Patch for Friend function linkage check (PR 15841)

[John McCall]

On Apr 24, 2013, at 7:14 PM, Rafael Espíndola <rafael.espindola at gmail.com> wrote:

> On 24 April 2013 21:25, Weiming Zhao <weimingz at codeaurora.org> wrote:
>> Hi Rafael,
>> 
>> 
>> 
>> Could you please check if the attached patch fixes the issue
>> (http://llvm.org/bugs/show_bug.cgi?id=15841) correctly?
>> 
> 
> I don't think this is the correct fix. getLanguageLinkageTemplate
> should follow the structure of [dcl.link]. I think the real problem is
> the context the friend function is getting.
> 
> I went looking in the standard, and what I found was [namespace.memdef] p3:
> 
> -------------------------------------------
> Every name first declared in a namespace is a member of that
> namespace. If a friend declaration in a non-local class first declares
> a class, function, class template or function template the friend is a
> member of the innermost enclosing namespace.
> ------------------------------------------
> 
> If I read that correctly,
> 
> extern "C" {
>  class Foo {
>    friend void bar();
>  };
>  void bar() {
>  }
> }
> 
> is equivalent to
> 
> void bar();
> extern "C" {
>  class Foo {
>    friend void bar();
>  };
>  void bar() {
>  }
> }
> 
> So I now think that clang is correct and the code should be rejected.
> Richard, do you agree? Is this a defect?

Linkage specifications don't actually have members under the standard.
Instead, they just set the language linkage of any non-class-member name
declared therein that has external linkage.  That would seem to apply to
a friend declaration.

John.