Android: add command mecanism to SDLActivity

[1bsyl]

Add some RPC commands so that SDLActivity defers most of SDL code internal execution it to the SDL main thread
see #14925

Most SDLActivity code is deferred to main SDL C thread with some command RPC.
except:

• lifecycle event (which remains in main SDL C thread).
• code to count recreation of activity because it is SDLActivity only.
• code to initialize (nativeSetupJNI / nativeQuit), because it occurs before C Thread.
• code which doesn't return void. (nativeGetHint()). because, value is needed immediately .
• nativeInitMainThread() / nativeInitMainThread(), already in C thread.

(SDL_Window *)Android_Window global variable is removed.
Android ActivityMutex is removed.
Android_LifecycleMutex, is used to lock the SDLActivity state. (Android_WaitActiveAndLockActivity()).
since it prevents the SDLActivity to add new lifecylce message, it prevents it to change state.
factorisation of EGLSurface/native_window handling. (use in surfaceChanged/Created/Destroy, and SDLCreate/DestroyWindow()

It seem to work quite nicely from my side. I tested various stuff, but not fully.
(onNativeFileDialog() was the less easy to update, and I really have to tested).
other functions are always simple. I guess, the ways it's written with named struct, prevent lot of type mistake.

Still a draft. please give a try.
I'll try it also in production maybe in a while

maybe 2 todo:

• I didn't add the HID native functions .. I am not familiar with them. but then didn't rely on SDL and use no locking.
  so maybe it makes not difference. (they could be added anyway afterwards).

• we probably lose some event timestamp precision because event are sent from Main thread, and not SDLActivity thread. maybe it's not a big deal.
  but if we need it for some command, we could add. (just enabled the timestamp, and then send it with the event).
  maybe we wait to see, if this is really useful

[1bsyl]

PR is still as draft because I need to test it in production,
but you can give a try as it is fully functional

[slouken]

Crash if video isn't initialized (also below)

[slouken]

Suggested change

	int android_keycode_to_SDL(int keycode)
	int Android_Keycode_to_SDL(int keycode)

[slouken]

The joystick inputs are specifically designed to be thread-safe and low latency, so they should probably bypass the RPC queue.

[1bsyl]

Updated with suggestion.

maybe we can add hint define to have the 2 differents behaviors. handling with rpc shouldn't be that low

[slouken]

We still need to pump events for lifecycle activity even if there's no window or video hasn't been initialized.

[slouken]

Updated with suggestion.

maybe we can add hint define to have the 2 differents behaviors. handling with rpc shouldn't be that low

It still involves an extra thread context switch which could be delayed due to scheduling. We don't need a hint, we just need the gamepad events to be passed through immediately.

[1bsyl]

ok,
are all the following concerned ?

 onNativePadUp,
  onNativeJoy,
  onNativeHat,
  nativeAddJoystick,
  nativeRemoveJoystick,
  nativeAddHaptic,
  nativeRemoveHaptic

?

[slouken]

ok, are all the following concerned ?

 onNativePadUp,
  onNativeJoy,
  onNativeHat,
  nativeAddJoystick,
  nativeRemoveJoystick,
  nativeAddHaptic,
  nativeRemoveHaptic

?

Yes, it would be all of the gamepad related functions. Really only the inputs are time sensitive, but we don't want input coming before nativeAddJoystick() has been processed, etc.

[1bsyl]

I've added a preprocessor define to toggle between gamepad under SDLActivity (now the default) or main C thread

[madebr]

This branch was (accidentally) created on the libsdl-org/SDL repo.
Depending on how long this PR will remain alive, perhaps close and reopen from a personal fork?

[1bsyl]

oops, sorry I didn't notice that.
I'll fix that tomorrow :)
thanks!

[1bsyl]

ok new PR is there ! thanks for catching

#14962