Transactional events for ML

  • Authors:

  • Laura Effinger-Dean;Matthew Kehrt;Dan Grossman

  • Affiliations:

  • University of Washington, Seattle, WA, USA;University of Washington, Seattle, WA, USA;University of Washington, Seattle, WA, USA

  • Venue:
  • Proceedings of the 13th ACM SIGPLAN international conference on Functional programming
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Transactional events (TE) are an approach to concurrent programming that enriches the first-class synchronous message-passing of Concurrent ML (CML) with a combinator that allows multiple messages to be passed as part of one all-or-nothing synchronization. Donnelly and Fluet (2006) designed and implemented TE as a Haskell library and demonstrated that it enables elegant solutions to programming patterns that are awkward or impossible in CML. However, both the definition and the implementation of TE relied fundamentally on the code in a synchronization not using mutable memory, an unreasonable assumption for mostly functional languages like ML where functional interfaces may have impure implementations. We present a definition and implementation of TE that supports ML-style references and nested synchronizations, both of which were previously unnecessary due to Haskell's more restrictive type system. As in prior work, we have a high-level semantics that makes nondeterministic choices such that synchronizations succeed whenever possible and a low-level semantics that uses search to implement the high-level semantics soundly and completely. The key design trade-off in the semantics is to allow updates to mutable memory without requiring the implementation to consider all possible thread interleavings. Our solution uses first-class heaps and allows interleavings only when a message is sent or received. We have used Coq to prove the high- and low-level semantics equivalent. We have implemented our approach by modifying the Objective Caml run-time system. By modifying the run-time system, rather than relying solely on a library, we can eliminate the potential for nonterminating computations within unsuccessful synchronizations to run forever.