[FR] improve bad diagnostic for `__REMOVED_IN` on `ALooper_pollAll` about ndk HOT 8 CLOSED

is plain old __attribute__((__deprecated__(("use foo instead")))) a better choice then, since it includes a user-provided diagnostic?

from ndk.

__attribute__((availability)) supports that too. We don't expose that in __REMOVED_IN() though, and that's the fix we need.

from ndk.

oh, yeah, that seems like a bug in __REMOVED_IN() --- i vote we require an explanation...

from ndk.

Don't know why I didn't think of that, or why we didn't do that in the first place :) +1

from ndk.

Status? SDL will not compile for android because of this bug

from ndk.

Status? SDL will not compile for android because of this bug

this bug won't fix any builds --- this bug will just give you a compiler diagnostic explaining that you need to switch from ALooper_pollAll() to ALooper_pollOnce().

from ndk.

but, yes, it looks like even SDL 3.0 relies on ALooper_pollAll(): https://github.com/libsdl-org/SDL/blob/12b6c17575a4b24539b7d17625232fff899c90cf/src/sensor/android/SDL_androidsensor.c#L73

from ndk.

It also uses ALooper_wake, so this is not a false positive. That's a bug in SDL.

EDIT: It may not be. The bug happens if you use pollAll in combination with both events signaled by callbacks and those signaled by wait. There's definitely a wait in that code. Whether there's a callback or not is not depends on how the sensor API is implemented in the OS, I think.

Okay, looking even more closely, it's not just in combination with callbacks, and wake and any other event, so yes, that probably does have the bug. See the discussion post I added to the OP for more info.

from ndk.