Algorithmic Game Semantics and its Applications: Final Report

ion refinement has proved to be a highly effective method to verify systems with very large state spaces. Since abstractions are conservative overapproximations, safety of any abstracted program implies the safety of the concrete program (though the converse is not true. In a SAS’05 paper [8], Ghica et al. have introduced a purely semantic approach to (data) abstraction refinement, based on game semantics, using a language called Abstracted Idealized Algol. The key feature of the language is the use of abstraction schemes at the level of data-types, which allows the writing of abstracted programs in a syntax similar to that of concrete programs. In fact, a concrete program is a particular abstracted program, in which all the abstractions are identities. A fully abstract game model for AIA is presented, and an abstraction refinement semi-algorithm based on concrete representations of strategies is proposed. Tool 1: Game semantics compiler for IA2+while To assess the practicability of the game-semantic approach to program verification, we have constructed a compiler that transforms an open procedural program into the finite-state machine representation of its fully abstract game semantics; very little user instrumentation of the source code is required. The tool was constructed in CAML; most of the back-end heavy duty finite-state machine processing was done using the AT+T FSM library. Ghica and Murawski’s experiments (in TACAS’04 [2]) confirm what is a common situation in software model checking: even though the asymptotic complexity of the algorithm is high, the worst-case scenario only happens in pathological cases; many common and useful programs can in fact be verified. A number of case studies (bubble sort,