Degenerate Triangles will result in NPE

[dbt-padberg]

Describe the bug

There are usages of Three.js's Triangle.getInterpolation(...) in this library that can result in a Null-Pointer-Exception.

The documentation (https://threejs.org/docs/index.html?q=Triangle#api/en/math/Triangle.getInterpolation) says:

Returns the value barycentrically interpolated for the given point on the triangle. Returns null if the triangle is degenerate.

What is a degenerate triangle? https://en.wikipedia.org/wiki/Degeneracy_(mathematics)#:~:text=A%20degenerate%20triangle%20is%20a,angle%20and%20two%20undefined%20angles.

E.g. https://github.com/search?q=repo%3Agkjohnson%2Fthree-mesh-bvh+intersection.normal&type=code

In my case the code or my meshes have provided a degenerate triangle which returned null.

It could be that other usages are affected too:
https://github.com/search?q=repo%3Agkjohnson%2Fthree-mesh-bvh+getInterpolation&type=code

To Reproduce

Sadly I am currently not able to provide an example. Hopefully I can provide some in a few days.

Expected behavior

There should be null checks and a logically fallback. Or at least some precheck to prevent this error.

Platform:

• Device: Desktop
• OS: MacOS
• Browser: Chrome
• Three.js version: ==0.169.0
• Library version: ==0.8.1 (+ three-bvh-csg == 0.0.16)

[gkjohnson]

Hello! Thanks for the report. An example would be useful here - though CSG will likely not be working nicely with degenerate triangles in the first place and should probably be removed before operating. However I understand that's not always easily doable or may even occur during a previous operation.

Up until about a month ago this is how three.js' raycasting worked, which this library aims to reproduce the results of exactly. The new barycoord attribute sampling implementation I added to three.js about a month ago should solve this issue but will instead result in a 0, 0, 0 barycoordinate value (and therefore 0, 0, 0 value for interpolated uv, normal, etc).

I'd prefer not to make breaking changes to accommodate such a recent three.js release (eg we should avoid using the getInterpolatedAttribute function) but if you'd like to submit a PR that updates the the raycasting to align with three.js' behavior (ie just add some null checks) we can get that merged.