window have been closed and deleted. This needs to be done in
	# 2 passes because we can't modify the list while iterating over it.
	kill_list = []
	for window in app.windows:
	if window != main_window:
	kill_list.append(window)
	# Then purge everything on the kill list.
	while kill_list:
	window = kill_list.pop()
	window.close()
	await main_window_probe.wait_for_window("Closing window")
	del window
	# Force a GC pass on the main thread. This isn't perfect, but it helps
	# minimize garbage collection on the test thread.
	gc.collect()
	main_window_state = main_window.state
	main_window.state = WindowState.NORMAL
	app.current_window = main_window
	await main_window_probe.wait_for_window(
	"Resetting main_window",
	minimize=True if main_window_state == WindowState.MINIMIZED else False,
	full_screen=True if main_window_state == WindowState.FULLSCREEN else False,
	)

I tried to increase state_switch_not_from_normal delay to 3, but it still fails. So, I am going to try 2 solutions:
* Add a waiting delay to the start of the test
* Manually reset the window state to NORMAL at the end of the test

My earlier guess was incorrect, a delay is required when closing the window from a non-normal window state.

[freakboy3742]

The complication here is that I can't reproduce the problem locally; apparently @mhsmith can though, so he may be able to provide better confirmation that this works.

However, this fix doesn't make a whole lot of sense to me. I can see how an end-of-test minimized state might cause problems - but that's not where the problem test was failing - it was failing at L812, after a wait_for_window has successfully completed asserting the initial state of the window under test, and a wait_for_window which should allow for the final state to be applied.

Yes, it's passed CI; but the problem was intermittent; I'll poke it a few times to see if I can manifest a failure.

[freakboy3742]

It's not obvious to me why this is possible

[proneon267]

Since, get_window_state() is being called in window_cleanup, so it might be getting called after window.close() is called on the window. Hence, leading to a situation where self.app might not exist.

[freakboy3742]

I can't argue with success though... 6 CI passes and no failure. I'll wait to see if the problem is resolved for @mhsmith, but otherwise this is looking promising.

[mhsmith]

Works for me. I don't know why my machine was different from yours, since I'm also on macOS 14.7. Maybe it's something to do with my screen configuration.

I can see how an end-of-test minimized state might cause problems - but that's not where the problem test was failing - it was failing at L812

Since it was failing on the second parametrization, maybe it was caused by insufficient cleanup from the first one?