CGAL users discussion list

["Sebastien Loriot (GeometryFactory)" <>]

Ben Haller wrote:
> On Jul 21, 2010, at 10:23 AM, Sebastien Loriot (GeometryFactory) wrote:
>
> > Ben Haller wrote:
> > >  Yeah, I think you're right.  I'm puzzled as to why using exact numbers would 
> > > solve the problem; the points in question violate the assumption (that 3+ points 
> > > will not be colinear) even with exact numbers, I would think.  But anyhow, I 
> > > have come to the conclusion that jittering the points slightly is a reasonable 
> > > workaround; I often have colinear points, and I don't really care which way the 
> > > Delaunay triangulation decides to handle them, as long as it picks a way and 
> > > doesn't raise.  :->
> > >  Thanks!
> > The problem here is not which triangulation you will get but
> > will you get a triangulation. Due to floating point computation,
> > you end up in a situation (because another predicate was badly
> > evaluated) where you call a predicate with points that
> > should form an independent set of points. Unfortunately this is not the
> > case and the predicate crashes.
>
>   I see.  Whereas with exact numbers, the first predicate is evaluated 
> exactly, and that leads to the avoidance of the second predicate...?  Well, 
> jittering seems to be working for me so far as a workaround, but if this ever 
> happens to me again, I'll switch to the exact number type.  Thanks!
>
> Ben Haller
> McGill University
If no construction are required, an alternative may be
using Exact_predicates_inexact_constructions_kernel.

S.

PS: Please answer to the list, the discussion may be useful to other people.