References/locations

References/locations

[Maciek Godek]

Hi,
it's me again, asking silly questions.
This time I would like to know if there's
a simple way to store a reference to
a variable in another variable -- say,
an element of vector or hash table.

I imagine this could look like this:
(define v #(1 2 3))
(define v1 (vector-location v 1))
v1
: 2
(set! v1 10)
v
: #(1 10 3)

I've tried to do it using a "procedure with
setter", but the problem is that set! doesn't
evaluate its first argument (as long as it's a
symbol), so I'd have to wrap everything
up in macros to obtain:
(set! (vector-location v 1) 10)

Besides I think that the names "hash-ref"
and "vector-ref" are confusing, since they
don't return references, but values (therefore
the names like "vector-get" or "hash-get" would
be more apropreate)

I also wonder if there's any point for allowing
locations to any sorts of variables (similar to
pointers in C or pointers to C++), that is,
(define x 10)
(define y (location x))
(set! y 20)
x
: 20

I expect that this idea isn't new in the world of
lisp, but I didn't find any discussion on the web.
There are certainly some issues related with locations:
- they require special behavior of set! and define:
(set! y 20) could work in two different ways,
depending on whether y is a location or not
- what about "locations of locations"?
- how would it affect performance?
- what about multiple threads?

Regards
M.

Re: References/locations

[Kjetil S. Matheussen]

"Maciek Godek":
> Subject: References/locations
> To: guile-user@gnu.org
> Message-ID:
> 	<e2ceda030808071850uc1640e6kf4ffc33699143fd2@mail.gmail.com>
> Content-Type: text/plain; charset=ISO-8859-1
>
> Hi,
> it's me again, asking silly questions.

Not at all, your questions are very good. :-)



> This time I would like to know if there's
> a simple way to store a reference to
> a variable in another variable -- say,
> an element of vector or hash table.
>
> I imagine this could look like this:
> (define v #(1 2 3))
> (define v1 (vector-location v 1))
> v1
> : 2
> (set! v1 10)
> v
> : #(1 10 3)
>
> I've tried to do it using a "procedure with
> setter", but the problem is that set! doesn't
> evaluate its first argument (as long as it's a
> symbol), so I'd have to wrap everything
> up in macros to obtain:
> (set! (vector-location v 1) 10)
>
> Besides I think that the names "hash-ref"
> and "vector-ref" are confusing, since they
> don't return references, but values (therefore
> the names like "vector-get" or "hash-get" would
> be more apropreate)
>

Never thought about that, but it sounds
correct. vector-get and hash-get would
probably be mroe apropreate names.



> I also wonder if there's any point for allowing
> locations to any sorts of variables (similar to
> pointers in C or pointers to C++), that is,
> (define x 10)
> (define y (location x))
> (set! y 20)
> x
> : 20
>

Well, here's a relatively clean way, I think:


(define-macro (location name)
    (define new-val (gensym))
    `(lambda (,new-val)
        (set! ,name ,new-val)))

(define old-set! set!)

(define-macro (set! a b)
   `(if (procedure? ,a)  ;; Needs a better check.
        (,a ,b)
        (old-set! ,a ,b)))


guile> (define x 10)
guile> (define y (location x))
guile> (set! y 20)
guile> x
20
guile>

Re: References/locations

[Stephen Compall]

"Maciek Godek" <pstrychuj@gmail.com> writes:
> I've tried to do it using a "procedure with
> setter", but the problem is that set! doesn't
> evaluate its first argument (as long as it's a
> symbol), so I'd have to wrap everything
> up in macros to obtain:
> (set! (vector-location v 1) 10)

Actually, if you (use-syntax (ice-9 syncase)), you should be able to
define lexical symbol-macros that expand a single symbol, even in a
set! place, something like:

(define v #(1 2 3))
(define-syntax v1
  (lambda (stx)
    (syntax-case stx ()
      (_ v))))

Obviously aliasing will serve:

(define-syntax alias-syntax
  (syntax-rules ()
    ((_ form)
     (lambda (stx)
       (syntax-case stx ()
         (_ (syntax form)))))))

(define-syntax defalias
  (syntax-rules stx ()
    ((_ new form)
     (define-syntax new (alias-syntax form)))))

(define-syntax let-alias
  (syntax-rules ()
    ((_ ((new form) ...) body ...)
     (let-syntax ((new (alias-syntax form)) ...) body ...))))

