This page has been validated.
Steele and Sussman
2
The Art of the Interpreter

LISP-like Languages

Of the hundreds or thousands of computer languages which have been invented, there is one particular family of languages whose common ancestor was the original LISP, developed by McCarthy and others in the late 1950's. [LISP History] These languages are generally characterized by a simple, fully parenthesized ("Cambridge Polish") syntax; the ability to manipulate general, linked-list data structures; a standard representation for programs of the language in terms of these structures; and an interactive programming system based on an interpreter for the standard representation. Examples of such languages are LISP 1.5 [LISP 1.5M], MacLISP [Moon], InterLISP [Teitelman], CONNIVER [McDermott and Sussman], QA4 [Rulifson], PLASMA [Smith and Hewitt] [Hewitt and Smith], and SCHEME [SCHEME] [Revised Report]. We will call this family the LISP-like languages.

The various members of this family differ in some interesting and often subtle ways. These differences have a profound impact on the styles of programming each may encourage or support. We will explore some of these differences by examining a series of small ("toy") evaluators which exhibit these differences without the clutter of "extra features" provided in real, production versions of LISP-like language systems.

The series of evaluators to be considered partially constitute a reconstruction of what we believe to be the paths along which the family evolved. These paths can be explained after the fact by viewing the historical changes to the language as being guided by the requirements of various aspects of modularity.

Structure of the Paper

Our discussion is divided into several parts, which form a linear progression. In addition, there are numerous large digressions which explore interesting side developments. These digressions are placed at the end as notes, cross-referenced to and from the text.

We exhibit a large number of LISP interpreters whose code differs from one to another in small ways (though their behavior differs greatly!). In order to avoid writing identical pieces of code over and over, each figure exhibits only routines which differ, and also contains cross-references to preceding figures from which missing routines for that figure are to be drawn.

Part Zero introduces the restricted dialect of the LISP language in which most of our examples are written. It also discusses the basic structure of an interpreter, and exhibits a meta-circular interpreter for the language.

Part One introduces procedural data as an abstraction mechanism, and considers its impact on variable scoping disciplines in the language. We are forced through a series of such disciplines as unexpected interactions are uncovered and fixed. Interpreters are exhibited for dynamic scoping and lexical scoping.

Part Two considers the problems associated with the decomposition