On the computation of the probability of a system failure