Using semantic differencing to reduce the cost of regression testing

An algorithm is presented that reduces the cost of regression testing by reducing the number of test cases that must be rerun and by reducing the size of the program that they must be run on. The algorithm uses dependence graphs and program slicing to partition the components of the new program into two sets: preserved points-components that have unchanged run-time behaviour; and affected points-components that have changed run-time behavior. Only test cases that test the behavior of affected points just be rerun; the behavior of the preserved points is guaranteed to be the same in the old and new versions of the program. Furthermore, the algorithm produces a program 'difference', which captures the behavior of (only) the affected points. Thus, rather than restricting the (large) new program on a large number of test cases, it is possible to certify the new program by running the (smaller) program 'differences' on a (smaller) number of test cases.<<ETX>>