CGAL users discussion list

[Marc Mörig <>]

On 07/26/2012 09:24 AM, Sebastien Loriot (GeometryFactory) wrote:
> On 07/26/2012 09:09 AM, Marc Mörig wrote:
> > Dear CGAL maintainers,
> >
> > in the file Sqrt_extension_type.h, there is the following code:
> >
> > std::pair<double, double> compute_to_interval() const
> > {
> > if (! is_extended_)
> > return CGAL_NTS to_interval(a0_);
> > CGAL::Interval_nt<false>::Protector p;
> > const CGAL::Interval_nt<false>& a0_int = CGAL_NTS to_interval(a0_);
> > const CGAL::Interval_nt<false>& a1_int = CGAL_NTS to_interval(a1_);
> > const CGAL::Interval_nt<false>& root_int = CGAL_NTS to_interval(root_);
> > const CGAL::Interval_nt<false>& x_int =
> > a0_int + (a1_int * CGAL::sqrt(root_int));
> >
> > std::pair<double, double> res(x_int.inf(), x_int.sup());
> > this->update_cached_value(res);
> > return res;
> > }
> >
> >
> > The Protector p sets a directed rounding mode for floating point
> > arithmetic, which is propagated to other code inside, e.g.
> > to_interval(a0_). In my case, this breaks code which assumes the default
> > rounding mode.
> Are you sure the problems is coming from that part? The previous
> rounding mode mode is restored when the protector is deleted.
>
> Sebastien.

I am pretty sure. a0_ is of a custom number type I wrote myself. Hence 
the directed rounding mode is in effect when my code is called. The 
problem goes away if I move the Protector creation three lines 
downwards. This is ok, since no floating-point operations are performed 
in the interval assignment anyways.

Marc

--
Marc Mörig                                 Email:  

Institut für Simulation und Graphik        Phone: +49 391 67-52858
Otto-von-Guericke Universität Magdeburg    Office:         G29-227