Bug in define/contract?

[Scott Moore]

Hi all,

The racket program below exhibits behavior that I think is a bug. When using define/contract (and I believe with-contract), contracts seem to get applied to values later than expected. In particular, in the following example I expect all uses of the variable “test” to see the wrapper added by the contract. Instead, applications see the projection but inspecting the value without invoking it does not.

Thanks,
Scott

#lang racket

(require rackunit)

(module test racket
(provide wrapped? (contract-out (test2 coerce-to-wrapped)) test)

(define-values (prop:wrapped wrapped? prop-wrapped-accessor) (make-impersonator-property 'wrapped))

(define (wrap value)
(let* ([wrapped (λ () (printf "Unwrapping ~a~n" value) value)])
(impersonate-procedure
wrapped
#f
prop:wrapped
#t)))

(define coerce-to-wrapped
(make-contract #:name "wrap"
#:projection
(λ (blame)
(λ (value)
(wrap value)))))

(define/contract (test)
coerce-to-wrapped
5)

(define (test2)
42))

(require 'test)

(check-true (wrapped? test)) ; Should return true!
(check-true (wrapped? test2)) ; Should return true!
(check-equal? ((test)) 5) ; test is in fact wrapped, must invoke twice to use the original procedure
(check-true (procedure? (test))) ; result of (test) is the unwrapped procedure

[Robby Findler]

The contract system (at a layer slight lower than define/contract or
provide/contract) call procedure-rename. And that appears not to
propogate the impersonator properties.

(wrapped? (procedure-rename (wrap (λ () 4)) 'hi)) ;; => #f

This PR is also relevant:

http://bugs.racket-lang.org/query/?debug=&database=default&cmd=view+audit-trail&cmd=view&pr=11222

> --
> You received this message because you are subscribed to the Google Groups "Racket Developers" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to racket-dev+...@googlegroups.com.
> To post to this group, send email to racke...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/racket-dev/452027A2-62FD-4D03-B700-89D04D3479A7%40fas.harvard.edu.
> For more options, visit https://groups.google.com/d/optout.

[Sam Tobin-Hochstadt]

A little further investigation -- the problem is in the C versions of
`impersonate-procedure` and `procedure-rename`, not the
keyword-handling ones.

#lang racket

(require (prefix-in k: '#%kernel))

(define-values (prop:wrapped wrapped? prop-wrapped-accessor)
(make-impersonator-property 'wrapped))

(define (wrap value)

(k:impersonate-procedure value #f prop:wrapped #t))

(wrapped? (k:procedure-rename (wrap +) 'x))

On Mon, Mar 9, 2015 at 5:14 PM, Robby Findler
<ro...@eecs.northwestern.edu> wrote:
> The contract system (at a layer slight lower than define/contract or
> provide/contract) call procedure-rename. And that appears not to
> propogate the impersonator properties.
>
> (wrapped? (procedure-rename (wrap (λ () 4)) 'hi)) ;; => #f
>
> This PR is also relevant:
>
> http://bugs.racket-lang.org/query/?debug=&database=default&cmd=view+audit-trail&cmd=view&pr=11222
>
> On Mon, Mar 9, 2015 at 3:26 PM, Scott Moore <sdm...@fas.harvard.edu> wrote:

>> --
>> You received this message because you are subscribed to the Google Groups "Racket Developers" group.
>> To unsubscribe from this group and stop receiving emails from it, send an email to racket-dev+...@googlegroups.com.
>> To post to this group, send email to racke...@googlegroups.com.
>> To view this discussion on the web visit https://groups.google.com/d/msgid/racket-dev/452027A2-62FD-4D03-B700-89D04D3479A7%40fas.harvard.edu.
>> For more options, visit https://groups.google.com/d/optout.
>
> --
> You received this message because you are subscribed to the Google Groups "Racket Developers" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to racket-dev+...@googlegroups.com.
> To post to this group, send email to racke...@googlegroups.com.

> To view this discussion on the web visit https://groups.google.com/d/msgid/racket-dev/CAL3TdON7pRgO%3Dc5AzSWV6AWb51BEb5CCv22JryhE2X%2Bd-Oh7ow%40mail.gmail.com.

[Robby Findler]

Okay, after a little prompting from Matthew I see that this is just a
plain old bug in the contract system. I've pushed a fix.

Thanks!

Robby

On Mon, Mar 9, 2015 at 3:26 PM, Scott Moore <sdm...@fas.harvard.edu> wrote: