Abstract

We solve a longstanding problem by providing a denotational model for nondeterministic programs that identifies two programs iff they have the same range of possible behaviours. We discuss the difficulties with traditional approaches, where divergence is bottom or where a term denotes a function from a set of environments. We see that making forcing explicit, in the manner of game semantics, allows us to avoid these problems.We begin by modelling a first-order language with sequential I/O and unbounded nondeterminism. Then we extend the model to a calculus with higher-order and recursive types, by adapting standard game semantics. Traditional adequacy proofs using logical relations are not applicable, so we use instead a novel hiding and unhiding argument