abuki Reading it again in the morning, it seems to me that the "correct" behavior is also wrong as the documentation states that If the track is empty, it is equivalent to calling setAnimation.
Maybe there is a difference between track playing empty animation and track which is empty = cleared? That should probably be more differentiated in the docs.

Yes, this makes all the difference! These are two completely different things, and it's very important to understand the differece. If you assumed they were the same, this explains your confusion.

The empty animation is an animation where nothing is keyed. It is an animation, which leads to mixing in and mixing out. If you clear a track or have no animations on a track, or mixing has ended, you have no animation playing. When nothing is playing, no mixing happens.

abuki That should probably be more differentiated in the docs.

Thanks for the feedback, we will consider if we can improve the docs.

However IMO the documentation is already rather differentiated, looking at SetEmptyAnimation:

"Sets an empty animation for a track, discarding any queued animations, and sets the track entry's mixDuration. An empty animation has no timelines and serves as a placeholder for mixing in or out."

IMO that would be very strange and complicated wording if it meant the same as "clears the track". An animation is an animation, the empty animation is not no animation .

Also AnimationState reads:

"Applies animations over time, queues animations for later playback, mixes (crossfading) between animations, and applies multiple animations on top of each other (layering)."

This documentation page is also rather explicit about when mixing happens and what the empty animation is for:
https://esotericsoftware.com/spine-applying-animations/#Empty-animations

Agreed, we'll make those changes to the API reference!

@abuki Thanks for pointing out the exact misleading passages.

Regarding 1:
Yes, please note that track and track entry may sound similar, but are as different as road and car. An empty car on a road is not the same as an empty road.

This in general applies to all our documentation, we strive to choose the wording as technically precise and short as possible, we never use "approximations" just to sound better. In your case I assume that you missed the important documentation about what the empty animation is, so that might be the main cause (which we are trying to improve).

Regarding 2:
We're improving the spine-unity documentation currently in this regard (it's updated and improved, but not yet final):
https://esotericsoftware.com/spine-unity-main-components#Setting-Animations
https://esotericsoftware.com/spine-unity-main-components#Empty-Animation-and-Clearing
We hope that these changes prevent such misunderstanding and missing important documentation sections in the future.

abuki To mix back to the setup pose, set an empty animation with no mix duration. The note about needing one more frame when mix duration is zero is trying to explain that it will take one more apply to do the mixing out, and then the update after that will be able to "end" the track entry (the end AnimationStateListener event).

You don't normally need to worry about this. You just set the empty animation and continue on as normal.

The animation always starts at track time 0. Do you mean the first time the animation is applied the track time is already at 0.14 seconds, so you don't see the first 0.14 seconds of the animation? I'm not seeing that with:

state.getData().setDefaultMix(1f);
state.setEmptyAnimation(0, 0);
state.addAnimation(0, "walk", true, 0);

You're right, there were a few bugs regarding that! We've fixed it in 4.3-beta.

The beta won't be considered stable until it's no longer beta. Can't say yet when that will be.

We might backport this fix to 4.2, as it's pretty nasty. I'll check with the team.