CML

Another approach to adding threads to SML is due to Reppy [19] in which a running program consists of threads which communicate and synchronise by passing synchronous messages on typed channels.

• spawn function takes a function as an argument and dynamically creates a new thread that evaluates the application of the argument function.
• channel function dynamically creates a (weakly) polymorphic channel. SML's Lexical scoping is used to share channels between threads, and to hide channels from other threads.
• send and accept operations deal with synchronous communications. They complement each other. To communicate to a channel a thread must rendezvous with another thread that wants to do complementary communication on the same channel.
• CSP style rendezvous mechanism with selective communication is implemented using events . According to Reppy, this achieves a level of abstraction that ensures security and ease of maintainance. events are a type of values that represent synchronous operations. base_events represent communication operations. receive and transmit functions build on these base_events . Combinators for associating actions to events and to build nondeterministic choices of events are provided. wrap combinator associates the action to an event. choose combinator builds the non-deterministic choice. A sync operator is provided to force synchronisation on an event. This formalism allows a programmer to create first class synchronisation and communication abstractions. Other synchronous operations like thread termination, low-level I/O support and time-outs can be expressed in terms of events . They are wait for termination, syncOnOutput and syncOnInput for I/O and waitUntil and timeOut for time-outs. CML has a uniform mechanism for combining synchronous operations. It provides guard and wrapAbort combinators to deal with commit points in communications. Concurrent versions of I/O primitives are provided.
• CML has been used in interactive systems to build a multi-threaded interface to the X protocol, called eXene , providing similar functions to the Xlib . Another area of application of CML is distributed systems programming. Especially interesting is the use of CML in Nuprl (an interactive proof system for ML) to distribute the parts of a proof among the pool of proof servers across a network of workstations.
• Implementation languages used is SML and its extensions (first-class continuations (for threads) and asynchronous signals (for preemptive scheduling)). Heap based implementations are particularly suited for light-weight threads (though stack-based ones are also possible).
• A CSP based formal semantics for CML has been given [20]. Some recent work in this area are reported in [6] and [3].