Incorrect Voronoi Diagram for simple test case

[kgoodrick-uu]

For a certain array of 5 points the Voronoi diagram "misses" a point. I have reproduced this on observable.

I originally reported this as a bug in Vega-Lite vega/vega-lite#9388 and then Vega vega/vega#3941, but after more digging I believe the bug is coming from this library.

[Fil]

Thanks! I can confirm that this bug comes from the jittering we use to solve collinear points:

d3-delaunay/src/delaunay.js

Line 63 in dc894a1

const p = jitter(points[2 * i], points[2 * i + 1], r);

Solved by increasing the factor 1e-8 to 1e-7. However, this in turn breaks another test. I found a sweet spot value at 1.5e-8 that makes all the tests happy, but it's clearly not a great solution (see #147).

[Fil]

solved by #147