6. The Use of Macros
An important characteristic of the SCHEME language is that its set of primitive constructs is quite small. This set is not always convenient for expressing programs, however, and so a macro facility is provided for extending the expressive power of the language. A macro is best thought of as a syntax rewrite rule. As a simple example, suppose we have a primitive GCD which takes only two arguments, and we wish to be able to write an invocation of a GCD function with any number of arguments. We might then define (in a "production- rule" style) the conditional rule:
(XGCD) => 0 (XGCD x) => x (XGCD x . rest) => (GCD x (XGCD . rest))
(Notice the use of LISP dots to refer to the rest of a list.) This is not considered to be a definition of a function XGCD, but a purely syntactic transformation. In principle all such transformations could be performed before executing the program. In fact, RABBIT does exactly this, although the SCHEME interpreter naturally does it incrementally, as each macro call is encountered.
Rather than use a separate production-rule/pattern-matching language, in practice SCHEME macros are defined as transformation functions from macro-call expressions to resulting S-expressions, just as they are in MacLISP. (Here, however, we shall continue to use production rules for purposes of exposition.) It is important to note that macros need not be written in the language for which they express rewrite rules; rather, they should be considered an adjunct to the interpreter, and written in the same language as the interpreter (or the compiler). To see this more clearly, consider a version of SCHEME which does not have S-expressions in its data domain. If programs in this language are
