[BZ,#16907] Do not disable __attribute__ when you include argp.h

  Hi,

As written in bug report:

"
When <argp.h> is included first and thus __attribute__ is not defined,
with -std=c90, c99 or c11, the __attribute__ is redefined as empty due
to the following code:
This might break headers included later, for example it breaks
-D_FORTIFY_SOURCE=2 by not defining strcpy() correctly [1]. Besides some
reformating, this code is more than 10 years old, and I don't know if
there is still a reason to redefine __attribute__ that way.
"

So is there any reason to keep fragments of code below? As we use
__attribute__ in stdio.h without checking gcc version I doubt that
compiling with gcc would work. I did not checked that though.



	* argp/argp-fmtstream.h: Do not define __attribute__.
	* argp/argp.h: Likewise.
  

On 18 May 2014 13:57, Ondřej Bílka <neleai@seznam.cz> wrote:
> Hi,
>
> As written in bug report:
>
> "
> When <argp.h> is included first and thus __attribute__ is not defined,
> with -std=c90, c99 or c11, the __attribute__ is redefined as empty due
> to the following code:
> This might break headers included later, for example it breaks
> -D_FORTIFY_SOURCE=2 by not defining strcpy() correctly [1]. Besides some
> reformating, this code is more than 10 years old, and I don't know if
> there is still a reason to redefine __attribute__ that way.
> "
>
> So is there any reason to keep fragments of code below? As we use
> __attribute__ in stdio.h without checking gcc version I doubt that
> compiling with gcc would work. I did not checked that though.

I believe this file is shared with gnulib so changes should ideally be
synchronized between the two projects.

>
>
>
>         * argp/argp-fmtstream.h: Do not define __attribute__.
>         * argp/argp.h: Likewise.
>
>
> diff --git a/argp/argp-fmtstream.h b/argp/argp-fmtstream.h
> index 1ed2834..f3f82df 100644
> --- a/argp/argp-fmtstream.h
> +++ b/argp/argp-fmtstream.h
> @@ -29,21 +29,6 @@
>  #include <string.h>
>  #include <unistd.h>
>
> -#ifndef __attribute__
> -/* This feature is available in gcc versions 2.5 and later.  */
> -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 5) || \
> -  defined __STRICT_ANSI__
> -#  define __attribute__(Spec) /* empty */
> -# endif
> -/* The __-protected variants of `format' and `printf' attributes
> -   are accepted by gcc versions 2.6.4 (effectively 2.7) and later.  */
> -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 7) || \
> -  defined __STRICT_ANSI__
> -#  define __format__ format
> -#  define __printf__ printf
> -# endif
> -#endif
> -
>  #if defined (__GNU_LIBRARY__) && defined (HAVE_LINEWRAP_H)
>  /* line_wrap_stream is available, so use that.  */
>  #define ARGP_FMTSTREAM_USE_LINEWRAP
> diff --git a/argp/argp.h b/argp/argp.h
> index 0868228..c1072a3 100644
> --- a/argp/argp.h
> +++ b/argp/argp.h
> @@ -35,21 +35,6 @@
>  # define __NTH(fct) fct __THROW
>  #endif
>
> -#ifndef __attribute__
> -/* This feature is available in gcc versions 2.5 and later.  */
> -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 5) || \
> -  defined __STRICT_ANSI__
> -#  define __attribute__(Spec) /* empty */
> -# endif
> -/* The __-protected variants of `format' and `printf' attributes
> -   are accepted by gcc versions 2.6.4 (effectively 2.7) and later.  */
> -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 7) || \
> -  defined __STRICT_ANSI__
> -#  define __format__ format
> -#  define __printf__ printf
> -# endif
> -#endif
> -
>  /* GCC 2.95 and later have "__restrict"; C99 compilers have
>     "restrict", and "configure" may have defined "restrict".  */
>  #ifndef __restrict
  

