GameActivity PATCH: fix deadlocks in java callbacks after app destroyed

[rib]

This ensures that any java Activity callbacks take into account the possibility that the android_app may have already been marked destroyed if android_main has returned - and so they mustn't block and wait for a thread that is no longer running.

Fixes: #97

[rib]

I've at least tested this with the agdk-mainloop example modified to exit early and so I've seen that successfully call onPause, onStop and onDestroy after the android_app was marked destroyed.

I'm going to go ahead an land this since I'd like to move on to landing some of the bigger changes like GameActivity 2.0.2 before the next Winit beta release.

@ardocrat it'd be good if you can test this to confirm it fixes the deadlock you were hitting.

Feel free to comment here even if though I'm going to merge this optimistically now.

[ardocrat]

Thanks, this case seems fixed now! After calling of eframe::Frame::close() I can see onDestroy() passed.

But now I have problem with winit, seems like EventLoop is not finishing, I got an error after relaunch: Creating EventLoop multiple times is not supported, I am using egui/eframe = 0.22.0 and winit = 0.28.

My log:

21:50:33.404  D  loadNativeCode_native
21:50:33.406  V  Creating: 0x722f9271f0
21:50:33.406  V  Callbacks set: 0x722f927240
21:50:33.406  V  Launching android_app_entry in a thread
21:50:33.406  V  android_app_entry called
21:50:33.406  V  android_app = 0x7137dbf040
21:50:33.406  V  config = 0x717f939d10
21:50:33.406  V  activity = 0x722f9271f0
21:50:33.406  V  assetmanager = 0x721f93fcc0
21:50:33.406  V  Config: mcc=250 mnc=99 lang=en cnt=GB orien=1 touch=3 dens=560 keys=1 nav=1 keysHid=3 navHid=0 sdk=30 size=2 long=2 modetype=1 modenight=2
21:50:33.434  V  onStart_native
21:50:33.434  V  Start: 0x722f9271f0
21:50:33.442  I  thread '<unnamed>' panicked at 'Creating EventLoop multiple times is not supported.', /home/user/.cargo/registry/src/git.luolix.top-1ecc6299db9ec823/winit-0.28.6/src/event_loop.rs:116:13
21:50:33.442  I  note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
21:50:33.443  E  Creating EventLoop multiple times is not supported.
21:50:33.443  D  write_work: cmd=1
21:50:33.443  V  android_app_destroy!

I guess this problem is related to winit exactly. I don't see if they are resetting this static variable somewhere:

I think I will kill process after onDestroy(). Thank you, @rib!

[rib]

I think I will kill process after onDestroy()

Just in case you might ever want to run other Android Services in the same process I'd try and avoid explicitly killing the process if you can avoid it.

The lifecycle for the activity is separate from the lifecycle of services that could be running in the same process.

On the Winit side of things then things should be a bit better in 0.29 hopefully, re: rust-windowing/winit#2900