Re: Uninitialized error with GCC mainline

[Markus Trippelsdorf <markus at trippelsdorf dot de>]

On 2017.03.03 at 16:13 +0100, Markus Trippelsdorf wrote:
> On 2017.03.03 at 01:55 +0000, Joseph Myers wrote:
> > As shown at 
> > <https://sourceware.org/ml/libc-testresults/2017-q1/msg00210.html>, 
> > building with GCC mainline fails with an uninitialized variable error on 
> > many (all?) architectures:
> > 
> > In file included from fnmatch.c:250:0:
> > fnmatch_loop.c: In function 'internal_fnwmatch':
> > ../locale/weightwc.h:103:28: error: '*((void *)&str+4)' may be used 
> > uninitialized in this function [-Werror=maybe-uninitialized]
> >     if (cp[nhere - 1] > usrc[nhere -1])
> >                         ~~~~^~~~~~~~~~
> > 
> > Building with GCC r245852 fails; building with GCC r245811, as in the 
> > previous build results 
> > <https://sourceware.org/ml/libc-testresults/2017-q1/msg00204.html>, passes 
> > with current glibc.  So this results from some GCC change, not a glibc 
> > change.  I haven't bisected to an individual revision.  We need to figure 
> > out if this is a compiler bug, or a glibc bug, or a case where in fact the 
> > uninitialized use cannot occur but the compiler reasonably cannot see it, 
> > in which case diagnostic pragmas should be used as elsewhere in this file.
> 
> There are many more similar warnings later on.
> This started with gcc's rr245840. It would be could if someone could
> analyze the issues and open gcc bugs if necessary.

The issue was fixed by gcc's r245976. I can now build glibc with gcc
trunk without any warning at -O3.

-- 
Markus