Refactoring trends across N versions of N Java open source systems : an empirical study

In the past few years, refactoring has emerged as an important consideration in the maintenance and evolution of software. Yet very little empirical evidence exists to support the claim about whether developers actively undertake refactoring, or whether as Fowler suggests that the benefits of doing refactoring are not short-term but too ‘long-term’ [8]. In this paper, we describe an empirical study of multiple versions of a range of open source Java systems in an attempt to understand whether refactoring does occur and, if so, which types of refactoring were most (and least) common. Fifteen refactorings were chosen as a basis (on seven Java systems) and the code analysed using an automated tool. Results confirmed that refactoring did take place, but the majority were of the simpler, less complex type. Interestingly, the most common refactorings empirically identified were those which, according to Fowler (and from a dependency graph of the ‘seventy two’ original refactorings), were central to larger more involved refactorings. One conclusion from the study is thus that developer time and effort for relatively large restructuring and testing of refactored code is prohibitive; making small, simple changes is preferred. A further conclusion from our study is that refactoring didn’t occur in the earliest or latest versions of the systems we investigated.