A Dynamic Failure Model for Predicting the Impact that a Program Location has on the Program

This paper presents a dynamic technique for predicting the e ect that a \location" of a program will have on the program's computational behavior. The technique is based on the three necessary and su cient conditions for software failure to occur: (1) a fault must be executed, (2) the fault must adversely a ect the data state, and (3) the adverse e ect in a data state must a ect program output. In order to predict the e ect that a location of a program will have on the program's computational behavior, the following characteristics of each program location are estimated: (1) the probability that a location of the program is executed, (2) the probability that a location of the program noticeably a ects the program state created by the location, and (3) the probability that the data states created by a location a ect the program's output. With estimates of these characteristics for each location in a program, we can predict those locations where a fault can more easily remain undetected during testing, as well as predict the degree of testing necessary to be convinced that a fault is not remaining undetected in a particular location.