Correct bounding sphere transform for glTF 2.0.

[emackey]

Fixes #6784.

[cesium-concierge]

Thanks for the pull request @emackey!

• ✔️ Signed CLA found.
• ❔ CHANGES.md was not updated.
  • If this change updates the public API in any way, please add a bullet point to CHANGES.md.

Reviewers, don't forget to make sure that:

• Cesium Viewer works.
• Works in 2D/CV.
• Works (or fails gracefully) in IE11.

---

I am a bot who helps you make Cesium awesome! Contributions to my configuration are welcome.

🌍 🌎 🌏

[lilleyse]

The pivot point is a bit off center in CV and 2D, but only for OffsetCube. It seems worse in 1.46 though, so maybe not relevant to this PR.

Also it would be good to get to the bottom of why the transform is applied first for computing the bounding sphere but last for the model matrix.

[lilleyse]

Besides that, this does fix the problem.

[emackey]

Also it would be good to get to the bottom of why the transform is applied first for computing the bounding sphere but last for the model matrix.

The (non-bounding-sphere) model matrix is being transformed by another matrix. The original model matrix is oriented for Cesium (+X forward, +Z up), then transformed to Y-up (still +X forward), and finally to +Z forward, +Y up. The resulting transformation matrix takes vertices expressed in that system and gets them into Cesium's preferred orientation.

The BoundingSphere calculations are taking a single point, the sphere's center point, and getting it into Cesium's coordinate system. So we're transforming a point by two different matrices, not stacking up the matrix multiplications into a single transform matrix.

For example, in the case of OffsetCube, the glTF mesh data is centered around [2, 4, 6], with 4 being "up" (glTF +Y) and 6 being "forward" (glTF +Z). We first transform +Z to +X, and get [6, 4, -2] as an intermediate center point. Then we transform +Y to +Z, and get [6, 2, 4]. Previously, 4 had been "up", and this is now +Z where Cesium expects. Previously 6 had been "forward", and this is now +X where Cesium expects.

[lilleyse]

Ah yeah that makes sense. In both cases the Z_UP_TO_X_UP transform is still applied first, but it just appears different because the model matrix stacks it last. Thanks!