Network-Based Analysis of Software Change Propagation

Abstract EN

The object-oriented software systems frequentlyevolve to meet new change requirements.

Understanding thecharacteristics of changes aids testers and system designersto improve the quality of softwares.

Identifying importantmodules becomes a key issue in the process of evolution.

Inthis context, a novel network-based approach is proposedto comprehensively investigate change distributions and thecorrelation between centrality measures and the scope ofchange propagation.

First, software dependency networksare constructed at class level.

And then, the number oftimes of cochanges among classes is minded from softwarerepositories.

According to the dependency relationships andthe number of times of cochanges among classes, thescope of change propagation is calculated.

Using Spearmanrank correlation analyzes the correlation between centralitymeasures and the scope of change propagation.

Three casestudies on java open source software projects Findbugs, Hibernate, and Spring are conducted to research the characteristicsof change propagation.

Experimental results showthat (i) change distribution is very uneven; (ii) PageRank, Degree, and CIRank are significantly correlated to the scopeof change propagation.

Particularly, CIRank shows highercorrelation coefficient, which suggests it can be a more usefulindicator for measuring the scope of change propagation ofclasses in object-oriented software system.