Re: GCC rejecting (I think) valid code

[Dale Johannesen <dalej at apple dot com>]

On Jan 23, 2005, at 6:05 AM, Andreas Schwab wrote:

> "Sam Lauber" <sam124@operamail.com> writes:
>
> > I compile this pair of declarations with GCC (which would
> > probably be found in a stdint.h):
> >
> > typedef short int16_t;
> > typedef unsigned int16_t uint16_t;
> >
> > The first declaration is OK, but GCC gives me an error on
> > the second one. That dosen't make sense, because part of
> > the C89 grammar is
> >
> > type-name:
> > (some types) typedef-name
> >
> > typedef-declaration: something like
> > `typedef' type-name typedef-name
>
> There is more to C than the syntax.  The standard also contains 
> semantic
> constraints that are to be applied on top of the syntax.  I can't check
> C89, but in C99 section 6.7.2 contains an exhaustive list of allowed
> type-specifier sets in a type-name, and this list only allows a single
> typedef-name not mixed with other type-specifiers for each type-name.

C90 has the same list in 3.5.2.  An example in 3.5.6 sheds some light 
on why:
consider

struct {
   unsigned int16_t:4;
};