Calling isActionPressed on an InputEvent instance crashes Godot

[PadraigK]

Thread 0 Crashed::  Dispatch queue: com.apple.main-thread
0   Godot                         	       0x108e13be4 0x104cb8000 + 68533220
1   Godot                         	       0x108e13b78 0x104cb8000 + 68533112
2   Godot                         	       0x108a2c01c 0x104cb8000 + 64438300
3   Godot                         	       0x108a4f064 0x104cb8000 + 64581732
4   libSwiftGodot.dylib           	       0x1295a299c InputEvent.isActionPressed(action:allowEcho:exactMatch:) + 1184 (InputEvent.swift:100)
5   libSimpleRunnerDriver.dylib   	       0x1131be654 PlayerController._input(event:) + 140 (PlayerController.swift:27)
6   libSwiftGodot.dylib           	       0x129951be4 _Node_proxy_input(instance:args:retPtr:) + 1092 (Node.swift:2115)
7   libSwiftGodot.dylib           	       0x129951798 @objc _Node_proxy_input(instance:args:retPtr:) + 12
8   Godot                         	       0x106ecbc9c 0x104cb8000 + 35732636
9   Godot                         	       0x106ef8438 0x104cb8000 + 35914808

override func _input(event: InputEvent) {
        event.isActionPressed(action: "hi")
    }

[PadraigK]

Note that calling Input.isActionPressed(action: "hi") works fine as a work around

[tishin]

There's an issue with reference handling in generated Node.swift:

let resolved_0 = gi.ref_get_object (args [0]!.load (as: UnsafeRawPointer.self))!
swiftObject._input (event: lookupLiveObject (handleAddress: resolved_0) as? InputEvent ?? InputEvent (nativeHandle: resolved_0))

As far as I can see, it expects the argument to be a pointer to Ref<InputEvent>, but actually gets a pointer to InputEvent itself, so I managed to fix it by simplifying argument handling:

let resolved_0 = args [0]!.load (as: UnsafeRawPointer.self)
swiftObject._input (event: lookupLiveObject (handleAddress: resolved_0) as? InputEvent ?? InputEvent (nativeHandle: resolved_0))

For some reason all refcounted objects get this wrapper in Generator:

if cmap.isRefcounted {
    argPrep += "let resolved_\(i) = gi.ref_get_object (args [\(i)]!.load (as: UnsafeRawPointer.self))!\n"
} else {
    argPrep += "let resolved_\(i) = args [\(i)]!.load (as: UnsafeRawPointer.self)\n"
}

I'm not sure if it's a general bug, or InputEvent handler is some sort of edge case scenario here.

[alicerunsonfedora]

This also appears to happen for touch events, too:

    public override func _input(event: InputEvent) {
        super._input(event: event)

        if event.isClass("InputEventScreenTouch") {
            LibAnthrobase.logger.debug("Received touch event: \(event)")
        }
    }

[migueldeicaza]

Sadly the commit history does not point out the reason for that change (be78878), and I do not recall the reason for that. It was too early on the binding development, and might have been an artifact of the way things worked in 4.0

I added a bad workaround as well when I was working on the editor in a similar area (651645e), and even filed a bug to get back to try to figure out the proper fix (#109).

I suspect the patch from @tishin is correct, and it should be much better than what we have now.

The Godot-CPP bindings use this ref_get_object, but I struggle to find out where it is being used, due to the convoluted usage pattern (include/godot_cpp/classes/ref.hpp

[migueldeicaza]

Ok, I have now verified that the C++ code does nothing with that binding code.

So I think that this can go in, and I will also revert the other incorrect workaround.