[cwg1699] clang interprets friendship rules lexically; other compilers do not

[llvmbot]

Bugzilla Link	44708
Version	9.0
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@dwblaikie,@DougGregor,@zygoloid

Extended Description

The following code:

class A {
  friend class B;
  int m;
};

class B {
  friend void f(A* p) { p->m = 32; }
};

compiles in clang version 9.0.0-2. If the function body is defined outside of the class, the compiler correctly issues an error about not being able to access the private member.

[zygoloid]

Well, per [class.access.base]p5, "A member m is accessible at the point R when named in class N if [...] m as a member of N is private, and R occurs in a [...] friend of class N"

The reference to p->m occurs in class B, which is a friend of class A, so arguably this is valid.

This question is the subject of CWG issue 1699 (http://wg21.link/cwg1699), and the consensus opinion when that was last discussed seemed to be that Clang's current behavior is correct.

[shafik]

It looks like this behavior has changed but dr 1699 looks like is still active and appears to be on EWGs plate

CC @erichkeane

[shafik]

Also see discussion here: https://lists.isocpp.org/core/2021/06/11087.php and https://lists.isocpp.org/core/2023/05/14339.php

[llvmbot]

@llvm/issue-subscribers-clang-frontend