NSF Award
9803880
9803880 Productivity benefits and quality improvement have come about as a result of designing and implementing software at ever higher levels of abstraction. So called "executable specifications"-high level functional descriptions of software-represent the current state of language development for software design. However, software is rarely implemented by directly translating such specifications, as the performance of the resulting product is usually poor. To remedy the performance problem without requiring human effort to reprogram a specification as an implementation, the use of program transformation has often been suggested, rewriting the specification according to rules that preserve its meaning but change its form. The technology of transformational improvement has made progress over a span of twenty years. Several functional language compilers now achieve substantial performance gains through the use of source-language transformations. Many improvements remain to be made, however. This research will build on recent and promising results that improve the performance of recursively defined functions by applying transformations based upon their underlying mathematics. Prototype systems have shown the potential of some of these methods, but these systems have never been fully implemented, integrated, or tested. In particular, this research will integrate a variety of known techniques that are expected to work more powerfully in concert than individually. In this project, a prototype transformation system will be extended so that it can be applied to programs written in a core subset of the programming language Standard ML, as well as in CAML, an ML dialect widely used in Europe. The research will help to determine whether high-level transformations can be effective enough in improving the performance of high-level specifications that software designers might quit worrying about many details of programming.***
