[systemd-devel] [PATCH 2/5] libsystemd:terminal :fix uninitialized warning

[Zbigniew Jędrzejewski-Szmek]

On Mon, Nov 03, 2014 at 01:41:20PM +0100, Lennart Poettering wrote:
> On Mon, 03.11.14 13:12, David Herrmann (dh.herrmann at gmail.com) wrote:
> 
> > Hi
> > 
> > On Thu, Oct 16, 2014 at 11:43 PM,  <philippedeswert at gmail.com> wrote:
> > > From: Philippe De Swert <philippedeswert at gmail.com>
> > >
> > > Remove the following warning during the compilation:
> > > src/libsystemd-terminal/grdev-drm.c: In function 'grdrm_card_hotplug':
> > > src/libsystemd-terminal/grdev-drm.c:1087:45: warning: 'fb' may be used uninitialized in this function [-Wmaybe-uninitialized]
> > > src/libsystemd-terminal/grdev-drm.c:1035:19: note: 'fb' was declared here
> > > ---
> > >  src/libsystemd-terminal/grdev-drm.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/src/libsystemd-terminal/grdev-drm.c b/src/libsystemd-terminal/grdev-drm.c
> > > index dba6db2..415755e 100644
> > > --- a/src/libsystemd-terminal/grdev-drm.c
> > > +++ b/src/libsystemd-terminal/grdev-drm.c
> > > @@ -1032,7 +1032,7 @@ error:
> > >
> > >  static void grdrm_crtc_expose(grdrm_crtc *crtc) {
> > >          grdrm_pipe *pipe;
> > > -        grdrm_fb *fb;
> > > +        grdrm_fb *fb  = NULL;
> > 
> > Ewww, this is not nice. It does fix the warning, indeed, but the
> > underlying problem is more generic. Lets look at this:
> > 
> > int some_function(void **out) {
> > ...
> >         r = ioctl(...);
> >         if (r < 0)
> >                 return -errno;
> > ...
> > }
> > 
> > gcc has no guarantee that "errno" is <0 if r is <0. Therefore,
> > whenever it inlines those functions (-O2 etc.), it will spew a warning
> > that "out" might be uninitialized if r<0 but errno==0. With -O0 gcc
> > doesn't complain as it probably does not optimize across functions.
> > However, with -O2 I get those warnings for "static" functions all the
> > time.
> > 
> > Not sure what to do here. I dislike initializing the pointer to NULL
> > as it might hide other real warnings. I'd prefer something like:
> > 
> > r = -SANE_ERRNO;
> > 
> > with:
> > 
> > #define SANE_ERRNO (abs(errno) ? : EMAGIC)
> > 
> > ...not sure what we do in other places, though. Lennart? Tom?
> 
> So far we went the simple way out and merged patches like the original
> one you are replying to.
> 
> Adding a call like this would make sane to me though:
> 
> static inline negative_errno(void) {
>        return _likely_(errno > 0) ? -errno : -EINVAL;
> }
> 
> Which is then invoked as:
> 
>       return negative_errno();
> 
> or so...
I like the assert more, because having errno <= 0 would be a signficant bug
in the system, not something that we want to ever hide. So basically this
is a way to tell the compiler what we already know, and assert is better.

I also filed https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61846, but
without much response so far.

Zbyszek