An empirical study of computation equivalence as determined by decomposition slice equivalence

Abstract

In order to further understand and assess decomposition slicing we characterize and evaluate the size of reductions obtained by computing equivalent decomposition slices from the perspective of the comprehender, maintainer, tester and researcher. The analysis was performed on 68 C language systems of sizes 100 to 50,000 lines. All decomposition slices were computed and compared for simple equality. From this data, we were able to determine with 95% confidence that the true mean percentage of equivalent decomposition slices is between 50.0% and 60.3%, with a p-value < 0.005. This has clear and significant impact for software testing, as any coverage method used for one of the variables used in an equivalence will apply to all variables in the class; for software comprehension as the number of items (variables) used for the understander is substantially reduced; for the software maintenance, as the number computational relationships is reduced; and for the researcher, in attempting to ascertain the underlying cause of this phenomena.