CGAL users discussion list

[Andreas Fabri <>]

On 01/03/2010 22:12, Eric Berberich wrote:
> Panagiotis Foteinos wrote:
> > Sadly, this is not an option for my code due to timing issues.
>
> Sadly robust geometric computing is not feasible with floating-point
> arithmetic. At least if you have (near-)degenerate situations. If you
> can ensure generic position floating-point might suffice. In fact you
> can even analyse which values of volumes should suffice ...
>
> eriC


The answer is philosophical too.  A more pragmatic view is
to use an Exact_predicates_inexact_constructions_kernel
for the Delaunay triangulation class, and to compute volumes
with an exact/exact kernel.  And then there is ongoing
work on filtered constructions where you try with floating
point  interval arithmetic, check if the interval fits
your needs, and if not or a floating point exception has
been thrown do the construction exactly.

andreas