ambiguous operator[] in variadic template

问题:

I'm trying to compile this example, where a variadic class template inherits from a variadic amount of bases, each of which implements a different operator[]:

#include <iostream>

template <typename T>
struct Field {
  typename T::value_type storage;

  typename T::value_type &operator[](const T &c) {
    return storage;
  }
};

template<typename... Fields>
struct ctmap : public Field<Fields>... {
};

int main() {
    struct age { typedef int value_type; };
    struct last_name { typedef std::string value_type; };

    ctmap<last_name, age> person;

    person[last_name()] = "Smith";
    person[age()] = 104;
    std::cout << "Hello World!" << std::endl;
    return 0;
}

When I compile with gcc (Debian 4.9.2-10), I get the following error

main.cpp: In function ‘int main()’:
main.cpp:22:23: error: request for member ‘operator[]’ is ambiguous
     person[last_name()] = "Smith";
                       ^
main.cpp:7:27: note: candidates are: typename T::value_type& Field<T>::operator[](const T&) [with T = main()::age; typename T::value_type = int]
   typename T::value_type &operator[](const T &c) {
                           ^
main.cpp:7:27: note:                 typename T::value_type& Field<T>::operator[](const T&) [with T = main()::last_name; typename T::value_type = std::basic_string<char>]
main.cpp:23:17: error: request for member ‘operator[]’ is ambiguous
     person[age()] = 104;
                 ^
main.cpp:7:27: note: candidates are: typename T::value_type& Field<T>::operator[](const T&) [with T = main()::age; typename T::value_type = int]
   typename T::value_type &operator[](const T &c) {
                           ^
main.cpp:7:27: note:                 typename T::value_type& Field<T>::operator[](const T&) [with T = main()::last_name; typename T::value_type = std::basic_string<char>]

Why is this ambiguous?

回答1:

A portable way do do what you want is roughly:

template<class...Ts>
struct operator_index_inherit {};
template<class T0, class T1, class...Ts>
struct operator_index_inherit<T0, T1, Ts...>:
  T0, operator_index_inherit<T1, Ts...>
{
  using T0::operator[];
  using operator_index_inherit<T1, Ts...>::operator[];
};
template<class T0>
struct operator_index_inherit<T0>:
  T0
{
  using T0::operator[];
};

then:

template<class... Fields>
struct ctmap : operator_index_inherit<Field<Fields>...> {
  using base = operator_index_inherit<Field<Fields>...>;
  using base::operator[];
};

here we linearly inherit from each of the types, and using operator[] on our parents.

If we could using Field<Fields>::operator[]...; we would not have to do this.

Some care has to be taken with constructors (which I did not take), but you might not need to do this.

live example.

---

What is actually going wrong depends on details of the standard I am less than certain of. Basically, you are mixing operators and inheritance and overloading in a complex way. Even if your code is standard compliant (which it may or may not be), it is compliant in a way that some compilers die on.

回答2:

The code is invalid and gcc is correct to reject it (clang 3.6.0 accepts it though - this is a bug). The rules for looking up an operator start with, from [over.match.oper]:

[...] for a binary operator @ with a left operand of a type whose cv-unqualified version is T1 and a right operand of a type whose cv-unqualified version is T2, three sets of candidate functions, designated member candidates, non-member candidates and built-in candidates, are constructed as follows:
— If T1 is a complete class type or a class currently being defined, the set of member candidates is the result of the qualified lookup of T1::operator@ (13.3.1.1.1); otherwise, the set of member candidates is empty.

The lookup rules for a member name are (since we're looking up ctmap<last_name,age>::operator[]), from [class.member.lookup]:

The lookup set for f in C, called S(f,C), [...] is calculated as follows:

If C contains a declaration of the name f, [...]

Otherwise (i.e., C does not contain a declaration of f or the resulting declaration set is empty), S(f,C) is initially empty. If C has base classes, calculate the lookup set for f in each direct base class subobject Bi, and merge each such lookup set S(f,B_i) in turn into S(f,C).

The following steps define the result of merging lookup set S(f,B_i) into the intermediate S(f,C):
— [...]
— Otherwise, if the declaration sets of S(f,B_i) and S(f,C) differ, the merge is ambiguous: the new S(f,C) is a lookup set with an invalid declaration set and the union of the subobject sets. In subsequent merges, an invalid declaration set is considered different from any other.
— [...]

Basically - we have two base classes here, both with an operator[]. Both declaration sets differ - so the merge is ambiguous. The way to disambiguate would be to introduce using-declarations bringing in all the base class member functions into the derived class, so that the initial lookup set finds everything.

To shorten your example:

struct A { void foo(char) { } };
struct B { void foo(int ) { } };

struct C : A, B { };

struct D : A, B {
    using A::foo;
    using B::foo;
};

With that hierarchy

C c;
c.foo(4);  // error: ambiguous lookup set for foo()

D d;
d.foo('x') // OK: calls A::foo()