[racket] selectively local-expand sub-expressions

From: Scott Klarenbach (scott at pointyhat.ca)
Date: Sun Jan 26 13:01:19 EST 2014

Thanks a lot Carl...this is very enlightening.  If I could impose one last
question before I go off and digest everything:

What is the "correct" approach to capturing the runtime values of any
references that may be bound by the enclosing environment, for splicing
into the final recursively-expanded expression?

;; example 1

(define y 3)
(define x 3)
(define-dsl-syntax (pred? x) (or (= 2 x) (> x y)))
(define-dsl-syntax (bad-pred? x) (or (= 2 x) (> x z)))

(recursive-expand (pred? 1))
'(or (= 2 1) (> 1 3))

(recursive-expand (bad-pred? 1))
>> unbound identifier: z

My naive approach was to collect all the identifiers in the expression body
that had bindings, compare them to the argument list of the macro with
bound-identifier=? to see which ones were explicitly introduced by the
user, and then eval the remaining ones at runtime in a second step in order
to splice them in.

I haven't tried this, and am sure people are cringing just by reading it,
lol.  I know there are tons of features like marking and syntax properties
and origins, etc which I don't yet understand, and which may provide a more
durable solution.

If needbe, I could explicitly provide to the macro the bindings I wish to
capture, like postgresql does with query params...ie,

(define-dsl-syntax (pred? x) (or (= 2 x) (> x $1)) #:capture (y)) ;; or
something

but for obvious reasons it is much better if these expressions just
expanded and automatically captured any referenced values in the same way
as would happen at runtime.

Thanks a lot.

Scott.




On Sat, Jan 25, 2014 at 6:00 PM, Carl Eastlund <carl.eastlund at gmail.com>wrote:

> Scott,
>
> I see what you're doing now.  You're not actually trying to use macro
> expansion at all; you're just using local-expand to substitute the
> definition of pred? where it occurs, so that you can make its macro
> definition also serve as its DSL definition.  That's sensible, but
> local-expand is still doing more than you want it to.  That's why I put in
> all the expansion caveats -- not because you necessarily meant to do full
> expansion, but because local-expand is pretty explicitly built for full
> expansion, and always tries to push as far as it can.  Any time the caveats
> about expansion don't apply, local-expand is probably a bigger gun than you
> need.
>
> Where local-expand is going to bite you is when the definition of pred?
> uses a macro at its top level.  For instance:
>
>   (define-syntax-rule (pred? x) (or (< x 3) (> x 7)))
>
> Here, local-expand is going to expand the use of (or ...), and any macro
> that (or ...) produces at its top level, until you reach a core form as the
> main expression, or something you've put in an explicit stop list.  That's
> not what you want, as I understand it -- you only want to expand pred?.
>
> So what to do when you want to apply one macro, but not perform general
> expansion?  Extract its transformer using syntax-local-value, and apply it
> to the expression.  You probably also want to apply a syntax mark before
> and after transformation, just to simulate the base level of hygiene the
> macro may be relying on.  It might not be necessary for simple definitions,
> but it can't hurt.
>
> I wrote up some code that does this, along with a test showing that it
> won't expand "or" too far.  It's also reasonably hygienic -- it won't be
> confused if someone defines a different macro named "pred?", for example.
> I don't know if that's a concern, but again, it can't hurt.  Anyway, you
> can find what I wrote here: https://gist.github.com/carl-eastlund/8626893
>
> Carl Eastlund
>
> On Fri, Jan 24, 2014 at 1:30 PM, Scott Klarenbach <scott at pointyhat.ca>wrote:
>
>> Just an update, I was able to make this work.
>>
>> #lang racket
>> (require (for-syntax racket/syntax syntax/stx))
>>
>> (define-syntax-rule (pred? x) (> 3 x))
>>
>> (define-for-syntax (recursive-expand stx)
>>   (let loop ([l (syntax->list stx)])
>> (cond [(stx-null? l) l]
>>   [(stx-pair? (stx-car l))
>>    (cons (loop (stx-car l)) (loop (stx-cdr l)))]
>>   [(equal? 'pred? (syntax->datum (stx-car l)))
>>    (local-expand (cons (stx-car l) (loop (stx-cdr l))) 'expression #f)]
>> ;; this works
>>   [else
>>    (cons (stx-car l) (loop (stx-cdr l)))])))
>>
>> (define-syntax (test stx)
>>   (syntax-case stx ()
>> [(_ x)
>>  (with-syntax ([expanded (recursive-expand #'x)])
>>    #''expanded)]))
>>
>> (module+ test
>>   (require rackunit)
>>   (check-equal? (test (or (< 10 x) (pred? y)))
>> '(or (< 10 x) (> 3 y))))
>>
>> The code I couldn't figure out last night was:
>> (local-expand (cons (stx-car l) (loop (stx-cdr l))) 'expression #f)]
>>
>> Thanks.
>>
>> --
>> Talk to you soon,
>>
>> Scott Klarenbach
>>
>> PointyHat Software Corp.
>> www.pointyhat.ca
>> p 604-568-4280
>> e scott at pointyhat.ca
>> 200-1575 W. Georgia
>> Vancouver, BC V6G2V3
>>
>> _______________________________________
>> To iterate is human; to recur, divine
>>
>
>


-- 
Talk to you soon,

Scott Klarenbach

PointyHat Software Corp.
www.pointyhat.ca
p 604-568-4280
e scott at pointyhat.ca
200-1575 W. Georgia
Vancouver, BC V6G2V3

_______________________________________
To iterate is human; to recur, divine
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.racket-lang.org/users/archive/attachments/20140126/f777ab80/attachment.html>

Posted on the users mailing list.