[Patch, libgfortran] PR 41219 Silence uninitialized variable warning

[Richard Guenther]

On Mon, Sep 7, 2009 at 12:13 PM, Martin Jambor<mjambor@suse.cz> wrote:
> On Sun, Sep 06, 2009 at 02:59:18PM +0200, Richard Guenther wrote:
>> On Sun, Sep 6, 2009 at 2:56 PM, Janne
>> Blomqvist<blomqvist.janne@gmail.com> wrote:
>> > Hi,
>> >
>> > the patch below fixes the "use of uninitialized variable" warning in
>> > unix.c. It's a spurious warning as the fstat() call will initialize
>> > the stat struct, but gcc can't see it.
>>
>> Huh, that looks odd.  I bet this is SRA at work.
>>
>
> I haven't examined this more than looking at the hunk below... but
> isn't statbuf sort of too much TREE_ADDRESSABLE for SRA?  I reckon
> that fstat cannot be optimized away...

I was thinking that SRA might insert a (useless) copy like

 SR1 = stat.A;
 SR2 = stat.B;
 stat.A = SR1;
 stat.B = SR2;
  fstat(&stat);

which then looks like uninitialized loads.  But I didn't investigate either.

Richard.

> Martin
>
>> Richard.
>>
>> > Ok for trunk?
>> >
>> > 2009-09-06   Janne Blomqvist   <jb@gcc.gnu.org>
>> >
>> >            PR libfortran/39667
>> >            * io/unix.c (fd_to_stream): Silence uninitialized warning.
>> >
>> >
>> > diff --git a/libgfortran/io/unix.c b/libgfortran/io/unix.c
>> > index 4ecba3a..e26f281 100644
>> > --- a/libgfortran/io/unix.c
>> > +++ b/libgfortran/io/unix.c
>> > @@ -759,6 +759,10 @@ fd_to_stream (int fd, int prot)
>> >    s->logical_offset = 0;
>> >    s->prot = prot;
>> >
>> > +   /* Silence compiler warnings about using unitialized variables.   */
>> > +   statbuf.st_mode = 0;
>> > +   statbuf.st_size = 0;
>> > +
>> >    /* Get the current length of the file. */
>> >
>> >    fstat (fd, &statbuf);
>> >
>> >
>> > --
>> > Janne Blomqvist
>> >
>