Degenerate triangles on boundary

[bsy-nicholasw]

I've been chasing an unexpected behavior and boiled it down to a very basic input. I'm adding points to an existing triangle (on boundary, inside, etc) and using Delaunator to retriangulate and Constrainautor to constrain certain edges. I've been getting degenerate triangles on the boundary.

Consider this input:

[[0, 0], [30.51591076416662, -5.272057753973058], [13.955446995823877, 16.74653639495955], [22.3355613751919, -3.8587860099360114]]

The first three points are the original triangle, and the last point is one I add, which was inserted on the edge [P0, P1]. Delaunator treats the 4th point as inside, instead of on the convex hull.

I get the following triangles:

[2, 3, 0], [3, 1, 0], [2, 1, 3]

And the hull:

[2, 1, 0]

A visual of the data, where the blue line is the degenerate triangle (note, I'm in a +Y up coordinate system, so everything's CW) -

[mourner]

Sorry for a late response — I've looked into this issue and it looks like the root cause is that P3 here is very close to the P0 to P1 line but not strictly collinear (probably due to floating point errors), and the deviation, while small, is not insignificant (something like 2e-11). So the question is where do we draw the line between a degenerate triangle and a very small but otherwise numerically significant one. Should there be a threshold parameter of sorts for Delaunator for cases like this? We probably can't filter things like this out by default.