Re: Invalid code in <limits>

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Richard Henderson <rth@redhat.com> writes:

| On Wed, Jan 29, 2003 at 12:06:18AM -0800, Mark Mitchell wrote:
| > The bottom line is that there's no need to make the compiler work
| > hard to deal with this stuff; it's easy to make a keyword whose value
| > is "__builtin_huge_valf () / 2 == __builtin_huge_valf ()" or
| > "__builtin_nanf ("") != __builtin_nanf ("")"; you just evaluate that
| > expression in the guts of the compiler where it can do whatever it
| > wants.
| 
| Ok, but how about
| 
| 	struct S {
| 	  static const double f = 1.0;
| 	  static const double g = __builtin_huge_valf ();
| 	};
| 
| (or whatever context is hardest for c++)?  If either
| works, they should both work.

Technically, the above are invalid constructs, *but* they are (or used
to be) supported by various compilers in different modes.  There is no
compelling reason why only integral-expression should be allowed.

Rewrite the above as

   struct A {
     static const double f;
     static const double g;
   };

   const double A::f = 1.0;
   const double A::g = __builtin_huge_valf();

extend with

   struct B {
      static const char  i = static_cast<int>(43);
   };

static_cast<> is no different from __builtin_huge_valf().

-- Gaby