CGAL users discussion list

[Camille Wormser <>]

Bernhard Kornberger wrote:
> If it is started by the command "./mat" it inserts the points
> from the badPoints.txt file into the traingulation. However,
> when it comes to the computation of the dual voronoi vertices,
> it outputs one time "nan nan nan".

This is a printing issue: you are using an exact number type, and the 
conversion to double fails.
You can still recover the exact values by doing:

Point q = T.dual(c_it);
std::cout << "v="
         << q.x().exact().to_double() << " "
         << q.y().exact().to_double() << " "
         << q.z().exact().to_double() << " " << std::endl;

The root of the problem is that the to_double function of Lazy_exact_nt 
does not detect NaN or Infinity properly:

if (compare_relative_precision(app,
   Lazy_exact_nt<ET>::get_relative_precision_of_to_double()) <= 0)
               return CGAL_NTS to_double(app);


If you replace the "<=" by "<", everything works fine, because 
compare_relative_precision returns 0 in this case. I attach the 
corresponding diff.

By the way, I have a question: the comparison of two NaNs in CGAL 
returns CGAL::EQUAL. Is it the way it should be?
Index: Lazy_exact_nt.h
===================================================================
--- Lazy_exact_nt.h     (revision 39817)
+++ Lazy_exact_nt.h     (working copy)
@@ -1079,7 +1079,7 @@
 
             // If it's precise enough, then OK.
             if (compare_relative_precision(app,
-                 Lazy_exact_nt<ET>::get_relative_precision_of_to_double()) 
<= 0)
+                 Lazy_exact_nt<ET>::get_relative_precision_of_to_double()) < 
0)
                 return CGAL_NTS to_double(app);
 
             CGAL_PROFILER(std::string("failures of : ") + 
std::string(CGAL_PRETTY_FUNCTION));