Re: type friction C <-> scheme

[Mark H Weaver <mhw@netris.org>]

Catonano <catonano@gmail.com> writes:

> 2017-06-10 0:33 GMT+02:00 Matt Wette <matt.wette@gmail.com>:
>
> I think Mike caught your real error: you were passing (list ‘* ‘int ‘*) and
>> it wants (list ‘* int ‘*).  `int’ is a variable defined by guile.
>>
>
> And isn't `*' a variable defined in guile too ?
>
>
>>
>> scheme@(guile-user)> (use-modules (system foreign))
>> scheme@(guile-user)> int
>> $1 = 8
>>
>
> yes, and
>
> scheme@(freexl common)> *
> $14 = #<procedure * (#:optional _ _ . _)>

'int' is a variable exported by (system foreign) whose sole purpose is
to represent a foreign type.

'*' is something completely different, namely a procedure to multiply
numbers.  IMO, it would be a nasty hack for the FFI to check
specifically for a multiplication procedure and interpret that as a
pointer type.

> I took a look at the guile-gcrypt code again
>
> it's full of things like
>
> '(* * ,int ,size_t *)
>
> that is, only asterisks (pointers) are quoted
>
> Other types are not
>
> So this must be an established convention
>
> It just seems unconsistent to me

I agree that it's inconsistent, but '* is shorter than any descriptive
variable name that we might have chosen for this purpose, and pointers
are common enough to warrant a short name.

On the other hand, if we had used symbols uniformly for all foreign
types, then it would have been impossible to bind your own type names
(e.g. for struct types) without adding a global registry of foreign
types, which would have various nasty problems such as the potential for
collisions between unrelated libraries.

     Regards,
       Mark