(Serious) sparc/solaris EH problem

[Teemu Torma]

I think the following problem should be corrected before the release.
If compiled with -O, i is not zero as one would expect.  The problem
does not happen if using sjlj exceptions, or if not optimizing.
I suspect that the register gets lost somehow.

$ c++ -O eh.cc
$ ./a.out
1: i = 0
2: i = 1
3: i = 1

Teemu

--------------------------------------------------------------------
#include <stdio.h>

int bar ()
{
  throw 100;
}

main ()
{
  int i = 0;
  try
    {
      printf ("1: i = %d\n", i);
      i = bar ();
    }
  catch (...)
    {
      printf ("2: i = %d\n", i);
    }

  printf ("3: i = %d\n", i);
  return i;
}
---------------------------------------------------------------------------