clang incorrectly diagnoses template operator* as invalid

[Quuxplusone]

Bugzilla Link	PR13869
Status	NEW
Importance	P enhancement
Reported by	Eli Friedman (efriedma@quicinc.com)
Reported on	2012-09-18 15:53:22 -0700
Last modified on	2018-01-23 09:38:15 -0800
Version	trunk
Hardware	PC All
CC	dgregor@apple.com, llvm-bugs@lists.llvm.org, rafael@espindo.la, richard-llvm@metafoo.co.uk, ville.voutilainen@gmail.com
Fixed by commit(s)
Attachments
Blocks
Blocked by
See also

Testcase (from g++.old-deja/g++.pt/overload8.C):

struct baz;
void operator*(baz&, double);
template <class T> inline T operator*(double s, const T &p);
void m(baz& a) { a * .5; }

gcc accepts this; clang gives an error "overloaded 'operator*' must have at
least one parameter of class or enumeration type".  Doug says clang should
accept this.

[Quuxplusone]

Same thing:

template <typename T, typename U>
void operator<<(T t, U u)
{
}

template <typename T>
void
operator<<(T t, int i)
{
}

struct S
{
};

int main()
{
    char buf[8];
    S s;
    buf << s;
    return 0;
}

gcc accepts, clang rejects, all the way up to the current trunk.

[over.oper]/6 says
"An operator function shall either be a non-static member function or be a non-
member function that has
at least one parameter whose type is a class, a reference to a class, an
enumeration, or a reference to an
enumeration."

I would certainly expect the violation of that "shall" to be a SFINAE
case, not a hard error.

[Quuxplusone]

This is not covered by SFINAE. Per [temp.deduct]p8,

"Only invalid types and expressions in the immediate context of the function type and its template parameter types can result in a deduction failure."

The issue occurs when we create the declaration of the function template specialization, which happens after we've already determined the function type (which is valid and doesn't result in any kind of failure).

While I agree with Doug that clang should accept this, I don't think this code is valid per the standard as written; taking this to CWG.