Question

Suppose you have the following code

template<typename T,void (T::*m)(int)>
    struct B{

    void f(T* a,int x){
        (a->*m)(x);
    }
};


struct A{


   template<typename X> void f(int){
  }

   void wrap(int i){
     f<char>(i);
   }

   B<A,&A::f<char> > y;
};


int main(){
   A a;
}

This definition

B<A,&A::f<char> > y;

works with gcc,but not with Visual studio 2010:

error C2440: 'specialization' : cannot convert from 'overloaded-function' to 'void (__thiscall A::* )(int)'

On the contrary,

 B<A,&f<char> > y;

works with visual studio but not with gcc.

Note that B<A,&A::f<char> > y; placed in the main, i.e.

  int main(){
       B<A,&A::f<char> > y;
    }

works for VS as well.

IS B<A,&f<char> > y; not standard? is there a way (apart wrapping the template function) to make the thing compile with both compilers?

===EDIT====

A possible, dirty solution is

#ifdef _WIN32
#define vsFix(a,b) b
#else
#define vsFix(a,b) a::b
#endif

 B<A,&vsFix(A,f)<char> > y;

Answer 1

IS B<A,&f<char> > y; not standard?

No. &f<char> is a pointer to function, not pointer to member function.

is there a way (apart wrapping the template function) to make the thing compile with both compilers?

Yes, by fixing the types, and passing pointer to object of type A :

template<typename T,void (T::*m)(int)>
struct B{

    void f(T* a,int x){
        (a->*m)(x);
    }
};


struct A{

   template<typename X> void f(int){
  }

   void wrap(int i){
     y.f(this, i);
   }

   B< A, &A::f<int> > y;
};


int main(){
   A a;
   a.wrap(5);
}