Would lazy eval be able to eliminate temps in doing operations such as: np.sum (u != 23)? That is, now ops involving selecting elements of matrixes are often performed by first constructing temp matrixes, and the operating on them.