Stepwise Refinement and Problem Solving

S tepwise refinement remains a popular approach to program-design implementation, primarily because of its flexibility. As originally proposed by Niklaus Wirth, the technique involves successively decomposing design decisions into target code. With each re6nement step, the designer must + note the interdependence of subproblems, + attempt to preserve the design’s functional correctness, and + defer decisions about representation details as long as possible.’ A more specific goal, stated by Harlan Mills, is to “divide, connect, and check an intended function by reexpressing it as an equivalent structure of properly connected subfunctions, each solving part of the problem, a n each simpler than the d original to further divide, connect and check.... . ..each refinement is taken as a working hypotheses for further investigation to be judged either sound or amended as its implication becomes clear.“’ In this article, we are interested in exploring whether or not the traditional notion of stepwise refinement can be reexpressed as a combination of general problem-solving activities that are based on paradigms taken from artificial intelligence research. This reexpression can form the basis for a more explicit view of programming as a problem-solving activ-