Gamepad and joystick input is broken

[rinOfTheStars]

What it says on the tin—while SDL recognizes and attaches itself to joysticks/gamepads properly, it seemingly does not properly poll input in gamepad mode, and trying to use joystick mode instead results in only ever receiving -32768 from Axis 4, regardless of the device used.

I can upload an example project (as a .zip; I don't want to set up a repository on github for the project I'm working on just yet) if that would help with figuring out what's wrong here.

[rinOfTheStars]

Update: looks like it's broken everywhere, or at least on all arm64 (little endian, maybe?) platforms. Get either no output at all or nonsense outputs from SDL running on a raspi box (so arm64 linux).

[revmischa]

Thanks for the report! I've been investigating this.

A few clarifying questions:

About the -32768 on Axis 4: On Linux, trigger axes (axis 4 and 5 on most controllers) report -32768 at rest using the raw Joystick API. This is expected behavior - triggers use the full i16 range from -32768 to 32767, with -32768 being "released" and 32767 being "fully pressed."

The Gamepad API should normalize these to 0-32767 instead, which is often more intuitive.

Questions to help debug:

1. When you say gamepad mode "doesn't properly poll input" - do you mean you're not receiving any events at all, or that the values are wrong?
2. For joystick mode - do other axes (0-3, the analog sticks) work correctly? Or is it just axis 4 that seems stuck?
3. What specific controller are you using?
4. Could you share a minimal example that reproduces the issue?

If you're only seeing -32768 on the triggers and not moving to higher values when pressing them, that would indeed be a bug. But if you're seeing -32768 at rest and the issue is that the Gamepad API isn't giving normalized values or any events at all, that would point to a different problem.

Also, there's a known upstream SDL3 issue (#11471) about axis value scaling on certain controllers, though that one is more about precision than values being completely stuck.

[rinOfTheStars]

Yeah, it’s only reporting the non-normalized Axis 4 (or Axis 4 and 6 on a controller with two triggers) when using the joystick subsystem, not reporting when the trigger is pressed with a different value, and not reporting any values at all when polling each named button/axis/hat in gamepad mode.

The three* devices I’ve tested are:

• 8BitDo Pro 3 (in all three modes, so both it pretending to be a NS or Xbox-via-XInput controller, and over their own version of the generic HID gamepad stuff); over both Bluetooth and USB on macOS, and over USB on Linux
• Xbox Adaptive Joystick, by itself over USB (so as a Game Input Protocol device); this only works on Linux for macOS jank reasons
• Xbox Adaptive Controller, both with and without the Joystick attached, and over both Bluetooth only on macOS and just USB on Linux again; iirc it acts like a regular HID device on Bluetooth and as a GIP device on USB

[rinOfTheStars]

There appears to have been some sort of... honestly I'm not sure what to call this; I guess it sort of counts as a regression? Or maybe as progress? In between 0.17.0 and 0.17.2: The Xbox Adaptive Joystick is incorrectly reported as a 'Generic Xbox Controller' and constantly reports -663 on the left x axis, and -2146 on the left y axis, every 'tick' over the gamepad api, instead of the previous bad value seen on the joystick api. Something similar occurs with the 8bitdo controller as well; though it reports -129 on the right y and left x axes, and 128 on the right x and left y axes instead. No other values are reported on either api, regardless of the actual state of the physical hardware

[revmischa]

I don't think anything related to joystick/gamepad was changed between those versions