Formal object oriented development of software systems using LOTOS

Abstract

Formal methods are necessary in achieving correct software: that is, software that can be proven to fulfil its requirements. Formal specifications are unambiguous and analysable. Building a formal model improves understanding. The modelling of nondeterminism, and its subsequent removal in formal steps, allows design and implementation decisions to be made when most suitable. Formal models are amenable to mathematical manipulation and reasoning, and facilitate rigorous testing procedures. However, formal methods are not widely used in software development. In most cases, this is because they are not suitably supported with development tools. Further, many software developers do not recognise the need for rigour. Object oriented techniques are successful in the production of large, complex software systems. The methods are based on simple mathematical models of abstraction and classifi cation. Further, the object oriented approach offers a conceptual consistency across all stages of software development. However, the inherent flexibility of object oriented approaches can lead to an incremental and interactive style of development, a consequence of which may be insuffi cient rigour. This lack of rigour is exacerbated by the inconsistent and informal semantics for object oriented concepts at all stages of development. Formal and object oriented methods are complementary in software development: object oriented methods can be used to manage the construction of formal models and formality can add rigour to object oriented software development. This thesis shows how formal object oriented development can proceed from analysis and requirements capture to design and implementation. A formal object oriented analysis language is defined in terms of a state transition system semantics. This language is said to be customer-oriented: a number of graphical views of object oriented relations in the formal analysis models are presented, and the specifi cations produced say what is required rather than how the requirements are to be met. A translation to ACT ONE provides an executable model for customer validation. This translation is founded on a precise statement of the relationship between classes and types (and subclassing and subtypes). The structure of the resulting ACT ONE requirements model corresponds to the structure of the problem domain, as communicated by the customer. The step from analysis to design requires an extension to the requirements model to incorporate semantics for object communication. A process algebra provides a suitable formal model for the specifi cation of communication properties. LOTOS, which combines ACT ONE and a process algebra in one coherent semantic model, provides a means of constructing object oriented design semantics. Design is de fined as the process of transforming a customer-oriented requirements model to an implementation-oriented design, whilst maintaining correctness. Correctness preserving transformations (CPTs) are defined for: transferring requirements structure to design structure, manipulating design structure and changing internal communication models. Design must be targetted towards a particular implementation environment. The thesis examines a number of different environments for the implementation of object oriented LOTOS designs. It illustrates the importance of understanding programming language semantics. We show how Eiffel can be used to implement formal object oriented designs. A case study which evaluates the formal object oriented models and methods, developed in this thesis, is reported. This identifi es re-use at all stages of software development and emphasises the role of structure: it improves understanding and communication, and makes validation and veri fication easier and better. The thesis shows that formal object oriented technology is ready for transfer to industry. These methods should be exploited sooner rather than later: object oriented development can incorporate formal methods without signi ficant cost, and formal methods can utilise the object oriented paradigm to manage complexity. The thesis provides a rationale for formal object oriented development and a set of conceptual tools which makes the development of software systems a true engineering discipline.