... And the patch.

2010/11/15 Fabien Chêne <fabien.chene@gmail.com>:
> Hello,
>
> 2010/8/20 Jason Merrill <jason@redhat.com>:
>> On 08/18/2010 03:05 PM, Fabien Chêne wrote:
>>>
>>> +diagnose_conflicting_using_decls (tree decl, tree real_decl, tree
>>> using_decl)
>>
>> I don't think we need to have special code to check for clashes between
>> using decls and other decls; such collisions ought to be caught by the same
>> code that catches two data members with the same name.  It looks like this
>> code in finish_member_declaration is preventing that:
>>
>>>  /* Enter the DECL into the scope of the class.  */
>>>  else if ((TREE_CODE (decl) == USING_DECL && !DECL_DEPENDENT_P (decl))
>>>           || pushdecl_class_level (decl))
>>
>> What happens if you just remove the USING_DECL check here?
>
> Sounds like it was  the right thing to do. I have had to modify a bit
> the parser in order to handle USING_DECLs.
> I also have modified supplement_binding() to handle USING_DECLs. It is
> a bit touchy, but it seems to work well.
> Please also double check that lookup/using31.C is correct, I am not sure.
>
>> Perhaps the checking code in handle_using_decl is also unnecessary.
>
> Definitely.
>
> Regtested x86_64-unknown-linux-gnu without regressions on an old tree.
> Regtesting in progress on a fresh tree.
> OK to commit if it succeeds ?
>
> gcc/ChangeLog
>
> 2010-11-15  Fabien Chêne  <fabien@gcc.gnu.org>
>
>        PR c++/26256
>        * dbxout.c (dbxout_type_fields): Ignore using declarations.
>
>
> gcc/testsuite/ChangeLog
>
> 2010-11-15  Fabien Chêne  <fabien@gcc.gnu.org>
>
>        PR c++/26256
>        * g++.dg/lookup/using23.C: New.
>        * g++.dg/lookup/using24.C: New.
>        * g++.dg/lookup/using25.C: New.
>        * g++.dg/lookup/using26.C: New.
>        * g++.dg/lookup/using27.C: New.
>        * g++.dg/lookup/using28.C: New.
>        * g++.dg/lookup/using29.C: New.
>        * g++.dg/lookup/using30.C: New.
>        * g++.dg/lookup/using31.C: New.
>        * g++.dg/lookup/using32.C: New.
>        * g++.dg/lookup/using33.C: New.
>        * g++.dg/lookup/using34.C: New.
>        * g++.dg/debug/using4.C: New.
>        * g++.dg/debug/using5.C: New.
>        * g++.old-deja/g++.other/using1.C: Adjust.
>        * g++.dg/template/using2.C: Likewise.
>
> gcc/cp/ChangeLog
>
> 2010-11-15  Fabien Chêne  <fabien@gcc.gnu.org>
>
>        PR c++/26256
>        * search.c (lookup_field_1): Get rid of the comment saying that
>        USING_DECL should not be returned, and actually return USING_DECL
>        if appropriate.
>        * semantics.c (finish_member_declaration): Remove the check that
>        prevents USING_DECLs from being verified by pushdecl_class_level.
>        * typeck.c (build_class_member_access_expr): Handle USING_DECLs.
>        * class.c (check_field_decls): Keep using declarations.
>        * parser.c (cp_parser_nonclass_name): Handle USING_DECLs.
>        * name-lookup.c (supplement_binding): Extend the `struct stat'
>        hack to handle using declarations.
>        (push_class_level_binding): Returns early for dependent USING_DECLs
>        and for USING_DECLs that refers to FUNCTION_DECL.
>
> --
> Fabien
>



-- 
Fabien