Remove degenerate polygons in PolySetBuilder

[kintel]

Also includes some PolySetBuilder refactoring.

Fixes #5000

Note: rotate_extrude() could be improved to better handle manifold corner cases. Deferring to #5046

[kintel]

@pca006132 In case you have any comments :)

[pca006132]

going through PolySetBuilder will lose manifoldness.

[kintel]

Riight, this is the wrong hammer for this job; we need something that cleans up a user-specified polyhedron. I'll think about it, perhaps just tessellate to triangles early for user polyhedrons.

[pca006132]

why should we have this if it can be called automatically?

[kintel]

Really just for API cleanliness, as most client code will use begin/end pairs to make it more readable.

[kintel]

We could make it required and rather assert if someone tries to do two beginPolygon() in a row.

[pca006132]

We could make it required and rather assert if someone tries to do two beginPolygon() in a row.

Sounds good.

[pca006132]

I think we probably should not do this in the PolySetBuilder, e.g. manifold outputs can have degenerate triangles but we don't want to pass it through PolySetBuilder as this loses manifoldness as well.

[pca006132]

Another concern (not specific to this PR): we are using assert for checks, but we do not build in debug mode in the CI, so these checks are not run automatically.

[kintel]

I think the only issue here is PolyhedronNode. Apart from that, this only affects code already using PolySetBuilder, and it's a cheap feature. The goal of the assert() is to catch programming errors, not run-time errors, so it's really just a convenience for developers.

That said, there might be a better/more general place to insert a polygon cleaner which doesn't collapse manifolds.
Btw., do you know a good nomenclature to describe this type of objects (manifolds where multiple different vertices have the same position in space)?

[pca006132]

Btw., do you know a good nomenclature to describe this type of objects (manifolds where multiple different vertices have the same position in space)?

No, but @elalish may know

[jordanbrown0]

we are using assert for checks, but we do not build in debug mode in the CI, so these checks are not run automatically.

Tidbit: It's been a long time, but the last time I tried to build in debug mode for MSYS2/Windows, the build failed - it hit some kind of Windows symbol table limit.

[elalish]

I'm not sure it has a specific name - it's just an artifact of various different manifoldness definitions floating around. I wrote up the major differences on our wiki. The short answer is that anything that doesn't allow disjoint vertices at the same location cannot maintain manifoldness through Boolean operations in general. It's closely related to why STL is not capable of reliably transmitting manifolds either. Hence why 3MF should be used instead, and now even glTF with my new EXT_mesh_manifold extension.