Developers may introduce poor code structures, often realized by code smells, along the software maintenance. However, only some of these structures eventually become obstacles to the addition of new features. Developers are forced to remove these obstacles before adding features. Identifying which poor structures are actual obstacles is hard, due to their subtlety and scattering across code elements. Such identification has been largely debated by developers in public platforms like Gerrit Code Review. Fully removing obstacles is also hard even when developers apply sets of interrelated code transformations. Despite enabling the feature addition, certain sets of transformations tend to introduce rather than remove code smells. The scarce knowledge on recurring obstacles, and how refactoring can cope with them, helps little in guiding the feature addition. This doctoral research aims to address the current scarceness via empirical studies with developers and projects. Our major goal is three-fold: (1) to assess past feature additions in order to elicit recurring obstacles; (2) to understand when interrelated transformations unexpectedly introduce poor code structures; and (3) to propose a refactoring recommender system. Contrarily to the few existing ones, our system aims to guide developers along the feature addition while removing poor code structures.

PhD student, OPUS Research Group at PUC-Rio