Re: gcc bug impacts glibc cvs

[Daniel Jacobowitz <dmj+ at andrew dot cmu dot edu>]

On Mon, Mar 12, 2001 at 11:06:59AM -0800, Ulrich Drepper wrote:
> Franz Sirl <Franz.Sirl-kernel@lauterbach.com> writes:
> 
> > Well, currently this test is quite simple, every GCC does it
> > potentially wrong so far. Maybe a fix can go into gcc-3.0, but most
> > probably it won't be in gcc-2.95.3.
> 
> On x86 the problem seems not to appear:
> 
> extern void *xyzzy __attribute__ ((__weak__));
> int
> main (void)
> {
>   return &xyzzy ? (xyzzy == (void *) 0) : 0;
> }
> 
> This succeeds with -O0, -O1, -O2, -O3.  Therefore I do not want a
> work-around to go into glibc unconditionally.
> 
> If the above test is not enough let's refine it.

I don't know how to easily reproduce the problem on x86, just due to
the vagaries of the scheduler; it appears to be cheap enough to do
short jumps that ?: doesn't get collapsed down into eager evaluation. 
Here's a testcase that does actually show the problem on PPC, though:

extern void *xyzzy __attribute__ ((__weak__));
int
main (void)
{
  return &xyzzy ? ((int)xyzzy) : 0;
}


(Your original test does not exhibit the problem, for reasons beyond my
immediate sight - again, scheduler quirks).

While it's certainly a bit too fragile of a test, could we at least use
this to autoconf for the bug until someone thinks of a better one?

-- 
Daniel Jacobowitz                           Debian GNU/Linux Developer
Monta Vista Software                              Debian Security Team
                         "I am croutons!"