Research On Fault Classification Oriented Spectrum Based Fault Localization

The process of fault localization is one of the most important part in software testing.How to locate the fault statements from a complex program accurately and quickly has become an important research direction.In the traditional approaches of software fault localization,the spectrum based fault localization is the commonly-used approach which mainly utilizes test coverage information and test cases execution results,and uses the suspicious formulas to calculate the suspiciousness values of each program entities in order to help programmers identify the location of faults.SBFL has assisted in finding faulty statements,but it has a low precision on fault localization,especially for complex programs.The program need to inspect the statements manually to identify the fault statements,and the work is still tedious and time consuming.This disadvantage limits the applications of SBFL in software fault localization.Aiming at the low precision of SBFL on fault localization,this thesis presents an improved localization approach,namely Fault Classification Oriented Spectrum Based Fault Localization(FC-SBFL).Firstly,a new fault classification approach that based on the characteristics of statements is proposed.This fault classification calculates the probability of being fault for each kind of statement by using the principle of statistical computation.Furthermore,this fault classification is combined with the traditional SBFL approach.Utilizing the function of fault classification can update the suspiciousness value of statements,further improve the accuracy of fault localization.In addition,this thesis analyzes the principle that is why FC-SBFL can improve the precision on fault localization,and then gives a formal description with the principle.In order to verify the validity of FC-SBFL approach,the empirical studies use seven packages of the Siemens Suite which are all open sources of C programs and select 132 faulty versions from SIR repository.We analyze and compare the precision of fault localization between FC-SBFL approach and SBFL approach on different suspiciousness formulas.The empirical results show that more than 79.0%faults have been identified in a better effectiveness when using FC-SBFL approach,and provide a powerful support for the practical application of FC-SBFL approach.