Towards a Theory of Software Evolution - And its Practical Impact

After years of study of software evolution processes, most recently through the FEAST projects, it appears that now there is a sufficient body of knowledge that can serve as basis for the development of an axiomatic theory of software evolution. Such body is represented by, for example, patterns and behavioural invariants observed in attributes of industrially evolved systems. This paper discusses the need for such a theory, its practical impact, the underlying concepts and outlines an strategy for its development. As illustration, a set of statements (definitions, axioms, theorems) is given that leads to the principle of software uncertainty and to software evolution management guidelines. 1. Software Evolution The phenomenon of software evolution, the consequence of an intrinsic need for continuing maintenance and further development of software embedded in real world domains, was first identified and studied in the early 70s [leh69,bel72]. Until recently, however, it did not arouse significant interest amongst academics or practitioners in the computer science, software engineering or software process communities. This situation has now changed. In recent years the interest in and study of the evolution phenomenon has spread as demonstrated by the recurrent appearance of the theme in events [e.g., pse98] and publications [e.g., pfl98,ben00]. Formulation of the FEAST hypothesis [leh94], three International FEAST Workshops [fea94,5], the results of the EPSRC funded FEAST/1 [leh96] and FEAST/2 [leh98] projects [feast], a recent (fourth) International FEAST 2000 Workshop with some 40 international participants and the publications that resulted [fea00a] have contributed to this development. Growing awareness is, in part, also due to the proliferation of desktop and personal computers, their increasing use by individuals, in industry, commerce, government and so on, and increasing use of the Internet. As users become ever more sophisticated and dependent on satisfactory system operation, the need for speedy, reliable, cost-effective evolution of their software through, for example, the provision of management rules, tools and guidelines [leh00a] has become increasingly apparent. It is forcing the continuing software evolution through upgrading or replacement that has become everyday experience for all serious computer users. Equally, increasing interest in business process improvement, resulting from the burgeoning integration of business processes and their internationalisation, has resulted in widespread recognition of the growing organisational dependence on software and, with it, the realisation of a need for business and software co-evolution [e.g., sebpc,soce]. In considering software evolution the FEAST group and a small number of other groups, e.g. [kem99], have viewed the word evolution as a noun [leh00c]. They are concerned with the properties of the evolution phenomenon, the what and the why. They seek to understand it by applying the scientific method of measuring, modelling, interpretation, hypothesis generation, as illustrated in figure 1. The primary goal has been to determine the underlying causes, attributes and practical impact of evolution on the software process and its products. Since the practical nature of the phenomenon as experienced in industry and by users was to be determined, examination of the evolution of a number of very different systems industry developed and supported systems was the first priority. As shown by its publications [feast], these studies have, and are still, yielded significant results including some that throw considerable light on the nature and attributes of software evolution. Figure 1 highlights the role of theory, the main theme of the present paper. Observation Measurement Patterns Models Interpretation Hypotheses