CGAL users discussion list

[Olivier Devillers <>]

If you really want to compute the Delaunay triangulation of integers in [0,2000],
then you do not really need the CGAL management of robustness issues:
If you use 64 bits integers for your coordinates, computation should be exact
(in sphere test for integers on d bits require 5d+5 bits=60 bits in your case).
When 5 points are co-spherical, you avoid an expensive certification of this fact.
It remains a bit costly, because you still have many cases where more than 4 points are
cospherical, and some special work has to be done to deal with such situations.


Another solution is to keep your coordinates in double and actually perturb them
(add a random value in [0 ,  0.00001] to each coordinate)

--
Olivier Devillers, chercheur à