Re: any value to "NORET_TYPE" macro?

["Satyam Sharma" <satyam.sharma@gmail.com>]

On 5/22/07, Satyam Sharma <satyam.sharma@gmail.com> wrote:
> On 5/22/07, Robert P. J. Day <rpjday@mindspring.com> wrote:
> > On Tue, 22 May 2007, John Anthony Kazos Jr. wrote:
> >
> > > >   given that:
> > > >
> > > > $ grep -r "define.*NORET_TYPE" *
> > > > include/linux/ext4_fs.h:# define NORET_TYPE    /**/
> > > > include/linux/linkage.h:#define NORET_TYPE    /**/
> > > > include/linux/ext3_fs.h:# define NORET_TYPE    /**/
> > > > $
> > > >
> > > > is there any obvious value to the 30 or so uses of that macro
> > > > sprinkled throughout the tree?
> > >
> > > Since it evaluates to absolutely empty code during pre-processing,
> > > there is no obvious value. The question is whether there is some odd
> > > hackish non-obvious value, I'd expect. (I'd also expect that to be
> > > another "no".)
> > >
> > > If something that evaluates to nothingness ("There was nothing
> > > left...not even a hole!") actually does anything, then somebody in
> > > the standards-compilers-users pipeline needs to be violently beaten
> > > for stupidity.
> >
> >
> > actually, one of the folks on the KJ list found this:
> >
> >   http://www.ussg.iu.edu/hypermail/linux/kernel/9605/1957.html
> >
> > which speaks thusly:
> >
> > ...
> > -#if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 5)
> > -# define NORET_TYPE __volatile__
> > -# define ATTRIB_NORET /**/
> > -# define NORET_AND /**/
> > -#else
> > # define NORET_TYPE /**/
> > # define ATTRIB_NORET __attribute__((noreturn))
> > # define NORET_AND noreturn,
> > -#endif
> > ...
> >
> >   so it looks like a thoroughly obsolete macro which can be tossed.
> > i'll make the patch and test it.
>
> AFAICT, NORET_TYPE must've been introduced to silence gcc _warnings_,
> and not do actually do anything useful that affects functionality in any way.
> So the way to "test" your patch would be to see if there is any increase /
> decrease in the number of *warnings* blurted out by gcc during kernel build
> (best would be to build with various gcc versions on various platforms :-)

But of course, John is *bang right* that if /**/ makes gcc do (or not do)
_anything at all_ (be it crying out a warning), then someone surely
deserves to get his head bashed in ... :-)