Research On Coincidental Correctness In Fault Localization

In recent years, automation of software debugging methods have rapidly developed. All kinds of fault localization methods assist software developers to debug. Compared to the traditional manual testing, automated software debugging methods are more efficient.However,among the test cases usually have a kind of tests called coincidence Correctness(Coincidental Correctness, referred to as CC), at the same time,this kind of test cases is often overlooked. Because of the existence of this kind of test cases, the accuracy of fault localization methods will be affected a lot,especially for Spectrum-based Fault Localization, referred to as SBFL.What is more important is that now usually fault localization methods do not consider this kind of test cases for the influence of the error precision.Therefore, in order to solve the related problem,This paper mainly research coincidence accuracy test cases for fault localization accuracy,the influence of validity and how to avoid coincidence test cases on the accuracy of fault localization.First of all, this paper will clarify that correctness coincidence test cases exist universally and have a bad effect on fault localization accuracy.Secondly, in order to reduce the coincidence test cases for locating methods SBFL error accuracy, to improve the error code lines of dubious value, this paper proposes an method based weight to test to inhance the fault localization accuracy.This method will distribute difference weight to passed tests according to different weight calculated function.In addition, this paper proposes a another method based Iterative prediction to improve the error code lines of dubious value.The main idea of the method is by iterative method, prediction accuracy of coincidence N is the number of test cases, and then the correctness suspicious coincidence that test cases to the candidate value for sorting, get rid of dubious value higher N coincidence correctness before test cases, until you find false statements, or seeking candidates for the number of test cases is less than the number of iteration to predict N.Finally, using the Siemens Suit test suite to test the effectiveness of the two methods and get to the conclusion that both of our methods are effective.