Async all the things

[wmww]

Turns out screencopy frames weren't being captured because we were doing non-async waiting in the tests. This makes all our context managers (including notably Program) async, fixing the issue. Assertions are put in place to fail tests that do not get a minimum number of frames.

[Saviq]

The gedit test still feels suspect to me… how could it only be 5 frames when it blinks the cursor every half second for 10 seconds?

test	frames
test_active_app[confined_shell-asciinema]	25
test_active_app[confined_shell-mir-kiosk-neverputt]	94
test_compositor_alone[confined_shell]	7
test_inactive_app[confined_shell-qterminal]	9
test_inactive_app[confined_shell-gedit]	5
test_inactive_app[confined_shell-mir-kiosk-kodi]	22

[AlanGriffiths]

how could it only be 5 frames when it blinks the cursor

Does it? Is the cursor over the text area?

[Saviq]

how could it only be 5 frames when it blinks the cursor

Does it? Is the cursor over the text area?

I meant the text cursor, the caret. That one blinks as long as there's focus?

[wmww]

The gedit cursor does eventually stop blinking, but I see it blink about ten times before then.

However I'm currently seeing 22 frames for gedit on ubuntu-frame, and more frames for other servers.

[Saviq]

The numbers are: ci(xvfb) native(local) xvfb(local)

	confined-shell	mir_demo_server	mir-kiosk	ubuntu-frame
test_active_app[*-asciinema]	23 27 29	22 22 24	28 35 35	20 23 26
test_active_app[*-mir-kiosk-neverputt]	76 591 705	124 1869 989	99 2359 938	91 1203 805
test_compositor_alone[*]	9 6 6	2 2 2	10 10 10	2 3 3
test_inactive_app[*-gedit]	6 46 51	2 40 58	10 46 47	2 22 22
test_inactive_app[*-mir-kiosk-kodi]	19 163 89	17 181 94	25 164 94	14 154 84
test_inactive_app[*-qterminal]	10 13 13	5 7 8	12 19 17	5 10 9

So the gedit case looks weirdest.

[wmww]

Very strange, confined-shell specifically doesn't seem to be producing any output when run on XVFB. It produces output just fine when run on normal X.

[wmww]

I've gotten closer to the cause. on xvfb-run confined_shell fails to terminate and needs to be killed. Other shells do not, and confined_shell does not when run on normal X (or at least XWayland). This is strange but it's pre-existing, main has the same behavior.

However, something about getting output when a process is killed (not terminated) is broken on this branch. It's not that waiting after the SIGKILL is timing out. The process exits quickly after being killed, it just doesn't give any output. Continuing to investigate.

[wmww]

That ended up being a bit tricky, but I think I have a good system now. The problem was that with asyncio it's only reasonable to call process.communicate() once, even if that call is cancelled. With the new system it's called when the program starts, but not awaited until the program is done. There's also a task that sends the signals to kill and later terminate the program, and it's cancelled when the program is done.

[Saviq]

bors merge

[bors[bot]]

Build succeeded:

• ProcessSnaps
• RunTests (3.10, self)