Abstract

Detecting refactoring instances occurred in successive revisions of software systems can provide wealthy information for several purposes, e.g., to facilitate the code review process, to devise more accurate code merging techniques, to help the developers of API clients to ease their adaptation to API changes, and to enable more accurate empirical studies on the refactoring practice. In the literature there are several techniques proposed for refactoring detection, supporting a wide variety of refactoring types. Yet, almost all of them have missed an extensively-applied refactoring type, i.e., Rename Local Variable refactoring. In addition, all these techniques rely on similarity thresholds (which are difficult to tune), or need the systems under analysis to be fully built (which is usually a daunting task), or depend on specific IDEs (which drastically limits their effectiveness and usability).
In this thesis, we extend the state-of-the-art refactoring detection tool, RefactoringMiner, by defining necessary rules and extending its core algorithms to tailor it for accurately detecting Rename Local Variable refactoring instances. We have evaluated the proposed technique on two large-scale open-source systems, namely dnsjava and Tomcat. Our comparison with REPENT, the state-of-the-art tool in detecting Rename Local Variable refactoring instances, shows that our approach is superior in terms of precision and recall. Moreover, to automatically create a reference corpus of refactoring instances which is required for the evaluation, we have built a fully-automated infrastructure (called RefBenchmark) that is able to invoke several refactoring detection tools and find the agreements/disagreements between their results.