Re: [Chicken-users] Multiple return values

[felix]

Peter Keller wrote:
> > [syntax] (define-location NAME TYPE [INIT])
> >   the same as `define-external', but not visible
> >   outside the current file.
>
>
> Why did you make this choice to keep the variable local to the current
> file? What would happen if you lifted this restriction?

`define-external' is exactly what you want.
`define-location' is really just a special case of this
form.

> > [syntax] (let-location ((NAME TYPE [INIT]) ...) BODY ...)
> >   creates a lexically bound C variable.
>
>
> I like the [INIT] bit since it allows you to translate between scheme
> objects and C types.
>
> Though, suppose something like this happens:
> (define foo 134324234234) ;; Suppose you used my GNU MP FFI...

[I got the hint... ;-)]

> (let-location ((i int foo))
>         ...)
>
> Now, chicken has to check to see if foo will "fit" into the integer i, if not
> then i has to get assigned INT_MAX or something like that.
> However, since chicken currently can't support numbers larger than a C_word
> it won't be a problem, but it should be thought about. And, uh, what about
> long long. :)

Oh, yes... I remember that one. ;-)

> What about arrays?
>
> ;; does this make sense?
> (let-location ((i integer (s32vector 1 2 3 4 5 6 7)))
>         ...)

No. `i' is an integer, but you assign a vector.

> > I still have to run a couple of tests (generated
> > FFI code has been slightly improved), but this should
> > be available soon.
>
>
> Will this affect current FFI code?

It should generate more compact and more efficient code
(just slightly).

> > Notes:
> > - the types given to define-location/let-location
> >   have to have known sizes (only simple or pointer types)
>
>
> I, and I'm sure others, really appreciate the additions you are doing
> to chicken in this area, Felix. Thank you.

Hey, you're welcome. I have to thank you and all others for
doing so much work for it. Your MP library, and the mass of code
Tony is grinding out will hopefully converge to something
genuinely competitive to other Scheme implementations (or even
other language environments).

> P.S. I suppose if I pester enough, a solution for structures might be
> figured out as well? *grin*

You mean for locations? Unfortunately we can not properly
compute the size of a struct. Too many factors come into
play, like struct-alignment, etc. But what we need is some
kind of `sizeof' operator...


cheers,
felix