Shane MacLaughlin

Update, it only appears to happen where there are a mix of negative and positive coordinates. Shifting prior to triangulation seems to correct the issue. Further testing ongoing.

Hi Andrew, the attached dataset reduces the couple of million points in the original sample down to 1163 points. Still trying to reduce this further but removing many more points...

Not much smaller but at 1090 points another test case that might be interesting as it suggests an issue with the convex hull [delaunator-bug3.txt](https://github.com/user-attachments/files/24014624/delaunator-bug3.txt) Zooming in a bit shows another...

A bit more testing here shows this issue still occurs on datasets with no negative coordinates, but much less frequently. Will continue with some more tests

Looking at the convex hull of the failed triangulation, we can see that it is not actually convex. This could be a good mechanism for automatically detecting the point at...

Implemented the above which seems to correct the problem on all test sets so far. Changes are as follows; **delaunator.hpp** ``` namespace delaunator { const double PointResolution = 0.00001; const...

Modified source attached. I'm working on Visual C++ 2022 (Version 17.13.7), targetting 64 bit Windows 10/11. Building using ISO C++17 Standard with precise floating point model. Testing carried out with...

Thanks for the feedback. I'll be sticking with the modified code myself for now as it seems to produce stable results for the type of data that I'm dealing with....