Software restructuring using hierarchical clustering

Abstract

Bad designs in software code have a significant impact on the total cost incurred in the development of software. This is because software code with bad designs has poor structure, which decreases its readability, understandability and maintainability. Software restructuring is thus a crucial activity in software development. Cohesion is an important measure in assessing the quality of software. The cohesion of a software module is the degree to which module components belong together. An ill-structured software code is characterized by low cohesion. Software restructuring techniques based on hierarchical agglomerative clustering (HAC) algorithms have been widely used to restructure large modules with low cohesion into smaller modules with high cohesion. These techniques generate clustering trees (or dendrograms) of the modules. The clustering trees are then sliced at different cut-points to obtain the desired restructurings. Choosing the appropriate cut-points is a difficult problem in clustering. This problem is exacerbated in previous HAC techniques as those techniques generate clustering trees which have a large number of cut-points. Moreover, many of those cut-points return clusters of which only a few lead to a meaningful restructuring. In this thesis, we propose a new hierarchical clustering technique for restructuring software at the function-level that generates clustering trees where the number of cut-points is reduced, and the quality of the cut-points is improved. To establish this we compare the results of our technique with those of four previous hierarchical clustering algorithms. We also develop an easy-to-use software tool that allows the user to generate clustering trees of functions using five different clustering algorithms, including the algorithm proposed in this thesis. Finally, we give a characterization of clusters returned by cut-points, in the context of software restructuring.