Some future directions in computational failure mechanics