clang accepts friend that g++/Comeau reject

[llvmbot]

Bugzilla Link	11989
Version	3.0
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@tkremenek

Extended Description

This code:

 class D {
     class E{
         class F{};
         friend  void foo1(D::E::F& q);
         };
     friend  void foo1(D::E::F& q);
     };

 void foo1(D::E::F& q) {}

 int main(int argc, char** argv) { return 0; }

is accepted by clang but rejected by g++ and Comeau. Gnu says they are right to reject (see http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52212), so I'm filing it with you. One of you guys is right; perhaps you can get together and agree who it is?

[tkremenek]

cloned to rdar://problem/10869563

[llvmbot]

@llvm/issue-subscribers-clang-frontend

[widberg]

This is still an issue. gcc 13.2 rejects the code while clang 17.0.1 and trunk (60c4f82) accept it. https://godbolt.org/z/3oje3KbeW

[Endilll]

EDG, GCC, and MSVC agree that F is not accessible in the second friend declaration: https://godbolt.org/z/9xTb6WP7T
But I wonder if their conclusion still holds against post-P1787 wording. That said, this paper doesn't seem to have changed much for qualified name lookup in friend declarations, but this is not an observation made with confidence.

[zyn0217]

EDG, GCC, and MSVC agree that F is not accessible in the second friend declaration: https://godbolt.org/z/9xTb6WP7T But I wonder if their conclusion still holds against post-P1787 wording. That said, this paper doesn't seem to have changed much for qualified name lookup in friend declarations, but this is not an observation made with confidence.

@Endilll P1787 changed the wording that originally justified the issue

-A name nominated by a friend declaration shall be accessible in the scope of the class containing the friend declaration.
+A member nominated by a friend declaration shall be accessible in the scope of the class containing the friend declaration.

- Access control is applied uniformly to all names, whether the names are referred to from declarations or expressions.

I presume the previous interpretation (that the name F used by outer friend declaration isn't made accessible by the inner friend declaration) no longer holds.

Shall we revisit this example?

[zyn0217]

FWIW we still have discrepencies after the PR with GCC:

https://godbolt.org/z/a4zdWadqh

https://godbolt.org/z/bGEdja56E