Research On Model Transformation Fault Localization Approach Based On Spectrum

Model-Driven Development(MDD)is currently widely used in service-oriented architecture information systems.As an important part and core technology of MDD,model transformation directly affects the success of software systems based on MDD.With the increasing complexity and scale of model transformations,the reliable guarantee of model transformation becomes more and more difficult.Debugging model transformation programs also face many challenges,and efficient and accurate fault localization approaches can help improve the debugging efficiency,so as to effectively guarantee the quality of model transformations.At present,researchers propose some automatic fault localization approaches to improve the debugging efficiency of model transformations.Among them,Spectrum-Based Fault Localization in model transformations(SBFL),as a representative dynamic analysis approach,mainly uses the rule coverage information and execution results to estimate the probability that each rule may be faulty.However,due to the characteristics of model transformations,the widespread existence of false-negative and false-positive results in the coverage information,which makes the fault localization results of the spectrum approach are not accurate.Therefore,we consider whether we can further improve the accuracy of fault localization by mining the covered range information of different test models.In this thesis,we propose a method to optimize the SBFL based on the impact degree of the test model and apply it to the fault localization of model transformations.This method evaluates the impact of different test models on fault localization based on the covered range of the test model,and then optimizes the fault localization results of the spectrum method based on the coverage rule information of the test models with different impacts.We will compare the proposed method with SBFL,and take the open-source model transformation projects as an example to verify the feasibility and effectiveness of the approach.The main work of this thesis can be summarized as follows:(1)The influence of rule inheritance relation on fault localization accuracy in model transformation is considered.Through static analysis of the inheritance relationship between rules,considering the influence of error propagation,the inheritance weight between rules is further calculated to solve the problem that it is difficult to distinguish true error rules owing to the same suspiciousness value in spectrum method.(2)Through the dynamic analysis of the model transformations execution,the covered range information of rules under different test models is used to evaluate the fault localization ability of different test models,and different weights are assigned to test models with different degrees of impact.The failing test model with smaller covered range is assigned more weight,and the rule weights are iteratively calculated according to the idea of the web page ranking algorithm,and then the localization results of the spectrum method are adjusted according to the coverage information of the weighted model to improve the accuracy of fault localization.(3)Related experiments are designed to evaluate the feasibility and location efficiency of the approach.We selected 12 suspiciousness calculation techniques that are widely used in the field of fault localization and carried out experiments on three open-source model transformation projects of different sizes.The experimental results show that compared with the SBFL approach,the fault localization effectiveness of the proposed approach can be improved by 25%on average.