Re: [Qemu-devel] [PATCH] kvm: workaround build break on gcc-7.1.1 / fedora26

[Greg Kurz]

On Thu, 3 Aug 2017 11:07:13 -0300
Philippe Mathieu-Daudé <address@hidden> wrote:

> On 08/03/2017 10:55 AM, Eric Blake wrote:
> > On 08/03/2017 08:34 AM, Philippe Mathieu-Daudé wrote:  
> >> Hi Greg,
> >>
> >> On 08/02/2017 11:47 AM, Greg Kurz wrote:  
> >>> Building QEMU on fedora26 with the latest gcc package fails:
> >>>
> >>>     CC      ppc64-softmmu/target/ppc/kvm.o
> >>> In file included from include/sysemu/hw_accel.h:16:0,
> >>>                    from target/ppc/kvm.c:31:
> >>> target/ppc/kvm.c: In function ‘kvmppc_booke_watchdog_enable’:
> >>> include/sysemu/kvm.h:449:35: error: ‘args_tmp[i]’ may be used
> >>> uninitialized  
> >   
> >>
> >> This sizeof() use looks unnatural to me. I wonder why not use size_t,
> >> since this is about sizeof()/ARRAY_SIZE(). The problem seems to come
> >> from the commit this cast was introduced (61c7bbd236):
> >>
> >> target-ppc/kvm.c:1302:21: error: comparison of unsigned expression < 0
> >> is always false [-Werror=type-limits]
> >>
> >> So I'd rather suggest this code, which looks more natural to read to me:
> >>
> >>      if (ARRAY_SIZE(args_tmp)) {
> >>          for (i = 0; i < ARRAY_SIZE(args_tmp) && ...  
> 
> Oops I forgot to suggest size_t i.
> 
> > 
> > For that matter, the existing code is doing:
> > 
> > int i;
> > i < (int)ARRAY_SIZE(args_tmp)
> > 
> > but wouldn't that be better as:
> > 
> > size_t i;
> > i < ARRAY_SIZE(args_temp)
> > 
> > I guess we have both the old compilers (per commit 61c7bbd2) and the new
> > to worry about; although I was unable to reproduce it on Fedora 26 on
> > x86_64 (is this an architecture-dependent compiler bug?)
> >   
> 

The following code snippet spits a warning with gcc-7.1.1-3.fc26 and -Wall on
ppc64le AND x86_64:

int foo(void)
{
    char empty_array[] = { };
    int i, ret = 0;

    for (i = 0; i < (int) (sizeof(empty_array) / sizeof(empty_array[0])); i++) {
        ret = empty_array[i];
    }

    return ret;
}

If I drop the (int), the warning goes away... and so does the build break
of qemu-system-ppc64 on my ppc64le host.

I don't have 4.7.1 or 4.6 compilers around but I could check 4.8.5
is okay with that change FWIW.

> Ok so let's stop losing time about compiler incoherent warnings, using 
> -Wno-type-limits for GCC < 5...
> So we can keep a sane/understandable codebase, using size_t and no (int) 
> cast.
> 

If I also convert 'int i' to 'size_t i' then I get the same error as
in commit 61c7bbd2:

error: comparison of unsigned expression < 0 is always
 false [-Werror=type-limits]

> Phil.

pgpbLE5zmZjpm.pgp
Description: OpenPGP digital signature