Dynamic impact analysis: a cost-effective technique to enforce error-propagation

This paper introduces dynamic impact analysis as a cost-effective technique to enforce the error-propagation condition for detecting a fault. The intuition behind dynamic impact analysis is as follows. In a specific test-case, if an execution of a syntactic component has a strong impact on the program output and if the output is correct, then the value of that component-execution is not likely to be erroneous. To capture this intuition in a theoretical framework the notion of impact is formally defined and the concept of impact strength is proposed as a quantitative measure of the impact. In order to provide an infrastructure supporting the computation of impact strengths, program impact graphs and execution impact graphs are introduced. An empirical study validating the computation of impact strengths is presented. It is shown that the impact strengths computed by dynamic impact analysis provide reasonable estimates for the error-sensitivity with respect to the output except when the impact is via one or more error-tolerant components of the program. Potential applications of dynamic impact analysis in the area of mutation testing and dynamic program slicing are discussed.