Historical Notes

This is my (Colin Smith’s) first real attempt at a project in Clojure, so I can’t say that everything I’ve done is idiomatic or the best possible. On the other hand, I’ve written a fair amount of Scheme, so it has been a lot of fun to consider how to best use Clojure’s strengths in implementing this. Of course, none of this project should be considered original work of my own; it is the brainchild of G.J. Sussman and his collaborators. But allow me take a moment to describe some of the pleasant surprises I experienced while doing this.

First of all, a rich set of persistent data structures makes all the difference. What few examples of mutability in Scmutils where present have all been removed. Many instances of using association lists and other O(n) data structures have been replaced by maps and sets, and the sorted variants of those.

Using defrecord and deftype simplified the handling of a lot of Scmutils objects that were simply the cons of a type tag and a value. Such types can implement IFn when they need to be callable, allowing me to completely dodge the apply-hook technique used in MIT/GNU Scheme, which is one of the main reasons the code would not have been easy to port to other variants of Scheme, let alone any other flavor of LISP.

The Scmutils code is essentially a monolith, given that Scheme has no module scoping. This Clojure implementation partitions the codebase into a variety of namespaces, with unit tests for each. This adds a bit of sanity to the code absent in the original, where it is not easy to see exactly how the user’s environment is constructed. Clojure’s namespace facility allows us to prepare the environment in which primitive operations like + and * are replaced with their namespace-qualified generic equivalents exactly at the point where this is necessary; but also allows the generic operations to be mixed with the native operations when this is useful.