Re: [Chicken-users] 32 bit integers?

[Kon Lovett]

On May 22, 2007, at 10:50 AM, Dan Muresan wrote:

> Hi,
>
> I've just realized that "normal" Chicken integers are 31 bits wide:
>
> csi> (inexact->exact 1073741824)
> [ nasty error ]
> csi> (inexact->exact 1073741823)
> 1073741823
> csi> (/ (log 1073741824) (log 2))
> 30.0
>
> Wider integers are promoted to flonums.

Or to bignums with the numbers egg (full numeric tower).

#;1> (use numbers)
; loading /usr/local/lib/chicken/1x/numbers.scm ...
; loading /usr/local/lib/chicken/1x/numbers-base.so ...
#;2> (inexact->exact 1073741824)
1073741824

Means you want to compile w/ generic-arithmetic & usual-integrations  
when using a foreign call.

> This is somewhat undesirable when interfacing to C libraries.

Especially if you didn't expect it. (It is documented but a tutorial  
on foreign function interfacing would be nice.)

If you use one of integer unsigned-integer integer32 unsigned- 
integer32 integer64 long unsigned-long as a foreign parameter/return  
type Chicken will automatically convert a flonum integer to/from the  
native type.

> Is there a native-integer type (like in OCaml)? Since the SRFI-4  
> vectors have exact 8, 16 and 32-bit integer elements, it might make  
> sense to have native integers too...

Sort of, see above. Note that SRFI-4 doesn't define any arithmetic.

Would the native integers be part of the numeric tower? If so this is  
a major piece of work. If disjoint from the number type then fairly  
straightforward to create but of limited utility.

Best Wishes,
Kon

> Cheers,
> Dan
>
>
> _______________________________________________
> Chicken-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/chicken-users