Using a many-objective approach to investigate automated refactoring

Abstract Context Software maintenance is expensive and so anything that can be done to reduce its cost is potentially of huge benefit. However, it is recognised that some maintenance, especially refactoring, can be automated. Given the number of possible refactorings and combinations of refactorings, a search-based approach may provide the means to optimise refactorings. Objective This paper describes the investigation of a many-objective genetic algorithm used to automate software refactoring, implemented as a Java tool, MultiRefactor. Method The approach and tool is evaluated using a set of open source Java programs. The tool contains four separate measures of software looking at the software quality as well as measures of code priority, refactoring coverage and element recentness. The many-objective algorithm combines the four objectives to improve the software in a holistic manner. An experiment has been constructed to compare the many-objective approach against a mono-objective approach that only uses a single objective to measure software quality. Different permutations of the objectives are also tested and compared to see how well the different objectives can work together in a multi-objective refactoring approach. The eight approaches are tested on six different open source Java programs. Results The many-objective approach is found to give better objective scores on average than the mono-objective approach and in less time. However, the priority and element recentness objectives are both found to be less successful in multi/many-objective setups when they are used together. Conclusion A many-objective approach is suitable and effective for optimising automated refactoring to improve quality. Including other objectives does not unduly degrade the quality improvements, but is less effective for those objectives than if they were used in a mono-objective approach.