ugly hack to work around #502

[pca006132]

#502 (comment)

This is an ugly hack to workaround issue #502, which seems to affect quite a lot of OpenSCAD users when they extrude from SVG and perform boolean operations.

The idea is that #502 is caused by skipping vertices incorrectly, so we try to triangulate the polygon without considering precision first, and triangulate it the second time if the first attempt failed. This makes some of the geometry checks necessary, so I modified some of the debug related code to make things compile and working without MANIFOLD_DEBUG=ON. Note that the performance with MANIFOLD_DEBUG=OFF is still very fast because we only enabled some checks in polygon.cpp, instead of all other checks that MANIFOLD_DEBUG=ON will enable.

What I am uncertain about:

1. Should we also do this when params.overlapChecks = false?
2. Should we merge this before we get a real fix for Difference of Cube and Complex SVG Results in Incomplete Render (openscad/openscad#4679) #502? Or should we wait a bit?

[pca006132]

I tried building openscad with this patch, and it did work.

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.07% 🎉

Comparison is base (7facd0e) 90.36% compared to head (09c1185) 90.44%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #507      +/-   ##
==========================================
+ Coverage   90.36%   90.44%   +0.07%     
==========================================
  Files          35       35              
  Lines        4431     4447      +16     
==========================================
+ Hits         4004     4022      +18     
+ Misses        427      425       -2     

Files Changed	Coverage Δ
src/utilities/include/public.h	67.44% <ø> (+0.38%)	⬆️
src/polygon/src/polygon.cpp	90.85% <100.00%> (+1.04%)	⬆️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[elalish]

Okay, I am more and more confused as to what this bug actually is. Is it a crash, or non-manifold output, or geometrically incorrect output? Are the images above correct or incorrect output?

[elalish]

Okay, rereading the original openscad bug, I think I've answered my own question: no crash, all output is manifold, but the geometry is off because of a bad triangulation. Let me see if I can fix the underlying problem before we put in a hack. Thanks for all the effort narrowing this down!