Up to 60% of the refactorings in software projects are constituted of a set of interrelated transformations, the so-called batches (or composite refactoring), rather than single transformations applied in isolation. However, a systematic characterization of batch characterization is missing, which hampers the elaboration of proper tooling support and empirical studies of how (batch) refactoring is applied in practice. This paper summarizes the research performed under the context of a Master's dissertation, which aimed at taming the aforementioned problems. To the best of our knowledge, our research is the first work published that provides a conceptual foundation, detection support and an large impact analysis of batch refactoring on code maintainability. To this end, we performed two complementary empirical studies as well as designed a first heuristic aimed at explicitly detecting batch refactorings. Our first study consisted of a literature review that synthesizes the otherwise scattered, partial conceptualization of batch refactoring mentioned in 29 studies with different purposes. We identified and defined seven batch characteristics such as the scope and typology of batches, plus seven types of batch effect on software maintainability, including code smell removal. All batches' characteristics and possible impacts were systematized in a conceptual framework, which assists, for instance, the proper design of batch refactoring studies and batch detection heuristics. We defined a new heuristic for batch detection, which made it possible to conduct a large study involving 4,607 batches discovered in 57 open and closed software projects. Amongst various findings, we reveal that most batches in practice occur entirely within one commit (93%), affect multiple methods (90%). Surprisingly, batches mostly end up introducing (51%) or not removing (38%) code smells. These findings contradict previous investigations limited to the impact analysis of each transformation in isolation. Our findings also enabled us to reveal beneficial or harmful patterns of batches that respectively induces the introduction or removal of certain code smells. These patterns: (i) were not previously documented even in Fowler's refactoring catalog, and (ii) provide concrete guidance for both researchers, tool designers, and practitioners.
Are Multi-Language Design Smells Fault-Prone? An Empirical Study
