Injecting variables into closures.

Injecting variables into closures.

[Kjetil S. Matheussen]

Hi,

I need (well) to inject variables into closures, and have
come up with the following macro to do so:

(define-macro (inject-variable name value)
   (define env (gensym))
   (define closure (gensym))
   (define vars (gensym))
   (define secondname (gensym))
   (define secondval (gensym))
   (define first-env (gensym))
   `((procedure->macro (lambda (,(gensym) ,env)
                         (cond ((null? (cdr ,env))
                                (let ((,closure (car ,env)))
                                  (set-car! ,env (cons ',name ,value))
                                  (set-cdr! ,env (list ,closure))))
                               ((null? (null? (cdr (car ,env))))
                                (let* ((,vars (car ,env))
                                       (,secondname (car vars))
                                       (,secondval (cdr vars)))
                                  (set-car! ,env
                                            (list (list ',name firstname)
                                                  ,value ,secondval))))
                               (else
                                (let ((,first-env (car ,env)))
                                  (set-cdr! ,first-env
                                            (cons ,value (cdr ,first-env)))
                                  (set-car! ,first-env
                                            (cons ',name (car ,first-env)))
                                  (set-car! ,env ,first-env))))))))


(let ()
   (display (defined? 'gakk))
   (inject-variable gakk 90)
   gakk)
=> #f90


Now I wonder, is this safe and portable? Or are there
other and better ways to do this?

Thanks for any answers.

Kjetil.



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Injecting variables into closures.

[Kjetil S. Matheussen]

On Thu, 29 Nov 2007, Kjetil S. Matheussen wrote:

>
> Hi,
>
> I need (well) to inject variables into closures, and have
> come up with the following macro to do so:
>
> (define-macro (inject-variable name value)
...
>                         (cond ((null? (cdr ,env))
...
>                               ((null? (null? (cdr (car ,env))))


#@^%$^&#!


(define-macro (inject-variable name value)
   (define env (gensym))
   (define closure (gensym))
   (define vars (gensym))
   (define secondname (gensym))
   (define secondval (gensym))
   (define first-env (gensym))
   `((procedure->macro
      (lambda (,(gensym) ,env)
        (define ,first-env (car ,env))
        (cond ((null? (cdr ,env))
 	      (let ((,closure ,first-env))
 		(set-car! ,env (cons ',name ,value))
 		(set-cdr! ,env (list ,closure))))
 	     ((not (pair? (cdr ,first-env)))
 	      (let* ((,vars ,first-env)
 		     (,secondname (car ,vars))
 		     (,secondval (cdr ,vars)))
 		(set-car! ,env
 			  (list (list ',name ,secondname)
 				,value ,secondval))))
 	     (else
 	      (set-cdr! ,first-env
 			(cons ,value (cdr ,first-env)))
 	      (set-car! ,first-env
 			(cons ',name (car ,first-env)))
 	      (set-car! ,env ,first-env)))))))


Sorry. :-)



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Injecting variables into closures.

[Kjetil S. Matheussen]

On Thu, 29 Nov 2007, Kjetil S. Matheussen wrote:

>
> (let ()
>   (display (defined? 'gakk))
>   (inject-variable gakk 90)
>   gakk)
> =>  #f90
>

By the way. I'm quite surprised the environments consist
of linked lists. Wouldn't it be much faster using hash
tables instead? Perhaps the variables are cached or something?




_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Injecting variables into closures.

[Ludovic Courtès]

Hi,

"Kjetil S. Matheussen" <k.s.matheussen@notam02.no> writes:

> By the way. I'm quite surprised the environments consist
> of linked lists. Wouldn't it be much faster using hash
> tables instead? Perhaps the variables are cached or something?

Yes, variable references are "cached" ("memoized") so that actual
variable lookup occurs only once.  So, for instance, in the following
piece of code:

  (let ((x 1)(y 2))
    (foo y))

the reference to "y" in "(foo y)" is replaced by an "iloc" that says
"this is a reference to the second variable in the first frame".  When
that iloc is encountered, the variable's value is fetched by doing
(roughly) two `list-ref's: one to get the frame and then one to get the
variable within the frame.

