Question

#include <iostream>

struct A
{
    bool f( int a ) { std::cout << "int" << std::endl; return true;};
    bool f( int a, int b ) { std::cout << "two int" << std::endl; return true;};
    bool f( float a ) {std::cout << "float" << std::endl; return true;};
    bool f( float a, float b ) {std::cout << "two float" << std::endl; return true;};
};

template <typename T>
struct Type
{
    typedef bool (A::*One)(T);
    typedef bool (A::*Two)(T, T);
};

template <typename T, typename Type<T>::One F >
void run(A & a)
{
    T tmp = 0;
    (a.*F)(tmp);
}


int main(int argc, char ** argv )
{
    A a;
    run<int, &A::f>(a);
    run<float, &A::f>(a);

    return 0;
}

Problem lies with syntax of typename Type<T>::One F. If I specify the actual method pointer instead of using my typedef from Type it works fine (see here and here).
Is there a way to use those typedefs from Type as template arguments?

Answer 1

My current guess is that both IBM xlC and Sun CC have bugs.

The versions which give errors are probably confused by the use of typename in a template argument context but used to indicates that a dependant name is a type instead of introducing a type argument. Note that these compilers aren’t conforming and don’t demand typename to indicates that a dependant name is a type even in other contexts where it is needed (even for Sun CC versions where your code compile).