Research On Memory Selection Technology In ART And Its Application In OOS Testing

With the rapid development of Object-Oriented Programming(OOP),a large number of Object-Oriented Software(OOS)has emerged.The reliability and quality of OOS have drawn a lot of attention.Among current software testing approaches,Random Testing(RT)is always the most widely used,but its failure-detection efficiency and effectiveness are not so satisfactory.Under the circumstances,T Y Chen has proposed the Adaptive Random Testing(ART)based on the improvement of RT,with research results showing that ART applied in OOS testing can make much better failure-detection effectiveness than RT.The key point in ensuring the failure-detection effectiveness of ART is to make sure that test cases are as evenly distributed as possible throughout the whole test input field during test execution.When ART is executing,two test case sets will be generated namely the executed test case set and the candidate test case set,and the test case in candidate test case set with the farthest distance away from the executed test case set will be selected as the next test case for execution.If one test case cannot find a failure,it will be added into the executed test case set after its execution.According to the above description of ART,we can draw the conclusion that the executed test case set will be continuously expanded during the process of testing execution,and which will lead to more testing time consuming or even reduce the failure-detection effectiveness.In order to solve such problem,this thesis is focused on studying a feasible and effective memory selection technology in ART and to apply the proposed approaches to OOS testing in order to improve the failure-detection effectiveness.In this thesis,two approaches are proposed namely the approach of ART with k-means clustering for OOS testing and the approach for determining the optimal cluster number k in ART with k-means clustering.Meanwhile,the experimental analysis has been carried out to prove the efficiency and effectiveness of the proposed two approaches for OOS testing.A k-means clustering Object Oriented Software Testing prototype Tool(kOOSTT)is also been designed and implemented in this thesis,and all experiments in this research have been carried out upon this tool.The main work of this thesis is given below.1.With the basic theories and the knowledge of related technologies of OOS testing based on ART,this research notes the importance of distance measure between the OO test cases in OOS testing field.In case of considering the special features of OO test cases,this research takes the method invocation sequence information into account,which is of great significance to the distance measure between OO test cases.Therefore,an OOS test case distance measure based on the Improved Wavelet Transform(IWT)and the approach of ART with k-means clustering for OOS testing named IWTClustering-ART(ART with Clustering based on Improved Wavelet Transform)are proposed in this thesis.The proposed distance measure based on IWT is used in the testing approach IWTClustering-ART to calculate the distance between OO test cases that will ensure the generated next test case can be capability in detecting failure effectively.In this study,the failure-detection effectiveness of the proposed IWTClustering-ART approach has been compared with WClustering-ART(ART with Clustering based on Wavelet Transform),OMISS-ART(ART with Object and Method Invocation Sequence Similarity),ARTOO(Adaptive Random Testing for Object-Oriented Software)and RT-ms(RT with method sequence),with the results showing that the proposed IWTClustering-ART has the best performance and it is of reliability and effectiveness in OOS testing.2.The approach of determining the optimal k value for clustering process in ART with k-means clustering is proposed in this thesis,and the experimental comparison analysis is carried out to show the effectiveness of the proposed approach.The optimal k value determination approach based on experimental process consists of two parts,namely an optimal k value solving model based on experimental process and an algorithm for optimal k value determination named kValue algorithm,which is suitable for determining the optimal k value for ART with k-means clustering.The kValue algorithm is devoted to determining the optimal k value for the clustering process of testing method IWTClustering-ART for specific tested subjects.Meanwhile,this research has carried lots of experimental analysis based on the proposed optimal k value solving model based on experimental process and the kValue algorithm,with the results showing that the approach of determining the optimal k value for clustering process in ART with k-means clustering is of feasibility and effectiveness,which indicates that the proposed approach is helpful with determining the most suitable k value for further improving the failure-detection effectiveness of the approach of ART with k-means clustering for OOS testing.Meanwhile,it will also provide some guidance to the following research work.3.A k-means clustering object oriented software testing prototype tool named kOOSTT is also been designed and implemented in this thesis.The main functional modules of kOOSTT include the class diagram entry module,the parameter analysis module,the optimal k value determination module,the testing execution module,and the testing result analysis module.kOOSTT is the experimental platform for the proposed approaches in this thesis,and some modules of which have achieved a high degree of automation during the testing process.In this research,the proposed kOOSTT is applied to to verify the effectiveness of the proposed approaches in OOS testing based on ART.The kOOSTT covers the entire testing process and it can be applied to verify the effectiveness of the proposed approach in OOS testing based on ART.