Ah, yeah. That was an alternative I was considering, though I wasn't sure
how to get the size of nativeint -- now I see: sizeof(value)! :)

Still not the ideal, but probably a little nicer than using 'hash' to
reveal abstracted values. ;)

BTW, thank-you Florent, for your articles on interfacing with C... I know
I've referenced them a few times.


On Thu, Apr 5, 2012 at 11:20 AM, Florent Monnier
<monnier.florent@gmail.com>wrote:

> Le jeudi 05 avril 2012 07:57:45, Anthony Tavener a écrit :
> > I'm trying to determine the size of a Bigarray, in bytes.
> >
> > I've tried several things... ending up with this as the only solution:
> >
> > --------------------------
> > let ba_char = Hashtbl.hash Bigarray.char;;
> > let ba_int8_signed = Hashtbl.hash int8_signed;;
> > let ba_int8_unsigned = Hashtbl.hash int8_unsigned;;
> > let ba_int16_signed = Hashtbl.hash int16_signed;;
> > let ba_int16_unsigned = Hashtbl.hash int16_unsigned;;
> > let ba_int32 = Hashtbl.hash Bigarray.int32;;
> > let ba_float32 = Hashtbl.hash float32;;
> > let ba_int64 = Hashtbl.hash Bigarray.int64;;
> > let ba_float64 = Hashtbl.hash float64;;
> > let ba_complex32 = Hashtbl.hash complex32;;
> > let ba_complex64 = Hashtbl.hash complex64;;
> > let ba_int = Hashtbl.hash Bigarray.int;;
> > let ba_nativeint = Hashtbl.hash Bigarray.nativeint;;
> >
> > let bigarray_bytes ba =
> >   let elt_bytes = match Hashtbl.hash (Array1.kind ba) with
> >
> >     | x when x=ba_char || x=ba_int8_signed || x=ba_int8_unsigned -> 1
> >     | x when x=ba_int16_signed || x=ba_int16_unsigned            -> 2
> >     | x when x=ba_int32 || x=ba_float32                          -> 4
> >     | x when x=ba_int64 || x=ba_float64 || x=ba_complex32        -> 8
> >     | x when x=ba_complex64                                      -> 16
> >     | x when x=ba_int || x=ba_nativeint                          ->
> >
> > Nativeint.size lsr 8
> >
> >     | _ -> failwith "Unknown Bigarray kind."
> >
> >   in (Array1.dim ba) * elt_bytes;;
> > --------------------------
> >
> > Which is a rather ugly hack! It feels dirty like Obj.magic, while being
> > completely inelegant, and even has a fail case! Can't get much worse. :)
> > Of course, internally, bigarray will know it's own size... sometimes the
> > price of abstraction... *sigh*
> >
> > I found a related Mantis entry from 6 years back:
> > http://caml.inria.fr/mantis/view.php?id=3962
> >
> > Well... does anyone have some better suggestions? Please? :)
> > Oh, I can make it look a little prettier by plugging in the constants
> eked
> > out by Hashtbl.hash. ;D
> >
> > Why do I want this? Interfacing with OpenGL (glcaml).
> >
> > Thanks for looking, and more again if you have a good tip!
>
> Hi Anthony,
>
> Had the same problem with glMLite.
> I use the function below, but if there is a better solution I would be
> interested too:
>
> val ba_sizeof: ba:('a, 'b, Bigarray.c_layout) Bigarray.Array1.t -> int
>
> external elem_size: ba:('a, 'b, Bigarray.c_layout) Bigarray.Array1.t -> int
>  = "ml_ba_elem_size" "noalloc"
>
> let ba_sizeof ~ba =
>  ((Bigarray.Array1.dim ba) * (elem_size ba)) ;;
>
> CAMLprim value ml_ba_elem_size( value _ba )
> {
>    struct caml_bigarray *ba = Bigarray_val(_ba);
>    int size;
>    switch (ba->flags & BIGARRAY_KIND_MASK)
>    {
>        case BIGARRAY_SINT8:
>        case BIGARRAY_UINT8:
>            size = 1; break;
>
>        case BIGARRAY_SINT16:
>        case BIGARRAY_UINT16:
>            size = 2; break;
>
>        case BIGARRAY_INT32:
>        case BIGARRAY_FLOAT32:
>        case BIGARRAY_COMPLEX32:
>            size = 4; break;
>
>        case BIGARRAY_INT64:
>        case BIGARRAY_FLOAT64:
>        case BIGARRAY_COMPLEX64:
>            size = 8; break;
>
>        case BIGARRAY_CAML_INT:
>        case BIGARRAY_NATIVE_INT:
>            size = sizeof(value); break;
>    }
>    return Val_int(size);
> }
>