Improving Cluster Test Selection Techniques Of Regression Testing By Slice Filtering

Regression testing is a software task conducted on modified software. Its purpose is to verify that modifications do not bring about adverse impact on the unchanged portion of the program. However, this task is very time-consuming since it requires rerunning all existing test cases. Regression test selection technique selects a subset of the existing test suite to retest and thus reduces the cost of regression testing.Cluster test selection is a novel regression test selection technique. It selects test cases basing on the similarities of their software behaviors. It conducts a cluster analysis on the execution profiles that are collected during previous software developing phases. Test cases exhibiting similar software traits are assigned into one group. Representative test cases are sampled from each of those groups to form a new test suite. Many sampling strategies can be employed, among which adaptive sampling strategy is an effective and popular sampling strategy.In this paper, program slicing is employed to improve the cluster test selection technique. First, a static slice is computed from the modified code. The intersection of the program slice and the execution profile of each test case is computed. This intersection is used to highlight the parts of software affected by modification. It reduces the size of the execution profiles of test cases significantly. Then cluster analysis is conducted on these intersected profiles. A few test cases will be sampled out from each cluster to form a small regression test suite. The slice filtering gives more potential chances to deal with large software by reducing the dimensions of cluster analysis. The scalability of the cluster test selection technique is significantly improved by slice filtering. The experiment shows that fault-revealing test cases tend to cluster together during cluster analysis and a regression test suite sampled from these clusters has sufficient fault detection capability.