Full name of submitter (unless configured in github; will be published with the issue): Hubert Tong

Reference (section label): expr.unary.op

Link to reflector thread (if any): N/A

Issue description:
There should be little controversy that the following should compile, but the wording is unclear: https://godbolt.org/z/aYo86aEer

struct A {
  union { int x; };
  int y;
};
struct B : A {};

template <typename> struct Q;
template <> struct Q<A> {};
template <typename T> Q<T> f(int T::*);

Q<A> g() { return f(&B::x); }
Q<A> h() { return f(&B::y); }

https://wg21.link/expr.unary.op under bullet 3.1 has:

If the operand is a qualified-id naming a non-static or variant member m of some class C, other than an explicit object member function, the result has type “pointer to member of class C of type T” and designates C​::​m.

In the above:

• x is a variant member both of the anonymous union and of A, and
• y is a non-static member of both A and of B.

It seems C in the wording is not quite unique.

Suggested resolution:

If the operand is a qualified-id naming a non-static or variant class member m of some class C, other than an explicit object member function, then let C be

• the class of which m is a direct member if m is not a variant member, and otherwise
• the class, that is not an anonymous union, of which m is a variant member.

The the result has type “pointer to member of class C of type T” and designates C​::​m. [Note: A qualified-id that names a member of a namespace-scope anonymous union is considered to be a class member access expression ([expr.prim.id.general]) and cannot be used to form a pointer to member.]