CGAL users discussion list

[Marc Glisse <>]

On Thu, 26 Jul 2012, Marc Mörig wrote:

> Dear CGAL maintainers,
>
> in CORE_Expr.h there is twice the call
>
> x.approx(53,1024);
>
> meant to improve the internal approximation in x to double quality.
> After a call x.approx(r,a) the internal approximation z is such that
>
> |x - z| <= max{ |x|2^{-r}, 2^{-a}  }
>
> So that call wastes precision if x is in the underflow range. It should be
>
> x.approx(53,1075);
>
> since the smallest (positive nonzero) double is 2^{-1074}.

Thank you for the detailed report. It makes sense, I have made the 
suggested change for 4.2.

--
Marc Glisse