Reconsidering Classes in Procedural Object-Oriented Code

Object-oriented software may show signs of procedural thinking because of lack of design or due to design erosion over a period of time. We refer to such a software as procedural object-oriented code. Huge classes, scarce class hierarchies and absence of classes for domain entities are hallmarks of procedural object-oriented code. Due to huge investments in such systems, software restructuring becomes necessary. To support code modularization, it is important to identify useful domain abstractions. In this paper, we present a tool-assisted technique to identify useful abstractions and class hierarchies in procedural object-oriented code. During this task, principal classes (draft classes) are identified. Afterwards, composition and association relationships are inferred for principal classes. Lastly, Formal Concept Analysis (FCA) is used to analyze hierarchical relationships between methods and attributes within principal classes. We validated our approach on several case studies and report our results on an industrial case.