Research On Software Fault Localization Method Based On Fault Propagation Context Analysis

Along with the increasing scale of software system, it becomes more and moredifficult to ensure the quality of the software. The probability of the software failure alsoincreases. The software failure means that the software expected outcome is differentfrom the actual outcome. After the software failure occurred, how to locate the faultystatements which contain bugs is what fault localization needs to solve. Now there aremany excellent software fault localization techniques. However, they all ignore thecoincidental correctness which means execution results of test cases that execute faultystatements indicate no failure information. The coincidental correctness is alwayscaused by the fault propagation. Based on the above considerations, this paper dividesthe fault localization into two steps, including fault-related statements localization andfault comprehension.The research contents of this paper include: Firstly, in order to get the executionpath information of the program as the input of subsequent modules, programinstrumentation is needed. However, existing program instrumentation methods are toocomplex. In this paper, we use lexical analysis and syntax analysis to realize a simpleand convenient program instrumentation method. Secondly, in order to do thefault-related statements localization, combined dependence network must be created. Soit is necessary to do program dependence analysis and create program dependence graphand transfer it. In this paper, we realize the traditional method of the programdependence analysis. Based on this, we create program dependence graph, transfer somespecific nodes and do the state specification. Thirdly, existing fault localization methodsnot only can not accurately locate the fault in most cases but also can not locate thefault-related statements. So we propose conditional probability, path probability andcombined dependence probability to enhance the accuracy of fault-related statementslocalization. In the end, coincidental correctness sharply affect the result of faultlocalization. Existing fault localization methods didn’t grasp its essence. So we proposethe fault comprehension based on fault propagation context analysis to eliminate itseffect.The method generates a ranked list of suspiciousness score of all statements.Higher the score is, more suspicious the corresponding statement is.In this paper, we use Siemens suit which has been widely used in the area ofsoftware fault localization to test. We compare our approach with other traditionmethods on localization effectiveness and stability. Experimental results indicate thatour approach greatly improves the localization effectiveness and stability and results presentation is more conducive to help developers to find the root cause of faults.