A new approach to domain testing

A new method of domain testing is proposed in this thesis, which, unlike the original method, is low cost, effective and applicable to any type of borders which define a path domain. To show that the new method is effective for detecting boundary errors at moderate cost, a thorough analysis of the problem and several probabilistic models have been presented. The basic idea of domain testing is to require points on and near the borders of a path domain to be tested. An efficient algorithm which will determine these points, given a path domain, therefore is crucial for successfully implementing this testing technique. The complexities of this problem under various types of borders is first presented. Then the algorithms to generate ON points for linear bounded domains, nonlinear bounded domains, and discrete domains are described. Although domain testing is a path oriented technique, in the past, there has been no research describing how to select a small subset of representative paths to test a given program. A new path selection criteria for this purpose is therefore proposed in this thesis. The criteria combine the advantages of both data flow testing and domain testing. An algorithm to permit the automatic generation of paths which satisfy a selected criterion is also presented. Traditionally, the existence of unexecutable paths is an unavoidable problem for such an automatic algorithm. However, the algorithm presented here deals with this problem by always enumerating executable paths to exercise. Finally, the applicability of the abstract model of domain testing for real applications are discussed.