ML Module Mania: A Type-Safe, Separately Compiled, Extensible Interpreter

ML provides unusually powerful mechanisms for building programs from reusable modules. Such power is not available in other popular languages, and programmers accustomed to those languages have wondered if a powerful modules system is really necessary. This paper explores the power of ML modules—including higher-order functors—via an extended programming example. The example solves a problem in the construction of interpreters: how to combine extensibility with separate compilation in a safe language. We focus on a kind of interpreter for which extensibility and separate compilation are especially important: the embedded interpreter. An embedded interpreter implements a reusable scripting language that can be used to control a complex application—like a web server or an optimizing compiler—which is written in a statically typed, compiled host language like ML. The interpreter becomes part of the application, so the application can invoke the interpreter and the interpreter can call code in the application. The idea was first demonstrated by Ousterhout (1990) and has been widely imitated (Benson 1994; Laumann and Bormann 1994; Ierusalimschy, de Figueiredo, and Celes 1996a; Jenness and Cozens 2002; van Rossum 2002). Sometimes a host language can also be used for scripting (Leijen and Meijer 2000), but often it is inconvenient or even impossible to make a host-language compiler available at run time. A scripting language and its interpreter must meet several requirements: