Avoid infinite recursion between AnimationPlayer and AnimationMixer

[matheusmdx]

In theory AnimationPlayer.current_animation shouldn't be setted directly but in normal cases that don't cause any problem unless the animation key set current_animation to an empty string, in this case will start an infinite recursion between AnimationPlayer calling AnimationMixer::_blend_process and AnimationMixer setting the current animation to a empty string which will never stop until the engine crashes. This fix makes AnimationMixer ignore these animation keys (that would do nothing anyways) and avoid the crash.

Fix: #98076

[TokageItLab]

It looks like the patch is too specialized for special cases IMO. How about separating set_current_animation into internal and exposed methods, and not allowing the exposed method to have an empty animation name?

[matheusmdx]

@TokageItLab I modified the pr to apply the changes u suggested, can you check again?

[TokageItLab]

I believe this is a clean way of fixing the issue.

I do have one concern. Looking at the summary of the issue #98076, does it only occur with an empty character and not when “[stop]” is set? If we can confirm that, I think this can be merged.

[matheusmdx]

Indeed the ["stop"] also triggers the issue, updated again and now i think is ready to merge

[AThousandShips]

Suggested change

	if (p_animation == "[stop]" || p_animation.is_empty()) {
	if (p_animation.is_empty() || p_animation == "[stop]") {

Prefer putting the cheaper condition first

[matheusmdx]

Done

[TokageItLab]

Indeed the ["stop"] also triggers the issue

In that case, we need to think about a different approach. This is because a set of "[stop]" could be called from the inspector.

BTW, is it a problem if stop() is called in a method track? How about a different approach, such as calling clear_cache() with deferred inside stop()?

[matheusmdx]

BTW, is it a problem if stop() is called in a method track?

Nothing happens, no crash but also the animation don't stop.