Empirical Evaluation of Strategies to Detect Logical Change Dependencies

摘要

Change impact analysis plays an immanent role in the maintenance and enhancement of software systems. There still exist many approaches to support change impact analysis. In the last years researchers try to utilize data in software repositories to gain findings for supporting miscellaneous aspects of software engineering, e.g. software evolution analysis or change impact analysis. In the context of change impact analysis, approaches (=strategies) try to detect logical dependencies among artifacts based on the version histories of files in the concurrent versioning system (e.g. CVS). They try to infer logical couplings of files (artifacts) based on co-changes (files which are frequently changed together). Based on these findings we want to contribute with the presentation of insights of deeper investigation of historical information in concurrent versioning systems in general. In this paper we have identified and described existing strategies to detect logical change couplings. These strategies will be illustrated by practical use cases. We have empirically evaluated these strategies based on versioning system repositories of two industrial projects. The analysis figures the absolute and relative contribution of dependency results per strategy. Furthermore we show overlappings of dependency results.