Missing mangling specification for parenthesized unresolved names

[hubert-reinterpretcast]

Consider the following case:

struct A {
  operator char *();
};

inline namespace N {
  char x(char *);
}
inline namespace M {
  char x(long long *);
}

template <typename T> void f(char (*)[sizeof (x)(T{})]) {}
template <typename T> void f(char (*)[sizeof x(T{})]) {}

short x(const A &);

void g(char (*p)[sizeof(char)]) { f<A>(p); }
void h(char (*p)[sizeof(short)]) { f<A>(p); }

What scope the name x belongs to is necessary unknown in the context of the calls from the template definitions.
As can be seen, both functions are callable with the same template arguments.
I do not see anything in the specification that provides a mangling for the parenthesized name.

Clang appears to handle this by using cp in place of cl for the call expression; however, this solution does not exactly handle the case where a unary & operator either forms a pointer or a pointer-to-member:

struct A { int x; };

char q(int *);
short q(int A::*);

template <typename T>
int f(char (*)[sizeof(q(&T::x))]) { return 1; }

template <typename T>
int f(char (*)[sizeof(q(&(T::x)))]) { return 2; }

int g(char (*p)[sizeof(char)] = 0) { return f<A>(p); }
int h(char (*p)[sizeof(short)] = 0) { return f<A>(p); }

[hubert-reinterpretcast]

Clang's cp versus cl is consistent with the specification's Dt versus DT treatment. Doing something similar with an alternative to the ad operator name is a reasonable resolution.

[zygoloid]

Clang's cp mangling was agreed on the old mailing list back in 2011. Looks like it never made it into the ABI document, though.

For &(dependent qualified-id), how about we add <operator-name> ::= ap for unary & with a dependent qualified-id as operand?

[rjmccall]

That seems reasonable to me. And yeah, we should add the old mangling.