Re: [Bug-gnubg] bad __attribute__ handling breaks w/newer standards, newer gcc, & glibc and causes infinite loop

[Philippe Michel]

On Wed, 7 Oct 2015, Michael Petch wrote:

> On 2015-10-06 14:00, Russ Allbery wrote:
> > /*
> >  * __attribute__ is available in gcc 2.5 and later, but only with gcc 2.7
> >  * could you use the __format__ form of the attributes, which is what we use
> >  * (to avoid confusion with other macros).
> >  */
> > #ifndef __attribute__
> > # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 7)
> > #  define __attribute__(spec)   /* empty */
> > # endif
> > #endif
>
> Hi Guys,
>
> Thanks for the information. Russ, I was considering doing this the way
> Bison seems to deal with __attribute__ . The generated code looks like:
>
> #ifndef YY_ATTRIBUTE
> # if (defined __GNUC__                                               \
>      && (2 < __GNUC__ || (__GNUC__ == 2 && 96 <= __GNUC_MINOR__)))  \
>     || defined __SUNPRO_C && 0x5110 <= __SUNPRO_C
> #  define YY_ATTRIBUTE(Spec) __attribute__(Spec)
> # else
> #  define YY_ATTRIBUTE(Spec) /* empty */
> # endif
> #endif
>
> Based on this I was considering something like:
>
> #ifndef __attribute__
> # if (defined __GNUC__ \
> && (2 < __GNUC__ || (__GNUC__ == 2 && 96 <= __GNUC_MINOR__))) \
> || defined __SUNPRO_C && 0x5110 <= __SUNPRO_C
> # define __attribute__(Spec) /* empty */
> # endif
> #endif
>
>
> It seems to also handle Sun compilers as well. Anyone have any
> objections to the latter solution?

There is a ax_c___attribute__.m4 macro in the autoconf archive that 
handles this totally differently, by checking if __attribute__(unused) 
compiles.

This may be a little careless since gnubg uses it for other things like 
alignment. Maybe the macro could be changed so that it optionally takes a 
list of attributes to check.

I have no idea if an approach is better. It probably doesn't matter much.