Cohesion is structural, coherence is functional: different views, different measures

Traditionally, the cohesion of a software component is considered to be a characteristic of its internal structure, and most cohesion measures proposed so far measure cohesion through the similarity of its constituent parts. However, cohesion may also be interpreted as an externally observed functional property, without regard for the component's internal structure. One way of measuring functional cohesion would be to measure the similarity of usage patterns of a component's external clients. One such measure is defined in this paper using a generic system model and its associated mechanism for calculating object sizes as the foundation. The new measure is simple to understand, easy to automate, and flexible enough to be used at different levels of abstraction. Moreover, it satisfies the most important properties that a cohesion measure is expected to satisfy. Examples are provided to illustrate the concept and its possible uses in analyzing and re-packaging of the components of a software system.

1 Introduction

The object-oriented paradigm is used extensively throughout the industry, yet is still in need of a sound and comprehensive set of measures. Although a number of different measures have been proposed specifically for the object-oriented software systems [12], [16], a universal agreement about the optimal set of measures and measurement techniques has yet to be arrived at. This statement holds for measures of size and complexity, but to an even greater extent for measures of quality. Among the most important factors of quality are measures of coupling and cohesion of software products. This holds not only for traditional structured design paradigm, but also for modern object-oriented approaches [1]. In theory, both measures should be used to assess the quality of the design well before the actual code is written, which makes them potentially very attractive for the purposes of early estimation [18]. Thus, they could help improve our understanding of software product properties and ultimately improve the Quality of software designs.