CGAL users discussion list

[Sebastien Loriot <>]

Martin Mörth wrote:
> Hello Everybody!
>
> I am calculating distances between points and segments. I isolated a
> case where the computation seems to be returning a bad value (double
> that is not a number).
>
> ] typedef CGAL::Simple_homogeneous< double > CgKernel;
> ] 
> ] typedef CgKernel::FT        CgFieldNumber;
> ] typedef CgKernel::Point_2   CgPoint;
> ] typedef CgKernel::Segment_2 CgSegment;
> ] 
> ] std::stringstream text
> ] (
> ] "-2.0527819641170391e+142 1.5399153216416277e+142 1.2075326642268196e+139
> " // p
> ] "-1.3832287034884996e+069 9.6657090977844965e+068 7.5794164096585707e+065
> " // s.min
> ] "2509.7485796014098 1275.2576940874201 1.0" // s.max
> ] );
> ] 
> ] CgPoint p;
> ] CgSegment s;
> ] 
> ] text >> p;
> ] text >> s;
> ] 
> ] CgFieldNumber field_number_dist = CGAL::squared_distance( p, s );
> ] double dist = CGAL::to_double( field_number_dist );
> ] 
> ] assert( _fpclass( dist ) == _FPCLASS_QNAN /* Quiet NaN */ );
>
> Approximate cartesian coordinates for the objects involved are:
>
> ] p     -1699.9804849431803 1275.2576946873996
> ] s.min -1824.9804849431800 1275.2576947042162
> ] s.max  2509.7485796014098 1275.2576940874201
>
> The point is very close to the segment. I am not sure whether
> producing a NaN is the behaviour one would have to expect given
> the kernel I use. 
Hello Martin,

I do not see your point. You are using a floating point number type
inside your kernel, so the expected behaviour is that you can get NaN.

You probably should have a look at http://www.cgal.org/philosophy.html


Sebastien.
> Maybe someone with more insight into the
> distance calculation implementation can shed some light on this.
>
> I am using CGAL 3.4 and VS 8.0 on Windows XP.
>
> Thanks in advance for any help!
>
> Martin
>