Clearly, a vector-like data structure (i.e., with O(1) access to
individual elements) would help here.

Thanks,
Ludovic.



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Injecting variables into closures.

[Kjetil S. Matheussen]

On Thu, 29 Nov 2007, Clinton Ebadi wrote:

> "Kjetil S. Matheussen" <k.s.matheussen@notam02.no> writes:
>
>> Hi,
>>
>> I need (well) to inject variables into closures, and have
>> come up with the following macro to do so:
>>
>> (define-macro (inject-variable name value)
>>   (define env (gensym))
>>   (define closure (gensym))
>>   (define vars (gensym))
>>   (define secondname (gensym))
>>   (define secondval (gensym))
>>   (define first-env (gensym))
>>   `((procedure->macro (lambda (,(gensym) ,env)
>>                         (cond ((null? (cdr ,env))
>>                                (let ((,closure (car ,env)))
>>                                  (set-car! ,env (cons ',name ,value))
>>                                  (set-cdr! ,env (list ,closure))))
>>                               ((null? (null? (cdr (car ,env))))
>>                                (let* ((,vars (car ,env))
>>                                       (,secondname (car vars))
>>                                       (,secondval (cdr vars)))
>>                                  (set-car! ,env
>>                                            (list (list ',name firstname)
>>                                                  ,value ,secondval))))
>>                               (else
>>                                (let ((,first-env (car ,env)))
>>                                  (set-cdr! ,first-env
>>                                            (cons ,value (cdr ,first-env)))
>>                                  (set-car! ,first-env
>>                                            (cons ',name (car ,first-env)))
>>                                  (set-car! ,env ,first-env))))))))
>>
>>
>> (let ()
>>   (display (defined? 'gakk))
>>   (inject-variable gakk 90)
>>   gakk)
>> => #f90
>>
>>
>> Now I wonder, is this safe and portable? Or are there
>> other and better ways to do this?
>
> Manipulating the environment is not going to be portable or safe. In

Well, it doesn't have to be. But I understand very well why
its not. :-)



> addition procedure->macro should be deprecated (but appears not to
> be?), and will be removed (necessarily) whenever someone finishes the
> phase separation in the evaluator.
>

Hope its not going to be. Its a convenient function.



> Is there a special reason to inject a value into the environment
> rather than let binding it?
>

I'm just playing with making a prettier module system:

(define-namespace bank)

(def-bank sum 0)
(def-bank (inc n)
   (inc! sum n))
(def-bank (dec n)
   (inc sum (- n)))

(bank/inc 50)
(bank/sum)
=> 50


I can also do it by doing a complete source transformations,
replacing sum with bank/sum everywhere and so on, but I would
rather avoid doing that.

Its just for fun.



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Injecting variables into closures.

[Ludovic Courtès]

Hi,

"Kjetil S. Matheussen" <k.s.matheussen@notam02.no> writes:

> Now I wonder, is this safe and portable? Or are there
> other and better ways to do this?

Neither `define-macro' nor `procedure->macro' is portable: both are
Guile-specific (SCM may have similar forms, though).

Another, simpler, Guile-specific way to do this would be:

  (module-define! (current-module) 'gakk 'some-value)

There's no R[56]RS-portable way to do this AFAIK.

Thanks,
Ludovic.



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: Injecting variables into closures.

[Kjetil S. Matheussen]

Ludovic Court?s:
>
> Hi,
>
> "Kjetil S. Matheussen" <k.s.matheussen@notam02.no> writes:
>
>> Now I wonder, is this safe and portable? Or are there
>> other and better ways to do this?
>
> Neither `define-macro' nor `procedure->macro' is portable: both are
> Guile-specific (SCM may have similar forms, though).
>

Thanks, but by "portable", I did (of course) mean portable
across various versions of Guile, not any of the scheme
standards...


> Another, simpler, Guile-specific way to do this would be:
>
>  (module-define! (current-module) 'gakk 'some-value)
>

Thanks, but that's probably not enough for what I
want to do. I need to simulate dynamic scoping
somehow, and I think its only possible to do that
by manipulating the environment. But I'm not
going to do that if its going to break with
forthcoming releases of Guile.



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user