Fix atomic builtins detection

[Jeremy Allison]

On Tue, Apr 26, 2016 at 12:46:59PM +0200, Jérémie Courrèges-Anglas wrote:
> Ralph Boehme <slow at samba.org> writes:
> 
> > On Mon, Apr 25, 2016 at 11:15:42AM -0700, Jeremy Allison wrote:
> >> On Mon, Apr 25, 2016 at 04:24:28PM +0200, Jeremie Courreges-Anglas wrote:
> >> > 
> >> > Hi,
> >> > 
> >> > lib/replace/wscript has this bunch of code:
> >> > 
> >> >     # Check for atomic builtins. */
> >> >     conf.CHECK_CODE('''
> >> >                     int main(void) {
> >> >                         int i;
> >> >                         (void)__sync_fetch_and_add(&i, 1);
> >> >                         return 0;
> >> >                     }
> >> >                     ''',
> >> >                     'HAVE___SYNC_FETCH_AND_ADD',
> >> >                     msg='Checking for __sync_fetch_and_add compiler builtin')
> >> > 
> >> > 
> >> > It is incorrect, since CHECK_CODE already wraps the code fragment in
> >> > main().  Here's what gets compiled:
> >> > 
> >> > 
> >> > [includes]
> >> >  int main(void) { 
> >> >                     int main(void) {
> >> >                         int i;
> >> >                         (void)__sync_fetch_and_add(&i, 1);
> >> >                         return 0;
> >> >                     }
> >> >                     ; return 0; }
> >> > 
> >> > 
> >> > Now the funny part is when you run this where __sync_fetch_and_add is
> >> > actually not available.  The inner main() function isn't called
> >> > anywhere, and at cc -O2 it is simply optimized out, thus the compile
> >> > test wrongly succeeds.  This can be seen for example on OpenBSD/hppa
> >> > (gcc-4.2.1).
> >> > 
> >> > Please find attached a patch to fix this issue for __sync_fetch_and_add
> >> > and atomic_add_32.
> >> 
> >> Looks good to me. Can I get a second Team reviewer ?
> >
> > lgtm. Can you push? Thanks!
> 
> Thanks you both.  Any chance to see this land in v4-3?

Yep - create a bug at bugzilla.samba.org and ask for
a back-port - will be a simple process.