(defalias v1 (vector-ref v 1))

(set! v1 42)
(vector-ref v 1) ; => 42

-- 
a Mr. Fleming wishes to study bugs in smelly cheese; a Polish woman
wishes to sift through tons of Central African ore to find minute
quantities of a substance she says will glow in the dark; a Mr. Kepler
wants to hear the songs the planets sing.
	--Carl Sagan, "The Demon-Haunted World"

Re: References/locations

[Maciek Godek]

Kjetil:

>> Hi,
>> it's me again, asking silly questions.
>
> Not at all, your questions are very good. :-)

That's werid. Every time I write a post I've got the
strange feeling that I'm thinking in the way that
should be forbidden :)

I can give you the background of this idea.
I've been trying to implement a vector-based
object-oriented system (I'm rather reluctant towards
goops). I imagine that it could be used in the
following manner (borrowed from C++):

(define sphere
  (class
   () ;; public slots
   (x y radius) ;; private slots
   ((move (dx dy) ;; methods
	  (set! x (+ x dx))
	  (set! y (+ y dy)))
    (scale (factor)
	   (set! radius (* radius factor))))))

Obviously, the "class" macro should convert methods
to lambdas. For "move" that would be:

  (lambda(self dx dy)
    (let((x (vector-loc self (get-hash-table properties 'x)))
           (y (vector-loc self (get-hash-table properties 'y)))
           (radius (vector-ref self (get-hash-table properties 'radius))))
      (set! x (+ x dx))
      (set! y (+ y dy)))

Obviously, the hash table reference would have to be
evaluated during macro expand to provide the maximum
efficiency.
(If you have any other solutions to this problem, I'm open
to suggestions)

>> Besides I think that the names "hash-ref"
>> and "vector-ref" are confusing, since they
>> don't return references, but values (therefore
>> the names like "vector-get" or "hash-get" would
>> be more apropreate)
>>
>
> Never thought about that, but it sounds
> correct. vector-get and hash-get would
> probably be mroe apropreate names.

On the other hand the world of scheme seems
to have gotten used to such convention.

> Well, here's a relatively clean way, I think:
>
> (define-macro (location name)
>   (define new-val (gensym))
>   `(lambda (,new-val)
>       (set! ,name ,new-val)))
>
> (define old-set! set!)
>
> (define-macro (set! a b)
>  `(if (procedure? ,a)  ;; Needs a better check.
>       (,a ,b)
>       (old-set! ,a ,b)))
>
>
> guile> (define x 10)
> guile> (define y (location x))
> guile> (set! y 20)
> guile> x
> 20
> guile>

That's amazing! Whenever I think of something (and imagine
how difficult would it be to implement), you come up with a
solution taking only a few lines of code :)
(I only have to think how would it perform in practice)

Thanks
M.

Re: References/locations

[Maciek Godek]

Stephen Compall:

> Actually, if you (use-syntax (ice-9 syncase)), you should be able to
> define lexical symbol-macros that expand a single symbol, even in a
> set! place, something like:
>
> (define v #(1 2 3))
> (define-syntax v1
>  (lambda (stx)
>    (syntax-case stx ()
>      (_ v))))

I don't really get this syntax-rules. Do you know of any good tutorial?
[I find the description in the revised report too difficult to understand]

> Obviously aliasing will serve:
>
> (define-syntax alias-syntax
>  (syntax-rules ()
>    ((_ form)
>     (lambda (stx)
>       (syntax-case stx ()
>         (_ (syntax form)))))))
>
> (define-syntax defalias
>  (syntax-rules stx ()
>    ((_ new form)
>     (define-syntax new (alias-syntax form)))))

ERROR: invalid syntax (syntax-rules stx () ((_ new form)
(define-syntax new (alias-syntax form))))
 => error-in-evaluation
(That's what guile says)

> (define-syntax let-alias
>  (syntax-rules ()
>    ((_ ((new form) ...) body ...)
>     (let-syntax ((new (alias-syntax form)) ...) body ...))))
>
> (defalias v1 (vector-ref v 1))
>
> (set! v1 42)
> (vector-ref v 1) ; => 42

Could you explain it in more detail? Why should it work?
(Or at least give me a link with such explanations)

Thanks in advance
M.

