Ian Price <ianpric...@googlemail.com> writes:
> The problem is one that occurs when hygienic and non-hygienic macros are
> mixed. Here, and-let* is the non-hygienic one. Basically, in a hygienic
> macro, you carry around a bunch of context to allow you to refer to the
> right variable names. Then defmacro comes along, strips all that away,
> and uses the raw symbol names.
>
> We can fix and-let* to be hygienic, that's pretty easy, but I'm not sure
> what you can do about this in general. It's not like you can pass in the
> gensymed name, because that will break the way people who for some
> reason still right defmacros expect them to work.
>
> Dunno, maybe I'm just missing some insight here.
This discussion came up on #scheme yesterday, and Eli Barzilay mentioned
that Racket tries to be more clever with its legacy defmacro by keeping
a hash associating input sexps with their syntax objects.
This is in no way a silver bullet, but for many macros, this is liable
to work out really nicely. An example implementation is below, though I
have a few notes indicating things needing fixed.
(define-syntax define-macro
(lambda (x)
"Define a defmacro."
(syntax-case x ()
((_ (macro . args) doc body1 body ...)
(string? (syntax->datum #'doc))
#'(define-macro macro doc (lambda args body1 body ...)))
((_ (macro . args) body ...)
#'(define-macro macro #f (lambda args body ...)))
((_ macro transformer)
#'(define-macro macro #f transformer))
((_ macro doc transformer)
(or (string? (syntax->datum #'doc))
(not (syntax->datum #'doc)))
#'(define-syntax macro
(lambda (y)
(define (recontextualize form context default)
(define (walk x)
;; is there any possibility of a circular syntax object?
(cond ((hashv-ref context x) => (lambda (x) x))
((pair? x)
(cons (walk (car x))
(walk (cdr x))))
((vector? x)
(vector-map walk x))
((symbol? x)
(datum->syntax default x))
(else x)))
(walk form))
(define (build-context form stx-form)
(define ctx (make-hash-table))
(define (walk x y)
(hashv-set! ctx x y)
;; is there any possibility of a circular syntax object?
(cond ((pair? x)
(walk (car x) (car (syntax-e y)))
(walk (cdr x) (cdr (syntax-e y))))
((vector? x)
(vector-for-each2 walk x (syntax-e y)))
;; Any other types needing handled?
))
(walk form stx-form)
ctx)
(define (vector-for-each2 f v1 v2)
(define len (vector-length v1))
(define v* (make-vector len))
(let loop ((i 0))
(unless (= i len)
(vector-set! v* i (f (vector-ref v1 i) (vector-ref v2 i)))
(loop (+ i 1))))
v*)
(define (vector-map f v)
(define len (vector-length v))
(define v* (make-vector len))
(let loop ((i 0))
(unless (= i len)
(vector-set! v* i (f (vector-ref v i)))
(loop (+ i 1))))
v*)
(define (syntax-e obj)
(syntax-case obj ()
[(first . rest)
(cons #'first #'rest)]
[#(value (... ...))
(apply vector #'(value (... ...)))]
[a (syntax->datum #'a)]))
doc ;; FIXME: may not be a docstring, and so would fail above
#((macro-type . defmacro)
(defmacro-args args))
(syntax-case y ()
((_ . args)
(let* ((v (syntax->datum #'args))
(ctx (build-context v #'args)))
(recontextualize (apply transformer v) ctx y))))))))))
This version of define-macro still fails on the original macros as
posted by Panicz Maciej Godek, but gives the "right" result using stis's
ck macro version.
At 2:30am, I'm not liable to get to the bottom of why till tomorrow, but
I think doing something like this is a positive step.
--
Ian Price -- shift-reset.com
"Programming is like pinball. The reward for doing it well is
the opportunity to do it again" - from "The Wizardy Compiled"
