Re: bug in check for stack growth direction in _AC_LIBOBJ_ALLOCA

[Bruno Haible]

Eric Blake wrote:
> Hmm, we'll need to backport this improved stack-direction testing to
> gnulib and libsigsegv, then

This is true for gnulib's alloca.c and c-stack.c.

The code in libsigsegv already contains a counter-measure against
tail recursion elimination:

------------------------------------ foo.c ------------------------------------
#include <stdio.h>
int
get_stack_direction ()
{
  auto char dummy;
  static char *dummyaddr = (char *)0;
  if (dummyaddr != (char *)0)
    return &dummy > dummyaddr ? 1 : &dummy < dummyaddr ? -1 : 0;
  else
    {
      dummyaddr = &dummy;
      {
        int result = get_stack_direction ();
        /* The next assignment avoids tail recursion elimination
           (IRIX 6.4 CC).  */
        dummyaddr = (char *)0;
        return result;
      }
    }
}
int
main ()
{
  printf ("%d\n", get_stack_direction ());
  return 0;
}
-------------------------------------------------------------------------------

But it doesn't prevent it safely! With gcc-4.6.0 for x86 and -O3 optimization,
I get the wrong result:

$ /arch/x86-linux/gnu-inst-gcc/4.6.0/bin/gcc -O2 foo.c
$ ./a.out 
-1
$ /arch/x86-linux/gnu-inst-gcc/4.6.0/bin/gcc -O3 foo.c
$ ./a.out 
1

Bruno
-- 
In memoriam Neda Agha-Soltan <http://en.wikipedia.org/wiki/Neda_Agha-Soltan>