On Mon, May 2, 2016 at 10:44 AM, Alex Knauth <[email protected]> wrote:
>
>> On May 2, 2016, at 10:00 AM, Sam Tobin-Hochstadt <[email protected]> 
>> wrote:
>>
>> The problem with your first example is that there are some extra
>> scopes on `#'x` from the context that it's in. If you use
>> `(datum->syntax #f 'x)` (producing a syntax object with no scopes on
>> it) then your tests pass.
>
> Trying to understand this. Ok, so, there are scopes in the context of the 
> transformer function, which are preventing this example from working. So, why 
> doesn't syntax-local-introduce remove those scopes? If it's supposed to say 
> "pretend that this syntax was part of the input," and an input syntax object 
> wouldn't have those scopes, should it remove them? Or would that cause other 
> problems?
>
> The functions produced by `(make-syntax-introducer)` take an extra argument, 
> which can be 'flip, 'add, or 'remove. Right now the default for 
> `(make-syntax-introducer)` functions is 'flip.

My initial statement was less precise than my second email.
`(syntax-local-introduce stx)` adds a single scope, using the
equivalent of `'flip` mode. That scope is "the current scope" for the
macro expansion step. We could add an optional argument to
`syntax-local-introduce` to control the mode -- the current
implementation of `local_introduce` in env.c calls
scheme_stx_flip_scope, but changing that to use a mode would be
relatively simple, as C code goes. I'm not sure how useful it would
be, though.

Sam

>> What `syntax-local-introduce` does is add precisely the scope that the
>> expander added to indicate the input to the macro that you're
>> defining.
>
> If it adds that scope, but doesn't take away the other scopes that were 
> messing up this example, does that mean `syntax-local-introduce` is in 'add 
> mode, in terms of the modes of `(make-syntax-introducer)` functions? Would it 
> mean anything to make `syntax-local-introduce` take an optional extra 
> argument like this?
>
> Would this example work if this use of `syntax-local-introduce` was in 'flip 
> mode?
>
> Alex Knauth
>
>> On Mon, May 2, 2016 at 9:53 AM, Matthew Butterick <[email protected]> wrote:
>
>>> Sorry to be dense, but I still don't see the "pretend that this identifier 
>>> was in the input" angle. Below, example 1 fails with an unbound-identifier 
>>> error, because the `syntax-local-introduce` identifier does not capture the 
>>> use-site reference. Whereas example 2 works, even though the `invoke-x` 
>>> macro ostensibly uses a hygienic reference.
>>>
>>> Having looked at the scope sets, I now understand WHY the behavior below 
>>> happens. But `syntax-local-introduce` clearly does something quite 
>>> different from, say, `(datum->syntax caller-stx datum)` (whose behavior I 
>>> think of as "pretend that this identifier was in the input").
>
>>> #lang racket
>>> (require rackunit)
>>>
>>> (module macros racket
>>>  (provide (all-defined-out))
>>>  (define-syntax (sli-x stx)
>>>    (syntax-case stx ()
>>>      [(_) (with-syntax ([x (syntax-local-introduce #'x)])
>>>             #'(define x 'syntax-local-introduce-x))]))
>>>
>>>  (define-syntax-rule (invoke-x) x))
>>>
>>> (require 'macros)
>>>
>>> ;; example 1
>>> (check-equal?
>>> (let ()
>>>   (sli-x)
>>>   x) 'syntax-local-introduce-x)
>>>
>>> ;; example 2
>>> (check-equal?
>>> (let ()
>>>   (sli-x)
>>>   (invoke-x)) 'syntax-local-introduce-x)
>>>
>

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to