Clear other axis that is in deadzone threshold

Clear the second axis when an axis event causes a second axis to be within the deadzone threshold.

I'm seeing drift on multiple gamepads due to what I believe is an issue with the deadzone filter, as describe by the below flow, with tiny_* here meaning drift values).

1. You press the gamepad axis - two events are received by gilrs
   1. axis_x = large_value_x - gilrs stores that as state, and sends event through
   2. axis_y = large_value_y - gilrs stores that as state, and sends event through
2. You release the gamepad axis - but it has a slight drift. Now, if the x axis event comes before y (and same thing the other way around), gilrs receives these two events:
   1. axis_x = tiny_value_x - gilrs sees that the new axis value would be new_axis_state = (tiny_value_x, large_value_y), and checks magnitude(new_axis_state) > deadzone_threshold, so it updates state and sends a axis_x=tiny_value_x event through.
   2. axis_y = tiny_value_y - gilrs sees that the new axis value would be new_axis_state = (tiny_value_x, tiny_value_y), and checks magnitude(new_axis_state) < deadzone_threshold - so it sets new_axis_state = (tiny_value_x, 0.0), and sends a axis_y=0.0 event.

Now the gilrs state is incorrectly (axis_x, axis_y) = (tiny_value_x, 0.0), and the latest axis_x event delivered was axis_x = tiny_value_x, which causes drift.

This change only applies the fix if update_state is true, as otherwise we risk an infinite loop, since we do not know what state we currently have. A more invasive change could be to add a field like Event::synthetic , so that the deadzone filter knows not to send the additional event once it encounters a synthetic event. Let me know if you want that approach, or some alternative solution.

changed the description

Thanks! Adding anything to Event would be breaking change and I would prefer to fix this keeping current API.

One other way to fix this without requiring update_state == true would be to keep filter specific state in GamepadData. If I'm not mistaken, we only require to keep one bool for each pair of left sticks, right sticks and dpad to determine if we already reset other axis value.

added 1 commit

• 8ff72413 - Make deadzone filter work better

Compare with previous version

@Arvamer Thanks! I have updated the PR to store flags in GamepadData if the last sent values were non-zero for each axis, and used that to know if they need to be reset. What do you think?

added 3 commits

• 83d327c4 - 1 commit from branch gilrs-project:master
• 66757c3f - Clear other axis that is in dead zone threshold
• ed73e630 - Make deadzone filter work better

Compare with previous version

enabled an automatic merge when the pipeline for ed73e630 succeeds

Looks good!

merged