SIGSEV @ SDL_joystick.c:2446

[isXander]

Android build of SDL3 causes a SIGSEV when disconnecting a controller.

void SDL_UpdateJoysticks(void)
{
    /* this needs to happen AFTER walking the joystick list above, so that any
       dangling hardware data from removed devices can be free'd
     */
    for (i = 0; i < SDL_arraysize(SDL_joystick_drivers); ++i) {
        SDL_joystick_drivers[i]->Detect(); // THIS LINE
    }
}

Attached is the log file generated by the JVM with info that may help.
hs_err_pid12511.log

I wouldn't know how to debug this, so please let me know what more info you need.

[slouken]

Are you able to reproduce this with testcontroller? I can create an apk of that if that's helpful.

[isXander]

That would be quite helpful. (aarch64)

[slouken]

Here you go!
https://www.libsdl.org/tmp/testcontroller.apk

[AntTheAlchemist]

@isXander I'm not able to reproduce this at all. Which gamepad caused this? Was it USB or Bluetooth? Which Android device? Are you able to reproduce it with a minimal test?

[isXander]

This happened on my PR branch. Maybe it could have possibly introduced a regression?

I was not able to reproduce on testcontroller. I think I may know why:

If you notice, when disconnecting a controller on Android, it takes up to 3 seconds for testcontroller to notice this and remove it. android/SDL_sysjoystick.c#ANDROID_JoystickDetect is the code responsible for this delay. It polls every 3 seconds.
Side note: the comment within this method says you have to do it this way because the API to hook correctly is >=16, but the android-project has a minimum API of 21?

My game polls events at 20 ticks per second, calling SDL_PumpEvents() as a result. My theory is that as a result, SDL tries to interact with the controller after Android no longer provides it via SDL_UpdateJoysticks. Taken from the JVM crash log I sent in the original issue post, here is the stack:

Native frames: (J/j=Java code, C=native code)
C  [libSDL3.so+0xcaee8]  SDL_UpdateJoysticks+0x22c (SDL_joystick.c:L2446)
C  [libSDL3.so+0xa8210] (SDL_events.c:L1402)
C  [libjnidispatch.so+0x14050]
C  [libjnidispatch.so+0x10a0c]
C  [libjnidispatch.so+0x7560]
C  [libjnidispatch.so+0x11024]
C  [libjnidispatch.so+0x141e4]
J 26320  dev.isxander.sdl3java.api.events.SdlEvents.SDL_PumpEvents()V (0 bytes) @ 0x0000006ff9f635e0 [0x0000006ff9f63540+0x00000000000000a0]
C  0x00000070938f9490

@isXander I'm not able to reproduce this at all. Which gamepad caused this? Was it USB or Bluetooth? Which Android device? Are you able to reproduce it with a minimal test?

This was on Android 15 (Pixel 9 Pro XL), with a Bluetooth-connected DualSense controller.

[slouken]

testcontroller polls every frame, probably around 60 FPS on that device. Are you thinking the lower rate is causing the issue?

[isXander]

I'm really not sure, could you provide the commands needed to build that testcontroller apk so I can have a mess around and see if I can repeat it? Any OS, whatever's easiest, I have access to them all.

[slouken]

Sure, I did something like this:

mkdir build
cd build
../build-scripts/create-android-project.py --variant symlink org.libsdl.testcontroller ../test/testcontroller.c ../test/gamepad* ../test/testutils* ../src/test/*

and then executed the gradle command that was spit out.