Factual Knowledge For Developing Concurrent Programs

We propose a system for the derivation of algorithms which allows us to use "factual knowledge" for the development of concurrent programs. From pre. liminary program versions the system can derive new versions which have higher performances and can be evaluated by communicating agents in a parallel architecture. The knowledge about the facts or properties of the programs is also used for the improvement of the system itself. I THE STRUCTURE OF THE SYSTEE We present some preliminary ideas for designing an interactive system which can be used for algorithm derivation. The components of the system are best understood by relating them to the AurstallDarlington methodology [2]. In that approach the programmer is first asked to produce a correct version of the program, and thenhehas tocare about efficiency issues. Ze then improves that preliminary version by perfcrming "eureka steps" and applying correctness preserving transformation rules [2] (maybe with the help of a machine for rule application). :je generalizethose concepts and we suggest the structure of a system (depicted in figure 1) where: i) the mathematical descriptions of the problems generalize the first correct program versions, ii) the factual knowledge [l] g eneralizes the eureka steps, and iii) the Logical System generalizesthemachine for the application of the transformation rules. For point i) we assume that the descriptions of the problems are constructive, that is, they correspond to executa.ble functional programs. We also assume that we nay have some constraints on their executions as, for instance, on the number of computing agents and their topological connections, on the space and time resources, etc. For point ii) we consider that during the development process the programmer acquires (maybe in an incremental way) the knowledge of some facts about the functions to be computed or the behaviour of the computing agents. Those new facts may or may not be logical consequences of the knowledge already available from the descriptions of the problems themselves. The Logical System of point iii) is more powerful than the traditional matching procedure, which applies the transformation rules and verifies the related validating conditions [3]. It is basically made out of three modules: a Knowledge Base in which new facts are incrementally added by the pro-grammer or the system itself, an Analyzer-SyntheAndrzej Skowron institute of Mathematics University of North Warsaw University Carolina at Charlotte PKiN IX p.907 Computer Science Department 00-901 Warsaw (Poland) Charlotte, NC 28223 (USA) Mathematical Descriptions of the Problems = Constructive Functions + Computational Constraints I . TFE LOGICAL SYSTmJ --I I Factual Knowledge on the Functions and the Computing Agents Efficient Concurrent Programswith Communicating