Re: [patch c++]:PR/15774 - Conflicting function decls not diagnosed (this time really for 15774)

[Kai Tietz <ktietz70@googlemail.com>]

2010/12/25 Jason Merrill <jason@redhat.com>:
> On 12/24/2010 05:24 PM, Kai Tietz wrote:
>>
>> Ok, I'll apply it tomorrow. No the error message doesn't display the
>> attribute as it displays a decl.  This would be possibly another task
>> to display attributes (at least a wished limited set) also for
>> declarations.
>
> I think we want to print whatever attributes cause the attributed entity to
> be considered distinct from the non-attributed version, rather than a
> redeclaration.
>
> Thinking about it that way, I think your patch actually isn't the right way
> to handle this: in pr15774-1.C, the two foos are the same foo, just the
> change in attributes is ill-formed.  So decls_match should return 1, but we
> should complain about attribute mismatch in duplicate_decls, perhaps just
> before the call to merge_decl_attributes.
>
> Jason
>

Hmm, not sure. Case 2 is simply ill-formed, but doesn't cause wrong
code. But case 1 in fact can produce wrong code. See as expample

extern void foo (int);
void doo(int a) { foo(a); }
void __attribute__((stdcall)) foo (int i) { }

As here in function "doo" the "foo" gets called with cdecl
calling-convention, but its implementation gets "stdcall" one, which
are incompatible and therefore lead to wrong-code.

The -2 case doesn't have this issue

Kai
-- 
|  (\_/) This is Bunny. Copy and paste
| (='.'=) Bunny into your signature to help
| (")_(") him gain world domination