Re: (C++) Casting of a local pointer-to-member variable

[Jason Merrill <jason at redhat dot com>]

>>>>> "Nathan" == Nathan Sidwell <nathan@codesourcery.com> writes:

> Chandra Chavva wrote:
>> * semantics.c (finish_parenthesized_expr): Call resolve_offset_ref()
>> to give appropriate error message if the qualified_id of a
>> pointer-to-member is enclosed in parenthesis.

> But it's not necessarily an error - we can only determine so when we've some
> context to provide type information.

No, but I think calling resolve_offset_ref is the right thing to do here.
AFAIK, the only purpose of (this use of) OFFSET_REF is to support
generating pointers to members; once that's no longer a possibility, we
don't need the OFFSET_REF anymore.  Getting rid of it here means we can
lose PTRMEM_OK_P as well.  Yay, simplification.

Actually, my preference would be to do away with (this use of) OFFSET_REF
entirely, and have a special parser rule for & nested-name-specifier
identifier.  Life would be simpler yet without resolve_offset_ref.

(Note that OFFSET_REF is also used for .* and ->*, which really ought to
have their own code.)

> I think you'll need to tweek the error message when we determine that we
> would've selected a member function, but can't because of the
> parentheses.

It actually ends up hitting

wa3.C: In function `int main()':
wa3.C:7: assuming pointer to member `void A::f()'
wa3.C:7: (a pointer to member can only be formed with `&A::f')

> Do you have a test case?

struct A {
  void f ();
};

int main ()
{
  void (A::*p)() = &(A::f);
}

Jason