Abstract

Software development usually involves a collection of properties, programs and data as input or output documents. Putting these three kinds of documents at the vertices of a triangle, one sees that all three sides of the triangle have been exploited in formal methods, and that they have often been used in both directions. However, richer combinations have seldom been envisaged, and formal methods often amount to a strict orientation of the figure by imposing functional dependencies (e.g., infering test cases from specifications). Moreover, undecidability problems arise when properties are expressed in full predicate logic (or similar formalisms) or programs are written in Turing-equivalent programming languages. We advocate that (1) formal methods should provide more flexible ways to exploit the developer's knowledge and o#er a variety of possibilities to construct programs, properties and test data and (2) it is worth restricting the power of logic formalisms and programming languages...

Citations