Triangulation issue: Zebra

[kintel]

The following OpenSCAD script yields a bad Manifold triangulation:

$fn=24;
text(text="R",font="naturePro",size=5);

Note: This single font glyph consists of ~120 separate outlines with >62K vertices.

Saving as SVG and reading that back yields the same issue (SVG attached for convenience).

The output after enabling MANIFOLD_DEBUG and PolygonParams().processOverlaps = false:
zebra.cc.zip

[elalish]

Still digging into what's wrong with this one, but I can see why it's a tough customer:

Amazingly this is not self-intersecting. However, this is also a great example of why the even-odd fill rule is such a disaster. Strictly positive fill rule would have made a much more sensible polygon.

[elalish]

@pca006132 Unfortunately minimize_testcase fails for this polygon - it returns a single inverted triangle, but as far as I can tell there's nothing with incorrect winding in the whole thing. I'd guess it's getting confused by the touching vertices.

[pca006132]

Make sense, that code was assuming non-intersecting (touching) polygons. Should try to make it more general one day...

[elalish]

Can we get a new polygon dump, since something must have changed compared to our existing tests?

[pca006132]

Yeah, I think it is probably related to precision.

[pca006132]

Trying to reproduce, so far I cannot get the result that @kintel got.

[pca006132]

cannot reproduce downstream anyway