r183722 - Implement DR85: Redeclaration of member is forbidden

[Richard Smith]

On Wed, Jul 3, 2013 at 12:59 PM, Kim Gräsman <kim.grasman at gmail.com> wrote:
> David, all,
>
> I have a test for a clang-based tool that ran clean before this
> commit, and I now get an unexpected diagnostic due to an intentional
> double declaration in the test;
>
> class Foo {
>   class NoUsage;  // Unnecessary -- defined inline later in the class.
>   class NoUsageDefinedOutOfLine;  // Necessary -- part of the public API.
>   class NoUsageDefinedOutOfLine;  // Unnecessary -- second fwd declare.
>   ...
> };
>
> Our tool removes duplicate declarations, so we're interested in being
> able to process code like this. Now I'm greeted with:
>   tests\fwd_decl_nested_class.cc:19:9: warning: class member cannot be
> redeclared [-Wgnu]
>     class NoUsageDefinedOutOfLine;  // Unnecessary -- second fwd declare.
>
> And this seems to be in the spirit of the fix. But:
>
> 1) Is -Wgnu really the right warning group for this? I'm not up to
> speed on what -Wgnu means, is it to warn that this is a GNU extension?

Yes, that's what -Wgnu means.

> 2) Should this warning be active for -std=c++98 or only -std=c++11? I
> get the diagnostic in both modes.

It should be active in both modes. This code has never been valid.

> 3) Other compilers (GCC, MSVC) seem to accept it, so I think I want to
> keep our current behavior until this is considered illegal by
> everyone. Is my best bet just to disable the warning to let Clang Sema
> accept this?

Sure. If you want to disable just this one warning but keep the rest
of -Wgnu enabled, please provide a patch adding a separate warning
flag for this one warning (as a member of the -Wgnu group).

> On Tue, Jun 11, 2013 at 5:51 AM, David Majnemer
> <david.majnemer at gmail.com> wrote:
>> Author: majnemer
>> Date: Mon Jun 10 22:51:23 2013
>> New Revision: 183722
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=183722&view=rev
>> Log:
>> Implement DR85: Redeclaration of member is forbidden
>>
>> Disallow the existence of a declaration of a member class that isn't a
>> forward declaration before it's definition.
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits