[Bug c++/27115] New: [4.0/4.1/4.2 Regression] ICE in cp_expr_size or miscompilation

[jakub at gcc dot gnu dot org]

struct S
{
  S () : d(1.0) {}
  S (const double &x) : d(x) {}
  S (const S &x) : d(x.d) {}
  const S &operator= (const S &x) { d = x.d; return *this; }
  bool operator< (const S &x) const { return d < x.d; }
  double d;
};

S
foo (S a, S b)
{
  S c = ({ (a < b) ? a : b; });
  return c;
}

extern "C" void abort ();

int
main (void)
{
  S a = 6.0, b = 8.0, c = 10.0, r;
  r = foo (a, b);
  if (r.d != 6.0)
    abort ();
  r = foo (c, a);
  if (r.d != 6.0)
    abort ();
  r = foo (c, b);
  if (r.d != 8.0)
    abort ();
  return 0;
}

ICEs in cp_expr_size at -O0 and is miscompiled at -O2.
After gimplification foo looks like:
S foo(S, S) (a, b)
{
  struct S & c.0;
  struct S D.2401;
  bool D.2405;
  struct S c;

  (void) 0;
  c.0 = <retval>;
  D.2405 = operator< (a, b);
  *c.0 = D.2401;
  return <retval>;
}
so D.2401 is undefined and D.2405 is unused.  Before gimplification it is:
{
  struct S c;

  (void) 0;
  <<cleanup_point <<< Unknown tree: expr_stmt
  (void) (*<retval> = TARGET_EXPR <D.2401, <<cleanup_point <<< Unknown tree:
expr_stmt
  if (operator< ((struct S *) a, (struct S &) (struct S *) (struct S *) b))
    {
      (void) *a;
    }
  else
    {
      (void) *b;
    } >>>
>>>) >>>
>>;
  <<cleanup_point return <retval>>>;
}
which looks ok.
With GCC 3.4.x the test passed at all optimization levels.


-- 
           Summary: [4.0/4.1/4.2 Regression] ICE in cp_expr_size or
                    miscompilation
           Product: gcc
           Version: 4.1.1
            Status: UNCONFIRMED
          Severity: critical
          Priority: P3
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jakub at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27115