Port bugfixes from JS

[kkaefer]

• Floating point precision can lead to incorrect triangulation earcut#16: mapbox/earcut@6547cc6 and mapbox/earcut@e457796
  • Test is failing even after the fix
• Race condition with tiny hole in huge polygon earcut#17: mapbox/earcut@d708db9
  • Test is failing even after the fix
• Strange issue triangulating certain polygons earcut#29: mapbox/earcut@44cc151
  • Test is failing even after the fix
• preserve colinear segments earcut#41: mapbox/earcut@deb0bfe
• Tessellation error with two holes (one touching outer vertex) earcut#45: mapbox/earcut@65c7da5

[jfirebaugh]

Other changes accumulated in JS:

• Do we need to port Kosta/fix index creation (fix for #13) earcut#14 and followups?
  • Nope, was already using indexing.
• Port mapbox/earcut@3839088
  • 3c1401c
• Do we need to port mapbox/earcut@5ae627e, mapbox/earcut@9961b54, mapbox/earcut@c461d09, mapbox/earcut@30b1732, mapbox/earcut@37c73ae?
  • No, these are all about flat array input, which is a JS-specific performance optimization that's not applicable to C++.
• Do we need to port mapbox/earcut@a169519 and mapbox/earcut@68da53f?
  • ⇢ Port Steiner point support #13

[mourner]

About mapbox/earcut#14 — since the C++ port used indexed output from the start, I assume it was correct from the beginning so no need for a port, but need to confirm with @kkaefer.

[jfirebaugh]

The fixes are now merged, however the test cases are not, because they fail. This needs investigation -- see the port-tests branch.

[mourner]

One more to port: mapbox/earcut@deb0bfe

[mourner]

Tried porting the last one, but it's a noticeable performance hit on C++. Is it OK to diverge JS and C++ implementations around this then?