Issue #23049: Pure python equivalent shouldn't imply more exactitude than is really there.