Object3D: Fix world matrix update bug in .attach()

[takahirox]

Description

This PR fixes a bug of world matrix update in Object3D.attach() and adds a unit test

Currently Object3D.attach() moves an object to under a new parent while keeping its world matrix values by doing the following.

1. Calculate the object's new position/quaternion/scale/local matrix
2. Call Object3D.updateWorldMatrix()
3. Add an object to the new parent

But Object3D.updateWorldMatrix() should be called after adding an object to the new parent. Otherwise, the object's world matrix will be wrong because the world matrix is calculated as the object is under the old parent.

This PR moves Object3D.updateWorldMatrix() after the line adding an object to the new parent.

You can check this change works by using the unit test this PR adds. Without the change in Object3D.attach() the test fails while with the change the test passes.

[takahirox]

No Object3D.updateWorldMatrix() call may be another option because object.matrixWorld values are maintained. (I thought calling it is for safety.)

[WestLangley]

Note that object.add() does not update the world matrices. So is that a 'bug', too?

I think this PR is an improvement, in any event.

[takahirox]

Note that object.add() does not update the world matrices. So is that a 'bug', too?

Yeah, .add() doesn't update the world matrices. I speculate it's intentional for optimization. Because in most of use cases user doesn't directly care world matrices. And the world matrices are efficiently updated in renderer (if scene.autoUpdate is true). So updating the world matrix at the time an object is added may be inefficient.

Anyways,

I think this PR is an improvement, in any event.

I think so too. Calculating the world matrix as the object is under an old parent doesn't make sense.

[takahirox]

I replaced object.updateWorldMatrix( false, false ) with object.updateWorldMatrix( false, true ) to update children's world matrices.

Another option is cutting off oblect.updateWorldMatrix() call because the world matrix should be maintained so .attach() should not have an effect to the object's and children's world matrices.

What is the original purpose of calling it in .attach()? Matrix4.decompose() is used inside so that we take care of the case where .decompose result will be wrong due to parents' non-uniform scale as mentioned in #3845? In that case, the world matrix values can change then yes we need to call object.updateWorldMatrix( false, true ).

[mrdoob]

Thanks!