Research On Defect Correlation And Its Application

The development of computer applications has been promoted by the development of computer software technology greatly, especially in the development and usage in some critical application system such as bank credit system and ATM etc., and software quality has become the most important attributes of software products. Software defects are widespread and harmful, not only to cause significant economic loss, even to endanger personal safety. The research on defects detection, assessment and prevention is an important computer fields to improve software quality and enhance software reliability.But software defects are not independent in actual testing process, and the correlation relation is failure correlation. The correlation relation is caused by control flow and data flow between them from the analysis of defects, and data flow is not considered in defects detection and assessment.Mostly, defects correlation is caused by the defect detected capability masked by other defects. Defects correlation affects the result of the test and makes the results not accurate enough as expected, and distorts the estimation results of software reliability assessment models. Software defect correlation is a universal phenomenon. Combined with the research on 863 project and from the standpoint of defects themselves, we make detailed analysis of defects correlation and gives the reasons of the software testing and reliability assessment out of action.With respect to software reliability assessment, by applying the generalized correlation to improve the existing reliability assessment models, we put forward the P-NHPP (Phase Nonhomogeneous Poisson Process) reliability model to make the reliability assessment parameters more accordance with the actual defects number. Considering data dependence for improving the software reliability models, P-NHPP as an improved model for NHPP (Nonhomogeneous Poisson Process) directs at the assumaption of independence between defects, and builds P-NHPP model. Starting with data acquisition for software reliability assessment, P-NHPP applied defect correlation to G-O model and is analysed by Mean Square of Fitting Error and Mean Error of Prediction. Experimental results showed that P-NHPP gets better fitting affection and has a fairly accurate prediction capability.With software automation testing, mutation testing is introduced to our research. As a testing strategy to evaluate the completude of test cases, mutation testing has been identified as a"fault-oriented"technique for unit testing, which is mainly used to generate complete test cases. Among them, the path-wise technique of test data generation is a high-efficiency technique for test cases generation, implements test data generation by building and solving constraint systems. However, most of path-wise generation techniques only take the control dependence among statements into consideration, viz, build constraint system by analyzing the control flow graph but neglecting the data dependence among statements. Considering both of them, this paper proposes a new domain reduction method named DRD (Domain Reduction Approach with Data Dependence) for improving the test data generation technique of domain reduction. DRD adds the data dependence to the program executive path and builds path with data dependence (PDD). On the base of PDD, DRD builds and solves the constraint system.Experimental results showed that this method improves the successful rate and execution efficiency of test data generation at a large extent, although it couldn't find the optimum solution for test data generation.