Segfault in jnadispatch.dll about jna HOT 4 CLOSED

I read this https://wiki.libsdl.org/SDL3/SDL_Event:

typedef union SDL_Event
{
    /* [...]
       Visual C++ will respect the push pack pragma and use 52 bytes (size of
       SDL_TextEditingEvent, the largest structure for 32-bit and 64-bit
       architectures) for this union, and GCC will use the alignment of the
       largest datatype within the union, which is 8 bytes on 64-bit
       architectures.
    */
    Uint8 padding[128];
} SDL_Event;

My reading of this is, that SDL itself will generate events with a size of up-to 52bytes, but that you have to expect that in the future 128bytes could be filled.

The clashes with your definition of the SDL_Event, which is at most 24bytes .

Basicly this: https://github.com/isXander/Controlify/blob/5d5f80dac3de16302c9e8543b868a187d592d9db/src/main/java/dev/isxander/controlify/controllermanager/SDLControllerManager.java#L66 asks SDL for an out-of-bounds write.

Closing this as questions should be directed to the google group.

from jna.

You may also notice in the same class you referenced that there is an event filter that only accepts joystick connection/disconnection event, so JNA never allocates more memory than those events.

from jna.

The problem is not what JNA does, but what the SDL runtime does. You are calling SDL_PollEvent and essentially pass a pointer to a memory location. SDL is now free to write 128bytes to that pointer. BUT nothing is telling JNA to allocate enough memory to hold 128bytes. So even if native is only working on a structure with 24bytes, it will need to copy it around. As the event might change its shape anytime, native must copy the full structure and that in turn means, you have an out of bounds write.

from jna.

Thank you so much for the assistance and explanation regardless of the incorrect forum, after some testing it looks like this issue is now all resolved. Thank you again!

from jna.