Re: [Boost-users] [mpl] how to find out whether a function is implemented

5 Aug 2005

      Zeljko Vrba  writes:
...
David Abrahams wrote:
...
However, reliably taking the address of an overloaded function
involves knowing the exact types in its signature.  I'm not sure how
you can do that.
I have googled a bit and found the following example in the Microsoft
C++ reference: http://tinyurl.com/cljnn
int Func( int i, int j );
int Func( long l );
...
int (*pFunc) ( int, int ) = Func;
Do you see any problems with this?
Not at all.  As I said, you need to know the exact signature.  The
problem is detecting that the function (or operator<) whose existence
you detect has the exact signature you care about.  

Maybe it could be tested by passing a convertible-to-T POD struct
through the function:

   // UNTESTED

   template <class T>
   struct convertible_to { operator T() const; };

   namespace tester
   {
     typedef char (&yes) [1];
     typedef char (&no) [2];

     struct not_found {};

     no is_found(not_found);
     yes i_found(...);

     not_found operator<(...);

     template 
     struct has_less1
     {
         static bool const value
           = sizeof(
               is_found(
                  convertible_to<A1>() < convertible_to<A2>()
               )
             ) == yes;
     };

     template <class T> T make();

     template 
     no is(T&,U&);

     template <class T>
     yes is(T&,T&);

     template ::value>
     struct has_less
     {
         static bool const value
           = sizeof( is(make<R>(), make<A1>() < make<A2>()) ) == yes;
     };

     template 
     struct has_less
     {
         static bool const value = false;
     };
   }

The problem with this is that you still can't tell whether the
operator< you found takes its arguments by reference or by value, and
without that information, you can't do the cast you want reliably.
...
I guess it should be possible to take an address of an operator by
replacing 'Func' with e.g. 'operator<'. I'll try it and report the
result.
I don't think this problem can be solved.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com