Redefining syntax

Variable to syntax

This example determines what the behavior of the usual implementations is when an identifier is defined as a variable and then redefined as syntax. For example:

(define (noodle) (foodle))
(define (foodle) 23)
(noodle) => ??

(define-syntax foodle (syntax-rules ()
  ((foodle) 17)))
(noodle) => ??

(define (noodle) (foodle))
(noodle) => ??

Essentially all the implementations that support syntax-rules behave the same way on the first and third calls to noodle, returning 23 and 17 respectively. The exception is Owl Lisp, which has a hyperstatic REPL that disallows all forward references.

The first definition of noodle refers to foodle, which is undefined, and is therefore assumed to be a variable. (Guile prints a warning at this point.) The second definition of noodle refers to the current (syntax) definition of foodle. The question is, what happens in the second call of noodle, when the first definition of noodle is being invoked, but the definition of foodle has changed out from under it?

Noodle continues to call the old definition of foodle and returns 23: Racket, Gambit, Chicken, Bigloo, Scheme48/scsh, SISC, Chez, Vicare, Larceny, Mosh, IronScheme, STklos, KSi, SigScheme, SXM, Chibi

Complains that a non-procedure is being invoked: Gauche, MIT, Guile, Kawa, SCM, Foment, Scheme 9, Sagittarius

Complains that foodle is not defined: Ypsilon

No support for syntax-rules: NexJ, JScheme, KSi, SigScheme, Shoe, TinyScheme, RScheme, S7, BDC, XLisp, Rep, Schemik, UMB, Elk, Llava, Sizzle, FemtoLisp, Dfsch, Inlab

Syntax to variable

Now we look at what happens when an identifier is defined as syntax and then changed to be a variable instead. Here's an example:

(define-syntax fard
  (syntax-rules ()
    ((fard a b) (- a b))))

(fard 1 2) => -1

(define (fard a b) (+ a b))

(fard 1 2) => ??

(apply fard (list 1 2)) => ??

fard is successfully redefined in both operator and operand positions: Racket, Gauche, MIT, Gambit, Scheme48/scsh, Guile, Kawa, SCM, Chez, Vicare, Larceny (prints a warning), Ypsilon, Mosh, KSi, Foment, Owl Lisp, Chibi, Sagittarius

fard is successfully redefined in operand position, but is still a syntax keyword in operator position: Chicken, Bigloo (prints a warning), STklos, Picrin

fard is successfully redefined in operator position, but is still a syntax keyword in operand position: Scheme 9

fard cannot be redefined: IronScheme, SXM

No support for syntax-rules: NexJ, JScheme, KSi, SigScheme, Shoe, TinyScheme, RScheme, S7, BDC, XLisp, Rep, Schemik, UMB, Elk, Llava, Sizzle, FemtoLisp, Dfsch, Inlab


Back to Scheme Surveys

Page source (GitHub)