Characteristics of method extractions in Java: a large scale empirical study

Extract method is the “Swiss army knife” of refactorings: developers perform method extraction to introduce alternative signatures, decompose long code, improve testability, among many other reasons. Although the rationales behind method extraction are well explored, we are not yet aware of its characteristics. Assessing this information can provide the basis to better understand this important refactoring operation as well as improve refactoring tools and techniques based on the actual behavior of developers. In this paper, we assess characteristics of the extract method refactoring. We rely on a state-of-the-art technique to detect method extraction, and analyze over 70K instances of this refactoring, mined from 124 software systems. We investigate five aspects of this operation: magnitude, content, transformation, size, and degree. We find that (i) the extract method is among the most popular refactorings; (ii) extracted methods are over represented on operations related to creation, validation, and setup; (iii) methods that are targets of the extractions are 2.2x longer than the average, and they are reduced by one statement after the extraction; and (iv) single method extraction represents most, but not all, of the cases. We conclude by proposing improvements to refactoring detection, suggestion, and automation tools and techniques to support both practitioners and researchers.