[PATCH] PR17337: Retained language linkage

[Alp Toker]

On 22/10/2013 20:26, Richard Smith wrote:
> +def ext_retained_language_linkage : Extension<
> +  "friend function %0 retaining previous language linkage is an
> extension">,
> +  InGroup<DiagGroup<"retained-language-linkage">>;
>
> I'd like this diagnostic to be clearer that
>
>   extern "C" void f();
>   struct S {
>     friend void f();
>   };
>
> is fine but
>
>   extern "C" void f();
>   extern "C++" {
>     struct S {
>       friend void f();
>     };
>   }
>
> is the extension. Maybe add a note pointing at the innermost
> surrounding linkage specification that we're ignoring?

I don't think that's the right way to look at this change.

My interpretation of the spec is very much that friend function
declarations don't have a linkage spec in the first place because they
don't introduce a function name:

[dcl.link]p4:
  In a linkage-specification,
  the specified language linkage applies to the function  types  of  all
  function declarators, function names, and variable names **introduced** by
  the declaration(s).


So the fact we're "ignoring" the innermost linkage spec is at best an
implementation detail that I'd prefer not to leak any further into a
diagnostic.

I know my view differs from yours in that I don't see this is a
correctness deviation or relaxation of the standard :-)

On the other hand, highlighting the selected outer linkage spec _would_
be nice to do when diagnosing linkage specs in general.

I have some patches back from when the linkage spec work started in 2010
to improve diags. Thinking to get back to it, but either way, digging up
and passing around the LinkageSpecDecl belongs in a separate patch.

>
> +      // The friend object kind isn't yet complete so check IDNS
> directly.
> +      if (New->getIdentifierNamespace() & Decl::IDNS_OrdinaryFriend) {
>
> We don't care whether it's FOK_Declared or FOK_Undeclared, so why not:
>
>   if (New->getFriendObjectKind() != FOK_None) {
>
> ?

Sure, let's do that.

So, I'm looking to land this patch with the getFriendObjectKind() change
and then contining with general improvement of linkage spec diagnostics.

Let me know if you feel strongly enough about the diagnostic to object
to landing this in the next day or so.

Alp.



>
>
> On Tue, Oct 22, 2013 at 10:46 AM, Alp Toker <alp at nuanti.com
> <mailto:alp at nuanti.com>> wrote:
>
>     Hello Richard,
>
>     With this patch, friend function declarations will retain the language
>     linkage specified for previous declarations instead of emitting an
>     error
>     diagnostic.
>
>     The feature is known to be compatible with GCC and MSVC and permits a
>     language to be specified indirectly where it cannot otherwise be
>     written
>     directly in class scope.
>
>     Further to the previous patch, this feature is now a clang
>     extension so
>     warnings can be enabled/disabled with a
>     -Wretained-language-linkage flag
>     while we seek clarifications to the language standard. Tests have been
>     moved to SemaCXX/linkage-spec.cpp.
>
>     Alp.
>
>     --
>     http://www.nuanti.com
>     the browser experts
>
>

-- 
http://www.nuanti.com
the browser experts

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20131022/2d3beda8/attachment.html>