[racket] module+ test... (require rackunit) results in unbound identifier when used in macro

On Wed, Jan 8, 2014 at 6:44 PM, Scott Klarenbach <scott at pointyhat.ca> wrote:

> Thanks guys.
>
> I changed it to this:
>
> (define-syntax (define/test stx)
>   (syntax-parse stx
> [(_ (proc:id arg:expr ...)
>  body:expr ...
> #:test test-body:expr ...)
>  (with-syntax ([id-str (symbol->string (syntax->datum #'proc))]
>  [ru-require (format-id stx "~a" "rackunit")])
>    #'(begin
>    (define (proc arg ...) body ...)
>    (module+ test
>  (require ru-require)
>  (test-case id-str test-body ...))))]))
>
> and it works now.  Is that what you meant Carl?
>

Yep, that's it.


>  The problem is that check-true (and check-false) are not in scope (as
>> you discovered) during expansion.
>
>
> I'm still a bit confused as to why.  I figured the (require rackunit)
> would have brought check-true into scope during expansion, much the same as
> if I just copied the (begin... form into another file without the macro.
>

The (require rackunit) does bring the names into scope... but _which_
scope?  That's always the important question with macros.  By default, a
binding form brings names into the scope in which the binding occurrence of
the names are written.  That's why if you write a local binding inside a
macro definition, only the code inside the macro can see it, while if you
bind a name that the user passed in, the user can see it.

With require, neither party writes the name; the name shows up in the
module definition.  Therefore, the require form has to figure out what
scope to bind the names in some other way.  So, for module names at least,
require uses the scope in which the module name is written to bind all of
the definitions provided by the module.  (Things get more interesting if
you use prefix-in or other import forms; they each have their own rules.)

In your example, what scope was the module name rackunit written in?
Inside the macro definition.  That's why writing #'(let ([check-true
<stuff>]) body ...)  and#'(begin (require rackunit) body ...) both make
check-true visible only inside the macro definition.  To make check-true
visible to the macro user, you have to change the context of the name
check-true.  In the first case, that means changing the context of the
identifier check-true; in the second, it means changing the context of the
module name rackunit.

So when things go wrong with macro bindings, the first question is always
"_which_ scope is this bound in?" and usually the second is "well, what
context does the binding occurrence have?".

I hope that helps, these are really murky waters, so if you're confused,
you're definitely not alone.  We're happy to ask questions frequently and
in detail.

Good luck!


>  On Wed, Jan 8, 2014 at 2:37 PM, Carl Eastlund <carl.eastlund at gmail.com>wrote:
>
>> I do not suggest rebinding the whole body like that; it could be
>> seriously problematic if the user refers to something not viable to the
>> macro definition.  What you need to rebind is the name "rackunit" itself
>> passed to require so that it has the context of the macro application.
>> That way the names bound by the require will be visible to the macro user.
>>  On Jan 8, 2014 5:29 PM, "Stephen Chang" <stchang at ccs.neu.edu> wrote:
>>
>>> The problem is that check-true (and check-false) are not in scope (as
>>> you discovered) during expansion. Here is one way to fix your problem
>>> by recapturing the test expressions to have the proper context. You
>>> can also use syntax-local-introduce instead of the lambda (someone
>>> correct me if this is not proper usage).
>>>
>>> #lang racket
>>> (require (for-syntax syntax/parse))
>>>
>>> (define-syntax (define/test stx)
>>>   (syntax-parse stx
>>>     [(_ (id arg ...) body ... #:test test-body ...)
>>>      (with-syntax ([id-str (symbol->string (syntax->datum #'id))]
>>>                    [(new-test-body ...)
>>>                     (map
>>>                      (λ (s) (datum->syntax #'here (syntax->datum s)))
>>>                      (syntax->list #'(test-body ...)))])
>>>        #'(begin
>>>            (define (id arg ...) body ...)
>>>            (print id-str)
>>>            (module+ test
>>>              (require rackunit)
>>>              (test-case id-str
>>>                         new-test-body ...))))]))
>>>
>>> (define/test (my-fn a b c)
>>>   (print a)
>>>   (print b)
>>>   (print c)
>>>   #:test
>>>   (check-true #t)
>>>   (check-false #t))
>>>
>>> On Wed, Jan 8, 2014 at 4:53 PM, Scott Klarenbach <scott at pointyhat.ca>
>>> wrote:
>>> > I have the following macro:
>>> >
>>> > (define-syntax (define/test stx)
>>> >   (syntax-parse stx
>>> > [(_ (id arg ...) body ... #:test test-body ...)
>>> > (with-syntax ([id-str (symbol->string (syntax->datum #'id))])
>>> >   #'(begin
>>> >   (define (id arg ...) body ...)
>>> >   (print id-str)
>>> >   (module+ test
>>> > (require rackunit)
>>> > (test-case id-str
>>> >   test-body ...))))]))
>>> >
>>> > Which handles some testing boilerplate and allows me to use it like so:
>>> >
>>> > (define/test (my-fn a b c)
>>> >   (print a)
>>> >   (print b)
>>> >   (print c)
>>> >   #:test
>>> >   (check-true #t)
>>> >   (check-false #t))
>>> >
>>> > The problem is that the (require rackunit) expression inside of
>>> (module+
>>> > test) is not being picked up after macro expansion.  Running either
>>> the code
>>> > or the tests (via raco test) results in: unbound identifier check-true.
>>> >
>>> > Adding (module+ (require rackunit)) to the top of the file solves the
>>> > problem and both the file and tests run as expected.
>>> >
>>> > What am I missing?
>>> >
>>> > 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
>>> >
>>> > ____________________
>>> >   Racket Users list:
>>> >   http://lists.racket-lang.org/users
>>> >
>>>
>>> ____________________
>>>   Racket Users list:
>>>   http://lists.racket-lang.org/users
>>>
>>
>
>
> --
> 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/20140108/9ab2d761/attachment-0001.html>