Skeleton IK 3D Interpolation

[Taiseibutsu]

Godot version

4.0.alpha6.official [e4f0fc5]

System information

Arch Linux, Nvidia Geforce 1080Ti

Issue description

Skeleton Ik 3D Interpolation does not assign the interpolation of the IK chain.
It assign it to 1 and only controls the override tip rotation.

0001-0313.mp4

Rotation also seemed identical as override tip magnet solution

0001-0313.mp4

Steps to reproduce

Normal use of
Skeleton3D.interpolation or Skeleton3D.set_interpolation()

Minimal reproduction project

skeleton3D_interpolation.zip

[Calinou]

Is the behavior in Godot 3.4.4 considered correct here?

[Taiseibutsu]

Yes, previous video was using magnet, that was it looked weird, the video for reference now it's default IK, used as it's similar on how the rotation tip behaves.

referenced old video with magnet usage:

0001-0313.mp4

[fire]

There is two systems for IK in godot engine.

The SkeletonIK3D is the older system. The newer system is pending modifications due to review by @lyuma and @TokageItLab.

So both are in limbo.

[TokageItLab]

@fire There is similar issue in #59554. CC @TwistedTwigleg

[lyuma]

I'll try and take a look at this, since I intend to keep SkeletonIK3D for 4.0 and it should be able to work the same as 3.5

My guess is these issues are casued by the change to make bone_pose include the bone_rest, while they were separate in 3.x

[patrykkalinowski]

Hey, is this planned to be fixed in 4.0 or do we wait for replacement system?

[lyuma]

Did a little experiment to port SkeletonIK3D to GDScript where the same bug can be replicated. It took only an hour or so.
skeletonik3dimpl.gd

I know it doesn't solve this bug, but if I'm really stuck, my plan is to do the same to the 3.x implementation. it may be easier to replicate the same behavior or debug if I have both implementations side-by-side.

People weren't particularly fond of SkeletonIK3D in 3.x either, but it might be a starting point.

[lyuma]

I found the bug.

I'm attaching a zip with the test 4.1 project with SkeletonIK3D node, GDScript port of 4.0 skeleton_ik_3d and GDScript port of 3.x skeleton_ik_3d (the only difference is how the rest pose forward direction is calculated)
skeleton3D_interpolation.zip

I already applied the bugfix to the gd files in the attached projects. Search for if true: and bring back the original condition or make it if false to reproduce the bug.

And finally, I performed a backport by hand of the tscn to 3.x so we can run it on the classic engine and confirm behavior is the same, which is now is.
skeleton_interpolation_3.x.zip

The cause of the bug is get_bone_pose_global_no_override was inadvertently broken in 2021 due to a typo in the forward port of 3.x #48251 to master #48166 , and then the typo was concealed later in #51368 due to removing what then appeared to be duplicate code.

See the bug here in skeleton_3d.cpp
3.x:

	if (b.parent >= 0) {
		b.pose_global = bonesptr[b.parent].pose_global * pose;
		b.pose_global_no_override = bonesptr[b.parent].pose_global_no_override * pose;

master:

	if (b.parent >= 0) {
		b.pose_global = bonesptr[b.parent].pose_global * pose;
		b.pose_global_no_override = bonesptr[b.parent].pose_global * pose;