Evaluating the effectiveness of risk containers to isolate change propagation

Abstract

Previous studies indicate that error-proneness risks can be isolated into risk containers created from architectural designs, to help detect and mitigate such risks early on. Like error-proneness, change propagation may lead to higher implementation and maintenance costs. We used automated tools to analyse four software development projects using three risk container types, each type based on a different architectural perspective. A strong and significant correlation between design change propagation and implementation co-change was observed for all three container types. We found that Design Rule Containers (DRCs), based on class diagrams, are the most effective for isolating change propagation because they have the least amount of container overlap, highest levels of internal coupling, highest co-change probability between classes that share containers, and the most change sets isolated in containers. Developers from two projects were able to justify why design dependencies had resulted in the top five DRCs being predicted to isolate the most change propagation. This and the previous error-proneness research suggests DRCs are an effective technique to detect and contain code maintainability risks at the design stage. These results provide some evidence that class diagrams are more useful than use case sequence diagrams for analysing maintainability risks in designs.