Re: References/locations

[Maciek Godek]

Kjetil:

> Well, here's a relatively clean way, I think:
>
> (define-macro (location name)
>   (define new-val (gensym))
>   `(lambda (,new-val)
>       (set! ,name ,new-val)))
>
> (define old-set! set!)
>
> (define-macro (set! a b)
>  `(if (procedure? ,a)  ;; Needs a better check.
>       (,a ,b)
>       (old-set! ,a ,b)))
>
>
> guile> (define x 10)
> guile> (define y (location x))
> guile> (set! y 20)
> guile> x
> 20
> guile>

I just realized that this is only a partial solution.
Continuing the above example,
guile> y
#<procedure #f (#syntmp-\ g1336-76}#)>

I don't know how to solve it: the problem is that
you actually can't bind a function call to a variable.

But even without it, the line
(define old-set! set!)
breaks:
ERROR: invalid syntax set!

I've been thinking of implementing this "location" stuff
as a smob, but you've got the point that it is (probably)
impossible to implement the location system without
redefining set! and define.

Regards
M.

Re: References/locations

[Clinton Ebadi]

"Maciek Godek" <pstrychuj@gmail.com> writes:

> I've been thinking of implementing this "location" stuff
> as a smob, but you've got the point that it is (probably)
> impossible to implement the location system without
> redefining set! and define.

You may want to read a few documents on functional programming to see
why your locations concept is not good. There are a number of
additional issues as well: one that comes to mind is the interaction
of the garbage collector with pointers into the middle of a vector.

What is so wrong with forms like `(set! (vector-ref foo index) ...)'?
It's generally bad practice to go around destructively modifying data
structures anyway.

-- 
Corinne: this is why we should have designated bath buddies
Corinne: to get places you cant reach because youre slippery and in
         case you get a lil tooo slippery and crack your head open
         someone can call the coast guard and save you

Re: References/locations

[Maciek Godek]

Clinton Ebadi:

>> I've been thinking of implementing this "location" stuff
>> as a smob, but you've got the point that it is (probably)
>> impossible to implement the location system without
>> redefining set! and define.
>
> You may want to read a few documents on functional programming to see
> why your locations concept is not good. There are a number of
> additional issues as well: one that comes to mind is the interaction
> of the garbage collector with pointers into the middle of a vector.

That's what I wanted to hear :)
(The other thing is the ambiguity: if we create a location to another
variable, say (set! y (location x)), then we can't redefine the meaning
of y easily:
(set! y 5) ; sets x to 5, which is unexpected behavior of set!
I think there would be many many more disadvantages)

> What is so wrong with forms like `(set! (vector-ref foo index) ...)'?

In scheme the only problem is that they don't work, unless we
redefine vector-ref:
(define vector-get vector-ref)
(define vector-ref (make-procedure-with-setter vector-get vector-set!))

> It's generally bad practice to go around destructively modifying data
> structures anyway.

Perhaps I'm not an experienced lisper, but I'd rather say it's generally
a bad practice to generalize overly: everything depends on what you're
doing. I'm currently working on an object system that is based on a local
state and it would be a stubbornness to implement it using purely
functional paradigm.

Regards
M.

Re: References/locations

[Clinton Ebadi]

"Maciek Godek" <pstrychuj@gmail.com> writes:
>> What is so wrong with forms like `(set! (vector-ref foo index) ...)'?
>
> In scheme the only problem is that they don't work, unless we
> redefine vector-ref:
> (define vector-get vector-ref)
> (define vector-ref (make-procedure-with-setter vector-get vector-set!))

(use-modules (srfi srfi-17))

(see "(guile)SRFI-17" in infotex)

-- 
emacsen: every copy of Emacs comes with a bag of pot and 5 hits of acid
emacsen: and a hotel coffee maker

Re: References/locations

[Stephen Compall]

"Maciek Godek" <pstrychuj@gmail.com> writes:
> By the way, the let-alias syntax you
> gave me in your former letter doesn't work with guile either.

I am curious about this; was this with use-syntax on syncase, or
use-modules/#:use-module?

Or maybe it has just been a while. :)

-- 
I write stuff at http://failex.blogspot.com/ now.  But the post
formatter and themes are terrible for sharing code, the primary
content, so it might go away sooner or later.