Workaround needed for silent kill() failure in OS X

Imported from http://bugzilla.roxen.com/bugzilla/show_bug.cgi?id=4793

Reported by Jonas Wallden jonasw@roxen.com

Test 349 from modules/testsuite seems to hang intermittently on OS X. A distilled test case:

class Fnord {

  int gnapp(int t)
  {
    int e;

    werror("#%d: STARTING\n", t);
    for (e = 0; e < 7; e++) {
      for (int d = 0; d < 150; d++) {
	object o = Process.create_process( ({ "/bin/sleep", "99999" }) );
	int err = kill(o->pid(), 9);
	o->wait();
      }
    }
    werror("#%d: DONE\n", t);
    return -1;
  }
  
  array start()
  {
    array a = ({ });
    for (int e = 0; e < 10; e++)
      a += ({ thread_create(gnapp, e) });
    return a;
  }
};


void main()
{
  werror("starting...\n");
  array res = Fnord()->start()->wait() - ({ -1 });
  werror("res: %O\n", res);
}

With tracing enabled in signal_handler.c we've determined that the kill() call in f_kill() returns without error but doesn't terminate the targeted process. The problem has yet to be reduced to a standalone C program, but one hypothesis is that the forked /bin/sleep has not yet started completely before the SIGKILL signal is delivered and that the OS X kernel therefore disregards the signal.