modifying pointers

modifying pointers

[Andy Wingo]

Hi Ludo,

You added a change to foreign-set! for %null-pointer. Would it not make
sense instead to make foreign pointers of type "void" unsettable?

Andy
-- 
http://wingolog.org/

Re: modifying pointers

[Ludovic Courtès]

Hi,

Andy Wingo <wingo@pobox.com> writes:

> You added a change to foreign-set! for %null-pointer. Would it not make
> sense instead to make foreign pointers of type "void" unsettable?

Not necessarily.  See the test that was added:

--8<---------------cut here---------------start------------->8---
  (pass-if "foreign-set! other-null-pointer"
    (let ((f (bytevector->foreign (make-bytevector 2))))
      (and (not (= 0 (foreign-ref f)))
           (begin
             (foreign-set! f 0)
             (= 0 (foreign-ref f)))
           (begin
             ;; Here changing the pointer value of F is perfectly valid.
             (foreign-set! f 777)
             (= 777 (foreign-ref f))))))
--8<---------------cut here---------------end--------------->8---

Here a ((void *) 777) pointer is created.

However, this is not allowed for ‘%null-pointer’ because (1) that would
change its semantics, and (2) it’s defined as ‘const’ anyway, so
modifying it leads to a segfault.

Thanks,
Ludo’.

Re: modifying pointers

[Andy Wingo]

Hi,

On Fri 09 Apr 2010 18:30, ludo@gnu.org (Ludovic Courtès) writes:

> Andy Wingo <wingo@pobox.com> writes:
>
>> You added a change to foreign-set! for %null-pointer. Would it not make
>> sense instead to make foreign pointers of type "void" unsettable?
>
> Not necessarily.  See the test that was added:
>
>   (pass-if "foreign-set! other-null-pointer"
>     (let ((f (bytevector->foreign (make-bytevector 2))))
>       (and (not (= 0 (foreign-ref f)))
>            (begin
>              (foreign-set! f 0)
>              (= 0 (foreign-ref f)))
>            (begin
>              ;; Here changing the pointer value of F is perfectly valid.
>              (foreign-set! f 777)
>              (= 777 (foreign-ref f))))))
>
> Here a ((void *) 777) pointer is created.

I'm just wondering if it is valid to create a ((void*) 777) pointer.
Under what condition is that useful?

If that is useful, OK; but under what condition is it useful to mutate
the pointer in a foreign pointer object? Why not create a new foreign
pointer object?

Andy
-- 
http://wingolog.org/

Re: modifying pointers

[Ludovic Courtès]

Hello,

Andy Wingo <wingo@pobox.com> writes:

> On Fri 09 Apr 2010 18:30, ludo@gnu.org (Ludovic Courtès) writes:
>
>> Andy Wingo <wingo@pobox.com> writes:
>>
>>> You added a change to foreign-set! for %null-pointer. Would it not make
>>> sense instead to make foreign pointers of type "void" unsettable?
>>
>> Not necessarily.  See the test that was added:
>>
>>   (pass-if "foreign-set! other-null-pointer"
>>     (let ((f (bytevector->foreign (make-bytevector 2))))
>>       (and (not (= 0 (foreign-ref f)))
>>            (begin
>>              (foreign-set! f 0)
>>              (= 0 (foreign-ref f)))
>>            (begin
>>              ;; Here changing the pointer value of F is perfectly valid.
>>              (foreign-set! f 777)
>>              (= 777 (foreign-ref f))))))
>>
>> Here a ((void *) 777) pointer is created.
>
> I'm just wondering if it is valid to create a ((void*) 777) pointer.
> Under what condition is that useful?

Under unusual scenarios where the program is communicated a pointer
value at the Scheme level, e.g., via ‘object-address’:

--8<---------------cut here---------------start------------->8---
scheme@(guile-user)> (object-address "sdf")
$1 = 43230400
scheme@(guile-user)> (define f (bytevector->foreign #vu8(10)))
scheme@(guile-user)> f
$2 = #<foreign pointer 43814744>
scheme@(guile-user)> (foreign-ref f)
$3 = 43814744
scheme@(guile-user)> (foreign-set! f $1)
scheme@(guile-user)> (foreign->bytevector f 'u32 0 1)
$4 = #u32(21) ;; scm_tc7_string
--8<---------------cut here---------------end--------------->8---

It opens the door to all sorts of crazy things.  :-)

> If that is useful, OK; but under what condition is it useful to mutate
> the pointer in a foreign pointer object? Why not create a new foreign
> pointer object?

There’s currently no procedure to create a foreign object from a bignum.

(Besides, ‘foreign-set!’ exists and has always been defined for all
types of foreign pointer objects.)

Thanks,
Ludo’.

Re: modifying pointers

[Andy Wingo]

Hi!

On Fri 09 Apr 2010 21:46, ludo@gnu.org (Ludovic Courtès) writes:

> Andy Wingo <wingo@pobox.com> writes:
>
>> I'm just wondering if it is valid to create a ((void*) 777) pointer.
>> Under what condition is that useful?
>
[...]
> It opens the door to all sorts of crazy things.  :-)

Cool :)

FWIW, let's kick that door off its hinges :-)))

>> If that is useful, OK; but under what condition is it useful to mutate
>> the pointer in a foreign pointer object? Why not create a new foreign
>> pointer object?
>
> There’s currently no procedure to create a foreign object from a
> bignum.

Perhaps we should define one then. It would be more expressive. Also it
avoids all kinds of shared-state problems.

> (Besides, ‘foreign-set!’ exists and has always been defined for all
> types of foreign pointer objects.)

Sure, but this might be an accident. In every case except `void',
foreign-set! modifies the pointed-to memory. Perhaps we need a
foreign-pointer accessor instead, which would work on pointers to values
as well. And a constructor of course.

Andy
-- 
http://wingolog.org/