Current implementation defines an adhoc concatenating iterator etc, which creates some additional overhead. Refactoring so that this internal API takes two sets results in less complexity and slightly more efficient code.