GCC rejects valid code?

[Joseph S. Myers]

On Thu, 24 Jan 2008, Roberto Bagnara wrote:

> While looking at the rules governing struct/union declarations in C, I
> stumbled
> upon this example:
> 
> union A {
>   int i;
>   float f;
> };
> 
> void
> foo(struct A** p) {
>   *p = 0;
> }
> 
> This is accepted by both Comeau and the Intel C compiler, but is rejected
> by GCC 4.1.2 and 4.3.0 on the grounds that
> 
> bug.c:7: error: ‘A’ defined as wrong kind of tag
> 
> My interpretation is that line 7 does not define `union A' with the
> wrong kind of tag;  it declares a (totally unrelated) `struct A'.
> However, I am not sure.  Should I file a bug report for this?

Tags have a single namespace, not three namespaces.  Because a declaration 
of the tag A is visible when "struct A" is used, and "struct A** p" does 
not define the contents of the type and is not just "struct A;", it refers 
to the previous type of that tag (6.7.2.3#9, paragraph numbering from 
N1256).  Referring to a union type with "struct" in turn violates 
paragraph 2 (added in C99 TC3 following DR#251).

-- 
Joseph S. Myers
joseph@codesourcery.com