Thank you!! this library is very good solution for me!
It seems very useful.


> On Mar 31, 2016, at 4:54 AM, Taylan Ulrich Bayırlı/Kammer 
> <[email protected]> wrote:
> 
> Park SungMin <[email protected] <mailto:[email protected]>> 
> writes:
> 
>>>> If I understood Park SungMin right, the problem is rather with the
>>>> representation of a huge array as a list?
>> 
>> correct! (sorry..my poor english)
>> 
>> I think huge c-array should be representation as bytevector or pointer
>> object.
>> 
>> (define type (list int (make-list 10 int)))
>> 
>> (define struct (make-c-struct type (list 42 (make-list 10 42))))
>> 
>> (parse-c-struct struct (list int '*))
>> => (42 #<pointer 0x2a0000002a>)
>> 
>> but….when i access to that pointer object
>> (pointer->bytevector (cadr (parse-c-struct struct (list int '*)))
>>                   (* 10 (sizeof int)))
>> 
>> => shutdown guile!
> 
> I see.  The problem here is, a struct like
> 
>    struct {
>      int a;
>      int b;
>      int c[2000];
>    } foo;
> 
> declares a memory region of 2002 consecutive ints, whereas
> 
>    struct {
>      int a;
>      int b;
>      int *c;
>    } bar;
> 
> declares a memory region of two consecutive ints followed by one
> pointer.  These are not the same thing.  The C language partly upholds
> an illusion that they're same because "foo.c" implicitly returns a
> pointer to the first integer in the array (i.e. the third integer in the
> struct, after a and b).  On the other hand, "bar.c" returns the pointer
> value that's stored in the struct.
> 
> When we tell Guile to treat foo like bar, it takes an int from the array
> and treats it like a pointer.  So dereferencing it segfaults.
> 
> I don't know what a simple solution would be.
> 
> My bytestructures library *might* help, though it's not properly
> integrated with the FFI yet.
> 
>    https://github.com/taylanub/scheme-bytestructures 
> <https://github.com/taylanub/scheme-bytestructures>
> 
> Here's an example of how to use it in this case:
> 
> |> ,use (system foreign)
> |> (define type (list int int (make-list 2000 int)))
> |> (define struct (make-c-struct type (list 1 2 (make-list 2000 3))))
> |> ,use (bytestructures guile)
> |> (define bs-type (bs:struct `((a ,int) (b ,int) (c ,(bs:vector 2000 int)))))
> WARNING: (guile-user): `int' imported from both (system foreign) and 
> (bytestructures guile)
> |> (define size (bytestructure-descriptor-size bs-type))
> |> (define bs-struct
>     (make-bytestructure (pointer->bytevector struct size) 0 bs-type))
> |> (bytestructure-ref bs-struct 'a)
> $2 = 1
> |> (bytestructure-ref bs-struct 'b)
> $3 = 2
> |> (bytestructure-ref bs-struct 'c 0)
> $5 = 3
> |> (bytestructure-ref bs-struct 'c 1)
> $6 = 3
> |> (bytestructure-ref bs-struct 'c 2)
> $7 = 3
> 
> As you see in the warning, the bytestructures library overrides some
> variables of Guile's FFI library.  I usually import the FFI library with
> 
>    (use-modules ((system foreign) #:prefix ffi:))
> 
> to work around that issue.  (So now the original int is "ffi:int".)
> 
> I don't know if the library is ideal for your use-case, but if you
> decide to use it or want to try it out, ask for help any time. :-)
> 
> Taylan

Reply via email to