On Mon, May 19, 2014 at 08:37:27AM +0100, Will Newton wrote:
> On 18 May 2014 13:57, Ondřej Bílka <neleai@seznam.cz> wrote:
> > Hi,
> >
> > As written in bug report:
> >
> > "
> > When <argp.h> is included first and thus __attribute__ is not defined,
> > with -std=c90, c99 or c11, the __attribute__ is redefined as empty due
> > to the following code:
> > This might break headers included later, for example it breaks
> > -D_FORTIFY_SOURCE=2 by not defining strcpy() correctly [1]. Besides some
> > reformating, this code is more than 10 years old, and I don't know if
> > there is still a reason to redefine __attribute__ that way.
> > "
> >
> > So is there any reason to keep fragments of code below? As we use
> > __attribute__ in stdio.h without checking gcc version I doubt that
> > compiling with gcc would work. I did not checked that though.
> 
> I believe this file is shared with gnulib so changes should ideally be
> synchronized between the two projects.
>

CCing gnulib. Could this be backported also there or do you have
different solution? 


> >
> >
> >
> >         * argp/argp-fmtstream.h: Do not define __attribute__.
> >         * argp/argp.h: Likewise.
> >
> >
> > diff --git a/argp/argp-fmtstream.h b/argp/argp-fmtstream.h
> > index 1ed2834..f3f82df 100644
> > --- a/argp/argp-fmtstream.h
> > +++ b/argp/argp-fmtstream.h
> > @@ -29,21 +29,6 @@
> >  #include <string.h>
> >  #include <unistd.h>
> >
> > -#ifndef __attribute__
> > -/* This feature is available in gcc versions 2.5 and later.  */
> > -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 5) || \
> > -  defined __STRICT_ANSI__
> > -#  define __attribute__(Spec) /* empty */
> > -# endif
> > -/* The __-protected variants of `format' and `printf' attributes
> > -   are accepted by gcc versions 2.6.4 (effectively 2.7) and later.  */
> > -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 7) || \
> > -  defined __STRICT_ANSI__
> > -#  define __format__ format
> > -#  define __printf__ printf
> > -# endif
> > -#endif
> > -
> >  #if defined (__GNU_LIBRARY__) && defined (HAVE_LINEWRAP_H)
> >  /* line_wrap_stream is available, so use that.  */
> >  #define ARGP_FMTSTREAM_USE_LINEWRAP
> > diff --git a/argp/argp.h b/argp/argp.h
> > index 0868228..c1072a3 100644
> > --- a/argp/argp.h
> > +++ b/argp/argp.h
> > @@ -35,21 +35,6 @@
> >  # define __NTH(fct) fct __THROW
> >  #endif
> >
> > -#ifndef __attribute__
> > -/* This feature is available in gcc versions 2.5 and later.  */
> > -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 5) || \
> > -  defined __STRICT_ANSI__
> > -#  define __attribute__(Spec) /* empty */
> > -# endif
> > -/* The __-protected variants of `format' and `printf' attributes
> > -   are accepted by gcc versions 2.6.4 (effectively 2.7) and later.  */
> > -# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 7) || \
> > -  defined __STRICT_ANSI__
> > -#  define __format__ format
> > -#  define __printf__ printf
> > -# endif
> > -#endif
> > -
> >  /* GCC 2.95 and later have "__restrict"; C99 compilers have
> >     "restrict", and "configure" may have defined "restrict".  */
> >  #ifndef __restrict
>
  

Ondřej Bílka wrote:
> CCing gnulib. Could this be backported also there or do you have
> different solution?

We solved this problem in a different way years ago, using something 
like this:

#if __GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR >= 7)
# define _GL_ATTRIBUTE_FORMAT(s) __attribute__ ((__format__ s))
#else
# define _GL_ATTRIBUTE_FORMAT(s)
#endif

and then using _GL_ATTRIBUTE_FORMAT ((__printf__, 2, 3)) later, instead 
of __attribute__ ((__format__ (__printf__, 2, 3))).  Presumably glibc 
could do something similar.  Glibc could use a different name; that's 
all right, gnulib will just switch to the name that glibc uses.