: Bunse, C.; Atkinson, C.

Abstract
The value of graphical modeling within the analysis and design activities of object-oriented development is predicated on the assumption that the resulting models can be mapped correctly, optimally and efficiently into executable (normally textual) code. In practice, however, because of the large potential mismatch in abstraction levels, the mapping of graphical models into code is often one of the weakest and most error prone links in the chain of development steps. This paper describes a practical approach for addressing this problem based upon the definition of a restricted extension of the UML known as the Normal Object Form (NOF). The basic purpose of the NOF is to provide a set of UML modeling concepts which are "semantically close" to those found in object-oriented programming languages. Highly abstract UML models can then be mapped into corresponding executable code by means of a series of semantically small refinement (intra-UML) and translation (extra-UML) translation steps, rather than in one large (often ad hoc) step. This not only increases the chances of a correct and optimal mapping, but also significantly improves the traceability of UML constructs to and from code constructs, with all the associated advantages for maintenance and reuse.