An exploratory study on change suggestions for methods using clone detection
Manishankar Mondal, Chanchal Kumar Roy, Kevin A. Schneider · Computer Science and Software Engineering · 2016
A number of studies investigated providing change suggestions to programmers on the basis of the evolution history of a software system. While existing studies provide change suggestions considering code fragment level or even line level granularities, we investigate providing change suggestions at the method level. Providing a suggestion to change the entire method at one time is intuitively more time saving for developers compared to providing suggestions separately for different fragments of a method. In this research we empirically investigate whether we can infer change suggestions at the method level by analyzing the past evolution history of a software system through detection of method clones, and if so, then how we can rank the method level change suggestions. According to our investigation on thousands of commits of seven diverse subject systems, we can provide change suggestions at the method level with up to 83% precision and 13.49% recall. Moreover, for up to 34% of the commits we can provide correct method level change suggestions. Compared to the existing fragment level change suggestion techniques, our method level change suggestion technique has promising precision and recall. We investigate the ranking of method level change suggestions and find that recency ranking (i.e., ranking on the basis of how recently the change suggestions appeared in the past) is a better choice than frequency ranking (ranking considering how frequently the suggestions appeared). We believe that while a method level change suggestion technique can never be a replacement for the existing fine grained change suggestion techniques, it can complement these existing ones.