Change scene on InputEventJoypadMotion in _input(event) disconnects joy controller

[toralfno]

Godot version

v3.4.stable.arch_linux

System information

Arch Linux 5.15.2-arch1-1

Issue description

If i call on change_scene() in _input(event): if event is InputEventJoypadMotion the controller disconnects.

func _input(event):
	if event is InputEventJoypadMotion:
		get_tree().change_scene("res://Scene.tscn")

It don't happens on InputEventJoypadButton. And if i add a Timer before the change the controller stays connected.

I have a Global scene that is autoloaded to monitor controllers.

Steps to reproduce

Run project and press an analog input on your controller and joy_connection_changed() will print out the device that disconnected.

Minimal reproduction project

Controllerbug.zip

[Calinou]

By "the controller disconnects", do you mean the controller disappears entirely from the OS' point of view? Or does Godot just stop receiving all controller inputs (while still receiving keyboard/mouse inputs)?

[toralfno]

Godot just stop receiving controller input from one controller. If you replay the project the controller is back.

It will receive input from other controllers and keyboard/mouse.

[Calinou]

Can you reproduce this in Godot 3.3.4? See also #55037 and #54066.

[toralfno]

Weird... on 3.3.4 it only happens on L2 and R2 (triggers).

[toralfno]

Tried on another computer with 3.3.2. Same result as original bug.

[toralfno]

Setting Input.set_use_accumulated_input to true fixes the problem.

[AgentSmith0]

Setting Input.set_use_accumulated_input to true fixes the problem.

This also solved my issue: #53334

[akien-mga]

I'd like to make sure we understand what the problem was here in different Godot version.

Godot 3.3.2 and 3.3.4 mentioned above should both default to input accumulation being enabled, so Input.set_use_accumulated_input(true) shouldn't make a difference there.

Then Godot 3.4+ did regression and input accumulation was disabled by default, so it can make sense that calling Input.set_use_accumulated_input(true) would improve things.

But then I don't understand the report that 3.3.2 seems to have the same bug as 3.4?