V3 PATCH: numeric_limits<> support, fix PR/3865

[Gabriel Dos Reis]

Roger Sayle <roger@eyesopen.com> writes:

| On Tue, 3 Sep 2002, Richard Henderson wrote:
| > On Tue, Sep 03, 2002 at 11:57:46AM -0600, Roger Sayle wrote:
| > > bool has_snan = !(__builtin_snan("") == 0.0);
| >
| > If -fsignaling-nan, this ought to result in sigfpe.
| > Thus you can't do that.
| 
| Is there any agreement on evaluating constant expressions containing
| signaling NaNs at compile time.  I agree that if the above statement
| where executed at run-time I'd expect a SIGFPE, but as an initializer
| to a global constant?

If the above expression is not evaluated at compile-time, it isn't
suitable for initializing numeric_limits<>::has_signaling_NaN since
C++ imposes the following constraint: 
 
  18.2.1/3
  For all members declared static const in the numeric_limits
  template, specializations shall define these values in such a way
  that they are usable as integral constant expressions. 

-- Gaby