Bounded Space Programming using Finite State Machines and Recursive Functions: the Hume Approach

Software engineering involves matching abstract software requirements to concrete implementations. Programming at a high-level of abstraction improves confidence in the correctness of functional requirements and reduces the cost of constructing software, but loses confidence in the correctness of behavioural requirements. This is especially serious for resource-constrained systems, such as real-time embedded or control systems, where tight runtime bounds must be maintained on both space and time behaviour. This paper describes Hume: a novel programming language based on a combination of finite state machine and recursive programming constructs. Hume is unusual in being structured as a series of levels, each of which exposes different machine properties. The paper provides a series of determinate cost models for the different Hume levels, and shows how Hume programs can be refactored to lower levels. In this way, appropriate programming abstractions can be chosen, without compromising on program properties such as costability.