The special floating point comparisons do not detect the "unordered" case when the imaginary components are not NAN, but one or both of the real components is NAN.

Also, I have a specification question: Does D define an ordering for complex numbers?  If not, should using the other relational operators be an error?

int main() {
  cdouble a = double.nan + 3i;
  cdouble b = double.nan + 4i;

  assert( ! (a <> b) );
  assert( ! (a <>= b) );
  assert( a !<>= b );
  assert( a !<> b );

  return 0;
}

David