Subtraction creates overlapping triangles about manifold

[pca006132]

Oh, I think it is probably coplanar triangles:

The reason I noticed this is because the offset result becomes a giant cthulhu... I just debugged it and found that the bug is caused by zero thickness cases causing the algorithm to blow up. Maybe I should close this?

[pca006132]

Thinking about it, this should be closed as the result is geometrically valid.

[pca006132]

I tried disabling tbb, change Split to use the public boolean API (instead of reusing Boolean3), and enabled address sanitizer. The result is the same.

[pca006132]

I did not optimize the decomposition algorithm, so it tends to generate cuts that precisely touch some triangles. I will try to optimize the algorithm, but it also reveals some issues in our boolean.

[elalish]

Can you be a touch more specific regarding what the overlapping triangles are? Is your only evidence that render? Technically what this library intends to generate is ε-valid geometry, which means there exists a perturbation of the vertices each by less than precision that means all points in R3 have winding number either 1 or 0. Do you believe this condition has been violated? It would also be nice to avoid folds, where triangles become coplanar, but this cannot in general be avoided. Is that the situation?

[pca006132]

The overlapping stuff are just the back faces, verified by meshlab.