Fix springbones physics interpolation stutter

[dog-on-moon]

When using physics interpolation, springbones do not calculate their target transforms using the correct transform. This causes noticeable stuttering when the skeleton's transform is being influenced by physics interpolation.

Before fix:

godot.windows.editor.x86_64_fsQFeN8k1g.mp4

After fix:

godot.windows.editor.x86_64_fYMlwB9s0F.mp4

Demo project: springbones_stutter_demo_2025-08-27_14-11-43.zip

[dog-on-moon]

Also I think the code here will also need a similar fix (noticed similar jittering with colliders), but Node3D::get_global_transform is const and Node3D::get_global_transform_interpolated isn't (which the compiler is upset about), and I'm not C++ savvy enough to know what the preferred way of addressing it is :P

godot/scene/3d/spring_bone_collision_3d.cpp

Lines 152 to 160 in 4ebf67c

	Transform3D SpringBoneCollision3D::get_transform_from_skeleton(const Transform3D &p_center) const {
	Transform3D gtr = get_global_transform();
	Skeleton3D *sk = get_skeleton();
	if (sk) {
	Transform3D tr = sk->get_global_transform();
	gtr = tr.affine_inverse() * p_center * gtr;
	}
	return gtr;
	}

[lawnjelly]

Node3D::get_global_transform is const and Node3D::get_global_transform_interpolated isn't (which the compiler is upset about), and I'm not C++ savvy enough to know what the preferred way of addressing it is :P

You may be able to simply make get_global_transform_interpolated() const, this is likely historical from before SceneTreeFTI (might be used by the ifdeffed backup path, but that's just for testing).

This is very probably fine but I haven't looked at all at this class (it may not be present in 3.x), so it could do with checking by someone more familiar with SpringBone3D.

Often things like this are processed on physics tick in which case the non-interpolated is fine, but if this is causing a problem presumably it is being called on the frame. get_transform_interpolated() should work either way and fallback to the normal method if called on a physics tick.

BTW I would also double check this function is not being called on invisible nodes, as get_global_transform_interpolated() can get rather expensive when called on invisible nodes (as the xform has to be calculated through the chain on demand).

[Repiteo]

Thanks!