Re: [C++] float, double & complex template non-type parms

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Fergus" == Fergus Henderson <fjh@cs.mu.oz.au> writes:

    Fergus> On 26-Feb-2001, Nathan Sidwell <nathan@codesourcery.com>
    Fergus> wrote:
    >> Hi, should we take the opportunity to deprecate non-type
    >> template parms of float, double or complex type in 3.0?

    Fergus> Why should they be deprecated?

The arguments in favor:

- They're non-standard.

- They don't work when cross-compiling, in general, due to weakness
  in the name-mangler and elsewhere.

- Their semantics are unclear.  For example:
  
    template <double d>
    class C {};

    template <>
    class C<1.0> {};

    C<1.1 - 0.1> c;

  Do we get the specialization or not?  The situation is much less
  clear than with integer arithmetic.

The arguments against:

- This is probably not an extension that causes conforming programs to
  behave differently.  Any time you use a template parameter that has
  `float' or `double' type, you are outside the standard so all 
  bets are off.

  Actually, I'm not sure what the overloading rules say here, and
  whether or not something like:

    template <class T, T t> void f(T t);

  is rejected as an overload candidate if `t' has type `float'.  If it
  *is* supposed to be rejected, then this extension might change
  the meaning of a conforming program, meaning that is not an
  extension, but a modification.

- Some other compilers do support them.

I, personally, support deprecating them.  Jason, what do you think?

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com