WETSoDA Framework Information Based Java Software Architecture Recovery

Transcription

1 WETSoDA 2017 Framework Information Based Java Software Architecture Recovery Xiaocong Li, Li Zhang, Ning Ge Beihang University, Beijing, China

2 OUTLINE Introduction Problem Statement Our Solution Experiment & Result Conclusion & Future Works

3 INTRODUCTION

4 INTRODUCTION Program comprehension is a major cost factor Lack of documents or the documents are out of date Architecture recovery methods are used Hierarchical clustering algorithm is used to divide source files into different modules The more features two entities have in common, the more likely they are in the same module

5 HIERARCHICAL CLUSTERING ALGORITHMS Examples Single Linkage Algorithm(SLA) Complete Linkage Algorithm(CLA) Weighted Combined Algorithm(WCA) scalable InforMation BOttleneck(LIMBO) We want to improve the accuracy using framework information Spring Struts Hibernate MyBatis

6 HIERARCHICAL CLUSTERING ALGORITHMS Feature selection Similarity calculation Entities mergence Repetition

7 Use the Spring MVC as an example PROBLEM STATEMENT

8 PROBLEM STATEMENT @Deprecated Model Model Controller Controller

9 PROBLEM STATEMENT @Deprecated Model Model Controller Controller

10 PROBLEM STATEMENT @Deprecated Model Model Controller Controller

11 PROBLEM STATEMENT Model1 Model2 Controller1 Controller2 Model Model Controller Controller

12 PROBLEM STATEMENT Module1 Model1 Controller1 Module2 Model2 Controller2

13 PROBLEM STATEMENT Module1 Model1 Controller1 Module2 Model2 Controller2

14 PROBLEM STATEMENT Module1 Model1 Controller1 Module2 Model2 Controller2 Model layer Model1 Model2 Controller layer Controller1 Controller2

15 PROBLEM STATEMENT Module1 Model1 Controller1 Module2 Model2 Controller2 Model layer Model1 Model2 Controller layer Controller1 Controller2

16 Use framework information OUR SOLUTION

17 FRAMEWORK INFORMATION Frameworks Spring Struts Hibernate MyBatis Annotations : Interfaces Super class Class name key words

18 FRAMEWORK INFORMATION Original FeatureVector v " = v " $, v " &,, v " ( Framework Feature Weight w * = w * $, w * &,, w * ( Weighted Feature Vector wv " = w * $ v " $, w * & v " &,, w * ( v " (

19 USE FRAMEWORK INFORMATION @Deprecated Model Model Controller Controller

20 USE FRAMEWORK INFORMATION @Deprecated Model Model Controller Controller

21 USE FRAMEWORK INFORMATION Model1 Model2 Controller1 Controller2 Model Model Controller Controller

22 PROBLEM STATEMENT Model layer Model1 Model2 Controller layer Controller1 Controller2

23 PROBLEM STATEMENT Model layer Model1 Model2 Controller layer Controller1 Controller2

24 EXPERIMENT AND RESULT

25 EXPERIMENT DESIGN Select projects using the Spring Framework ongithub Select framework specific features Run the algorithm with different feature weights Compare resultswith originalarchitecture

26 SELECTED PROJECTS Project Modules File Number LOC act cms gen oa sys SpringBlog

27 SELECTED FEATURES Usage Annotation Usage Annotation Usage Annotation Usage Annotation Implementation of Serializable Interface Implementation of JpaRepository Interface

28 ACCURACY CALCULATION Initial architecture $ M "(", = {M "(", &, M "(",,, M ( "(", } Recovered architecture M (/0 = {M $ (/0, M & (/0,, M ( (/0 } Accuracy Similarity M "(",, M (/0 = ;BC : ;<;= < < ; :; <?@ ; ;BC : ;<;=

29 RESULT ACCURACY WEIGHT OF DOMAIN FEATURES act cms gen oa sys SpringBlog

30 RESULT ACCURACY WEIGHT OF DOMAIN FEATURES act gen oa sys SpringBlog

31 RESULT ACCURACY WEIGHT OF DOMAIN FEATURES act cms gen oa sys SpringBlog

32 RESULT ACCURACY WEIGHT OF DOMAIN FEATURES cms

33 CONCLUSION AND FUTURE WORKS

34 CONCLUSION & FUTURE WORKS Framework information can be used to improve architecture recovery algorithm Future works More features should be used Recommend the weight using machine learning Apply the method to other systems Large-scale system should be tested

35 Q&A