Object-oriented cohesion as a surrogate of software comprehension: an empirical study

Abstract

The concept of software cohesion in both the procedural and object-oriented paradigm is well known and documented. What is not so well known or documented is the perception of what empirically constitutes a cohesive 'unit' by software engineers. In this paper, we describe an empirical investigation using object-oriented (OO) classes as a basis. Twenty-four subjects (drawn from IT experienced and IT inexperienced groups) were asked to rate ten classes sampled from two industrial systems in terms of their overall cohesiveness; a class environment was used to carry out the study. Four key results were observed. Firstly, class size (when expressed in terms of number of methods) did not tend to influence the perception of cohesion by any subjects. Secondly, well-commented classes were rated most highly amongst both IT experienced and inexperienced subjects. Thirdly, the empirical study suggests that cohesion comprises a combination of various class factors including low coupling, small numbers of attributes and well-commented methods, rather than any single, individual class feature per se. Finally, the research supports the view that cohesion is a subjective concept reflecting a cognitive combination of class features; as such it is a surrogate for class comprehension.