Uniqueness typing for resource management in message-passing concurrency

Abstract

We view channels as the main form of resources in a message-passing programming paradigm. These channels need to be carefully managed in settings where resources are scarce. To study this problem, we extend the pi-calculus with primitives for channel allocation and deallocation and allow channels to be reused to communicate values of different types. Inevitably, the added expressiveness increases the possibilities for runtime errors. We define a substructural type system, which combines uniqueness typing and affine typing to reject these ill-behaved programs.

Compositional Reasoning for Explicit Resource Management in Channel-Based Concurrency

“…. .. Resource-centric type disciplines such as [12,40] preclude name matching for a particular resource once all the permissions to use that resource have been used up; this feature is essential to statically reason about a number of basic design patterns for reuse. For such type settings, it turns out that the client implementations C 2 and C ′ 2 exhibit the same behaviour because the return channel used by both clients for both server interactions is private, i.e., unknown to the respective servers; as a result, the servers cannot answer the service on that channel before it is receives it on either srv 1 or srv 2 .…”

“…In [12] we developed a substructural type system based around a type attribute describing channel uniqueness, and this was used to statically ensure safe computations for Rπ. In this work, we weave this type information into our framework, imbuing it with an operational permissionsemantics to reason compositionally about the costed behaviour of (safe) processes.…”

“…The key novel structural rule is however tRev, which allows us to change (revise) the object type of a channel whenever we are guaranteed that the type assumption for that identifier is unique. These rules are recalled from [12] and the reader is encouraged to consult that document for more details.…”

“…In [12] we prove the usual subject reduction and progress lemmas for this type system, given an (obvious) error relation. Example 2.3.…”

“…We thus choose some fresh channel d, i.e., d M ∪ (Mσ Γ ) ∪ dom(Γ) 12 , and define σ ′ Γ as σ Γ , except that it maps d to c and also maps σ −1 Γ (c) (i.e., the channel name that mapped to c in σ Γ ) to σ Γ (d), since this channel is not mapped to by d anymore (in order to preserve bijectivity):…”