On second-order logic