Research On Program Fault Localization Based On The Execution Path Clustering

Software, as the soul of computer, will affect the normal work once it faults. It may lead to huge losses, even to the safety of people’s lives and property. Program errors are the major cause of software faults, so locating the program error automatically, quickly and accurately can effectively improve the software reliability. In recent years, great achievements have been made in the field of the program fault localization technology. The TBFL method based on the test is a kind of intelligent method in locating program error by using test-covering information. Test-based fault localization (TBFL) method requires a large number of test cases to collect the test-covering information which means a large amount of redundancy. It will get a poor result of fault localization when the calculation method is sensitive to the test cases.On the basis of previous studies, the FCM clustering method was applied to handling with the program execution paths according to the limitation of the existing mainstream methods. This method firstly collected all the test cases status information to save a trace files with LOUPE tools. Secondly, the trace files were analyzed using the MATLAB and then program executing path was express with a multivolume vector. On the basis of the general architecture of FLOC, program executing path was collected with the clustering algorithm. A large number of the same or similar execution paths were divided to fewer typical paths with the FCM algorithm. The typical clustered paths were selected as the test information. Finally, fault localization was implementing by the difference contrast or combining existing difference calculation methods.The algorithm of FCM was introduced to locate program fault in this paper. By the FCM, a large number of experiments were done on the Siemens data set and compared with Tarantula, SOBER and SBI algorithm. The experiments showed that the clustering method could reduce the sensitivity to the test case and could get the same effect with large numbers only using fewer test cases, so the method could reduce the path redundancy and improve the effect of position.