Learning to Rank Extract Method Refactoring Suggestions for Long Methods

摘要

Extract method refactoring is a common way to shorten long methods in software development. It improves code readability, reduces complexity, and is one of the most frequently used refactorings. Nevertheless, sometimes developers refrain from applying it because identifying an appropriate set of statements that can be extracted into a new method is error-prone and time-consuming.In a previous work, we presented a method that could be used to automatically derive extract method refactoring suggestions for long Java methods, that generated useful suggestions for developers. The approach relies on a scoring function that ranks all valid refactoring possibilities (that is, all candidates) to identify suitable candidates for an extract method refactoring that could be suggested to developers. Even though the evaluation has shown that the suggestions are useful for developers, there is a lack of understanding of the scoring function. In this paper, we present research on the single scoring features, and their importance for the ranking capability. In addition, we evaluate the ranking capability of the suggested scoring function, and derive a better and less complex one using learning to rank techniques.