A methodolgy for programming with concurrency

A programming calculus is presented which will yield programs with simple, suitable, and safe concurrency. The program design consists of three steps: (i) specification of a finite problem by a pre/post condition pair (2) formal refinement of a totally correct solution which can be implemented sequentially (3) declaration of program properties which allow relaxations in sequencing (+ concurrency) For infinite problems programs can be executed repeatedly without correctness problems. For the choice of refinement formal guidelines can be used. The derived solutions have the following properties: (a) dynamic concurrency processes do not have to be statically declared (b) only correct concurrency can be specified exclusion is not explicitly programmed (c) conditional concurrency no conditional delays (d) stepwise proofs of parallel correctness without auxiliary variables (e) simply derived freedom from deadlock and starvation without appealing to a fair scheduler Address of authors: University of Toronto, Computer Systems Research Group, 121 St. Joseph's St., Toronto, Ontario, Canada, M5S IAI Research supported by the Social Sciences and Humanities Research Council of Canada