SkeletonIK3D bone roll axis is not working correctly

[fracteed]

Godot version

4.0.2

System information

Windows 10 960m

Issue description

When using the skeltonIK3d on a simple bone chain, the roll axis of bones are not working correctly. As can be seen in this clip comparing 3.5 to 4.0.2, when moving the ik target on the x axis, the bone roll of the shin bone twists in a very strange way.

When the ik target is just moving up and down on the y axis 4.0.2 looks the same as 3.5. I have tried many ways of setting the roll axis in blender before exporting, but this seems to make no difference in godot.

ik_roll_bug.mp4

@lyuma I also tried this with your recent fix in master and it does not fix this issue, unfortunately. I did try the scorpion demo scene and I can see that it resolved that. Simple 2 or 3 bone chains seem to exhibit this odd roll issue...maybe the long chain on the scorpion tail is covering up the issue?

On a side note, I am getting tons of crashes trying to test out simple ik with the standard tool script. Especially just clicking in the skeletonik3d node.

Steps to reproduce

• import a simple 2 or 3 bone skeleton such as a leg
• add a skeltonik3d node
• fill in the node parameters and add a marker3d as target
• add a tool script to start off the ik solve
• move the ik target on the x axis and observe the twisting

Minimal reproduction project

This has the 3.5, 4.0.2 file and the blend file

ik_roll_bug.zip

[GeorgeS2019]

@fracteed confirmed the reported bug using the latest commit (less than 1 day)

[lyuma]

Thanks for all the testing, and for clarifications in the comments on #77469. Sorry I forgot to post an update last week, but I made progress and verified that this was indeed a regression in 4.x that was not present in 3.x and seems to still be an issue regardless of humanoid retarget and regardless of the other bugfixes.

I was out for a week but I'm pretty confident I will be able to get to the bottom of this issue by the time 4.1 is released. I'll keep you all updated.

[fracteed]

Thanks for the update @lyuma . I am very much looking forward to this, and will do some extensive testing when there is a new PR!

[lyuma]

Can you verify the updated PR on #77469 and see if it works for you? The issue was 3.x called Basis::rotated() and 4.0 calls the new rotate_to_align, which must do a left-multiply to match Basis::rotated.