Re: [PATCH] Use extern inline wrappers instead of macros for some checking functions

[Jakub Jelinek <jakub at redhat dot com>]

On Wed, Mar 09, 2005 at 08:06:57PM +0100, Jakub Jelinek wrote:
> This patch:
> a) fixes getwd, pread and pread64 that were broken before
> b) starts using extern inline wrappers instead of macros for those checking
>    routines that have no GCC builtins.  Although programs using say
>    somestruct->read (buf, 1, 2, 3);
>    without #undef read or using (somestruct->read) (buf, 1, 2, 3);
>    are broken, there are simply too many of them.
>    The wrappers for routines that have checking builtins in GCC
>    (like memcpy, strcpy or e.g. printf) should stay as is, so that warnings
>    about them are reported on the place where the bug occurs instead of
>    always in say bits/string3.h.
> c) improves test coverage of these functions
> Tested with GCC 4.0-RH with:
> -#if _FORTIFY_SOURCE > 0 && __GNUC_PREREQ (4, 1) && __OPTIMIZE__ > 0
> +#if _FORTIFY_SOURCE > 0 && (__GNUC_PREREQ (4, 1) || (defined __GNUC_RH_RELEASE__ && __GNUC_PREREQ (4, 0))) && __OPTIMIZE__ > 0
> so that tst-chk{2,3} actually test something.

Roland asked off-line if I could use __typeof for the __*_alias prototypes.
Unfortunately, it seems I can't.
With
extern ssize_t __read_chk (int __fd, void *__buf, size_t __nbytes,
                           size_t __buflen) __wur;
extern __typeof (read) __read_alias __asm__ ("read") __wur;

extern __always_inline __wur ssize_t
read (int __fd, void *__buf, size_t __nbytes)
{
  if (__bos0 (__buf) != (size_t) -1
      && (!__builtin_constant_p (__nbytes) || __nbytes > __bos0 (__buf)))
    return __read_chk (__fd, __buf, __nbytes, __bos0 (__buf));
  return __read_alias (__fd, __buf, __nbytes);
}

All GCCs I have tried (3.2.3-RH, 3.3.4-RH, 3.4.3-RH, 4.0 branch) treat
__typeof (read) as a "call" and thus this triggers a warning in
all those compilers:
/usr/include/unistd.h:312: warning: 'read' declared inline after being called
/usr/include/unistd.h:312: warning: previous declaration of 'read' was here

So I'll just post an __always_inline patch on top of this one.

	Jakub