JFeature: Know Your Corpus

Software corpora are crucial for evaluating research artifacts and ensuring repeatability of outcomes. Corpora such as DaCapo and Defects4J provide a collection of real-world open-source projects for evaluating the robustness and performance of software tools like static analysers. However, what do we know about these corpora? What do we know about their composition? Are they really suited for our particular problem? We developed JFEATURE, an extensible static analysis tool that extracts syntactic and semantic features from Java programs, to assist developers in answering these questions. We demonstrate the potential of JFEATURE by applying it to four widely-used corpora in the program analysis area, and we suggest other applications, including longitudinal studies of individual Java projects and the creation of new corpora.