Callback from animation tree gets called forever if in the end of an animation

[fbcosentino]

Godot version:

3.2.1 stable

OS/device including version:

Windows 7, Intel 64bit, GLES2

Issue description:

If an animation comes from AnimationTree, and the track has a callback in the end of the animation, that callback will be called in repetition forever. If the animation is called from the AnimationPlayer directly it doesn't happen. If the callback is not in the end of animation, it doesn't happen as well. Only happens when called from AnimationTree and the key is in the end of animation. Tested with state machine root only.

Steps to reproduce:

• Create AnimationPlayer
• Create animation
• Add callback track
• Add keyframe invoking a callback in the end of the animation duration
• Create AnimationTree
• Setup for the AnimationPlayer and state machine
• Add the animation as start animation

callback will be called in infinite loop

Minimal reproduction project:

In the example below two callbacks are set. The only difference is callback 1 key frame is in the middle of animation, callback 2 is in the end. Callback 1 gets called only once, callback 2 gets called forever. AnimCallbackTest.zip

[Sl3dge78]

Here are a few results from my preliminary investigation:

In the Animation Tree code, there is a check that doesn't call the method if delta is zero : https://github.com/godotengine/godot/blob/05395cb14fa0fbd270f6d8c63bcb7b80f45f4dfe/scene/animation/animation_tree.cpp#L969-L972

Seems great, however, delta is almost never 0 ; it still has a value set when going from 1 to 1 because it doesn't get updated line 106 here (step is the delta): https://github.com/godotengine/godot/blob/05395cb14fa0fbd270f6d8c63bcb7b80f45f4dfe/scene/animation/animation_blend_tree.cpp#L88-L107

So far I've tried two things :

• Change step to be the real delta (ie if we go outside the range of the animation by .2 with a delta of .3, it gets set to .1) > The method still gets called every frame
• Force change step (to try) to be 0 if we go overbounds > Method never gets called; obviously.

To me the issue is that a new Animation State doesn't get created if we go from 1 to 1 (because it's not a change, so the blend_animation doesn't get called), and as we read the delta from the states, it always reads a value, so never skips the method call.

• We could say that not triggering a method on the last frame of an animation is ok. So get a range that's [0 - End[. But that's dumb
• I'll try looking into creating an Animation State that goes from 1 to 1 with a step of 0, but it doesn't seem super clean and I don't know if it could create other issues with other track types...
• Calculate the delta in an other way and change the check in Animation Tree. But i don't know to what. Maybe recheck if time has changed and not trust the value in the animstate? But that would make the presence of delta in the animstate useless...
• Or I'm probably just missing something obvious.