Explanation on Member Detection code

Question

It's not the concept as a whole, but rather one of the methods it uses to determine if a class has an n data-member. Here is the full code; an ordinary use of SFINAE for member detection.

template <typename T>
struct has_X {
    struct Fallback { int X; };
    struct Derived : T, Fallback {};

    template <typename U, U> struct S;

    template <typename C> static char (&f(S<int Fallback::*, &C::X> *))[1];
    template <typename C> static char (&f(...))[2];

    public:
        const static bool value = sizeof(f<Derived>(0)) == 2;
};

The part where Derived inherits from both Fallback and T confuses me because when we do the overload of f, &C::X is &Derived::X. But shouldn't this overload always be chosen because isn't Derived guaranteed to have X since it inherits from Fallback which has that data-member?

Maybe I'm overlooking something. However, this single piece of code has shown and taught me things I never knew, so maybe there is something to this. What I would expect is for that overload to always be chosen (not the one with the ...) because Derived should always have X since it inherits from Fallback. But this is not the case. Can someone please explain why?

Answer 1

Fallback has one data member named X, but Derived will have two if T also has a member named X, in which case Derived::X cannot be taken unambiguously. So if T does not have X, the first overload is used, and if T has X, the second more general version is used. This is why you can tell these cases apart depending on the size of their return types.