Research Of Software Fault Localization Technique Based On Dependency Observation Matrix

Software testing is time-consuming and inefficient task in the process of software development, while the testing time is mainly decided by fault localization, i.e., finding out the primary cause of the software failure. Although a number of automated fault localization techniques have been proposed by experts and scholars and some have delivered promising results, multiple faults and the number of test cases may cause problems. Motivated by the success of probabilistic and statistical methods applied in other research fields, many experts and scholars recently exploit some specific test cases to execute the program and obtain the corresponding information combined, and incorprate them into a probabilistic method to locate the faulty statements. These methods have more obvious improvement compared to formerly method in terms of effectiveness.Surveying the state-of-the-art fault loclization techniques, we propsoe dependency observation matrix to improve Abreu’method. We obtain the program dependency through the static analysis of abstract syntax tree dump (AST-dump) file, and then instrument the program to construct the dependency matrix in order to get the confilict sets and the mininal hitting sets. We utilize the Bayes’s formula to update the probability of each hitting sets and compute the most suspicious one. Finally we use the depency pairs in the most suspicious set to pinpoint the faulty statement. Based on this idea,we designed and implemeted a protype system for software fault localization.To estimate the effectiveness of our approach, we make use of the well-known Siemens software benchmark as our experimental subject, and take the famous metric score-the ratio of the lines that must be checked and the total lines-for comparison. We compute the score values for each version of the programs in Siemens, and summarize the final results to compare to the2types of fault localization techniques. Our approach is more effetive and promising than other appraoches through the experimental result, and more important, our appraoch can aid developers to find out the root causes of software failure because of the use of the program dependency information.