A GRAPH-THEORETIC APPROACH TO THE VALIDATION OF SOFTWARE MAINTENANCE MODIFICATIONS.

摘要

The literature in the areas of software management and software engineering admits to a possible reduction in he reliability of software after modifications have been made. The validation of software, modifications is commonly referred to as retest, and has yet to be adequately resolved.;The approach taken here was to define alternative retest strategies, develop technique for their implementation using graph theory and 0-1 integ programming, examine their performance characteristics, and discuss practical implementation considerations. The six defined strategies were: (1) Rerun all previously executed tests. (2) Retest all testable paths through the changed code. (3) Rerun all tests which execute the changed code. (4) Retest all statements reachable from the changed code. (5) Retest all statements reaching to or reachable from the changed code. (6) Retest all statements reaching to the changed code and setting changed data, and reached from the changed code and using changed data.;Though strategy 2 was shown to be impractical, techniques were developed to implement each of the remaining five strategies. Then each strategy was implemented on three sample computer programs and its performance observed.;The results showed that using strategies 4, 5, and 6 generally reduced the amount of retesting as compared with the nominal strategies of 1 and 3. Even though there is a cost associated with running the retest model, the total cost of retesting (which includes rerunning tests and implementing the retest model) was shown to be significantly less for strategies 4, 5, and 6.;This dissertation addresses techniques for the validation of software modifications made during the maintenance phase. The general question addressed is "How can software modifications be validated?", and more specifically, "What quantitative techniques can be used to implement these strategies? Are these techniques feasible? If so, what are the implementation considerations?" These questions are critical to software practitioners, because the ad hoc retest selection techniques of the past have proved woefully inadequate.