Re: 1008 segfaults in genattr

[mrs at wrs dot com (Mike Stump)]

> To: Robert Lipe <robertl@dgii.com>
> Cc: egcs@cygnus.com
> From: Jason Merrill <jason@cygnus.com>
> Date: 20 Nov 1997 22:03:51 -0800

> >>>>> Robert Lipe <robertl@dgii.com> writes:

> > 4) Make the test case call _exit() instead of exit(), but just 
> > 	let code in the field learn this the hard way.

> I think I'm going to go with #4.

I'd prefer either a fixed atexit, or if we can stick in code to the
testcase to detect infinite looping, if that is the problem, and call
abort (or _exit in that case).

The reason is that the compiler uses atexit internally by itself to
implement C++ semantics, and if it doesn't work well enough, the
semantics it presents to the user will be flawed (in my opinion).
Now, the ANSI C++ may state the users program is flawed, and it is,
they should not be calling exit from global dtors, but under the label
of quality of implementation, we have a chance to do what I think is
the right thing, and it is reasonable to have a testcase for it.

Is it sufficient to change the testcase like the below?  If it is,
then that is the solution I prefer, along with a // execution test
fails - XFAIL *-*-sco*, or whatever is needed to mark it.

// prms-id: 9732

int count;
int bail = 0;

struct base {
  base () { ++count; }
  ~base () { --count; }
  base(const base&o) { ++count; }
};

class D {
public:
  ~D() {
    if (bail++)
      {
	// On some Linux boxes, we run the dtor for d twice,
	// once before exit, and once after!
	abort ();
      }
    else
      {
	if (count != 0)
	  abort ();
	exit (0);
      }
  }
} d;

base base_object;

base base_returning_function ();

const base& base_ref = base_returning_function ();

int main () {
}

base base_returning_function () {
  base local_base_object;
  return local_base_object;
}