[racket] Box: When and why?

On Oct 21, 2010, at 5:48 PM, Greg Hendershott wrote:

> Thank you, I understand now.
> 
> It's interesting how little I've need to consider, so far, whether
> Racket parameters are call-by-value or call-by-reference.
> 
> If you don't use set! on parameters, one reason to consider it
> disappears. I've been sticking to a functional style (after an initial
> awkward "imperative un-training" period).

Racket is call-by-value. Period. Always. -- Matthias


> Also whenever I've needed to return > 1 value, it's been very natural
> to use values (for 2 or 3) or a struct (for more).


p.s. Amen. 



> 
> On Thu, Oct 21, 2010 at 11:41 AM, Matthias Felleisen
> <matthias at ccs.neu.edu> wrote:
>> 
>> Boxes are the closest thing to addresses you can find in Racket.
>> You need them when you want to mimic a classical by-reference
>> function parameter protocol. For example,
>> 
>>  (define (return-nothing x y z a-box)
>>   (set-box! a-box ComputeSomethingWith-x-y-z)
>>   (void) ;; just to make clear that nothing else is returned
>>   )
>> 
>> You'd call this elsewhere with
>> 
>>  (define my-personal-box (box #f))
>>  ...
>>  (return-nothing my-personal-box)
>>  (unbox my-personal-box)
>> 
>> You can imagine that you have some other ways of referring to
>> my-personal-box (say in a list of all your boxes) which you may
>> then use to implement other services (saving all boxes to a file).
>> 
>> Some of our old methods that go down to C++ use this protocol.
>> 
>> In your 'p.s.' example, there is no reason whatsoever to
>> use a box over set! and variables.
>> 
>> -- Matthias
>> 
>> 
>> 
>> 
>> 
>> On Oct 21, 2010, at 11:31 AM, Greg Hendershott wrote:
>> 
>>> P.S. Maybe a simpler example I should have used:
>>> 
>>> (define next-web-parameter-id
>>>  (let ([i (box 0)])
>>>    (lambda ()
>>>      (begin0 (unbox i)
>>>              (set-box! i (add1 (unbox i)))))))
>>> 
>>> In both examples the box is involved in a closure lambda pattern.
>>> 
>>> This example, my old C/C++ brain wants to interpret it as a kind of
>>> thread-safe increment of a variable. But what's really going on?
>>> 
>>> On Thu, Oct 21, 2010 at 11:22 AM, Greg Hendershott
>>> <greghendershott at gmail.com> wrote:
>>>> Where can I find more information about the use scenarios for "box"?
>>>> 
>>>> The Guide is terse:
>>>> 
>>>>>> 
>>>> 3.11 Boxes
>>>> 
>>>> A box is like a single-element vector. It can print as a quoted #&
>>>> followed by the printed form of the boxed value. A #& form can also be
>>>> used as an expression, but since the resulting box is constant, it has
>>>> practically no use.
>>>> <<
>>>> 
>>>> This explains what a box is, but not why or when you would want to use one.
>>>> 
>>>> I see box used as in this example from servlet-dispatch.rkt:
>>>> 
>>>> (define (dispatch/servlet
>>>> ...
>>>>  (define servlet-box (box #f))
>>>> ...
>>>>  (filter:make
>>>> ...
>>>>    (lambda (url)
>>>>      (or (unbox servlet-box)
>>>>          (let ([servlet
>>>>                 (parameterize ([current-custodian (make-custodian)]
>>>>                                [current-namespace
>>>>                                 (make-servlet-namespace
>>>>                                  #:additional-specs
>>>>                                  default-module-specs)])
>>>>                   (if stateless?
>>>>                       (make-stateless.servlet
>>>> servlet-current-directory stuffer manager start)
>>>>                       (make-v2.servlet servlet-current-directory
>>>> manager start)))])
>>>>            (set-box! servlet-box servlet)
>>>>            servlet))))))
>>>> 
>>>> And I'm scratching my head, not understanding the point of using a box
>>>> as opposed to a simple non-boxed value.
>>>> 
>>>> My question isn't about this code per se; only an example. Generally,
>>>> in what situations would you use a box, and why?
>>>> 
>>>> Thank you.
>>>> 
>>> _________________________________________________
>>>  For list-related administrative tasks:
>>>  http://lists.racket-lang.org/listinfo/users
>> 
>>