[Haskell-beginners] Ffi, memory allocation and failure

[Antoine Latter]

On Wed, Feb 9, 2011 at 8:57 AM, C Gosch <ch.gosch at googlemail.com> wrote:
> Hi Edward, hello to all list members,
> thank you for your comments, I will keep that in mind. There will be
> operations on matrices (and other objects) which modify the objects,
> and those will be in IO or ST, I think.
> However, I am planning to provide operations with matrices which guarantee
> that each involved matrix is not changed, but which may return new matrices.
> All matrices have to be allocated, and that allocation can fail ... how does
> one usually go about this problem in Haskell?
> Cheers,
> Christian
>

I would check for allocation failure in the C wrapper and return a
status code or error message back to Haskell. Thus the prototype of
the C wrapper might be:

> result_code_t c_create_matrix(t1 arg1, t2 arg2, c_matrix *result);

Then you could either throw a Haskell exception (which is allowed from
pure code - but can only be caught in IO) or do everything under
'Maybe'.

If you expect allocation failure to be rare and exceptional, I would
lean towards the exception. I haven't had the occasion to wrap C
libraries were I expected this sort of thing.

Antoine

> 2011/2/9 Edward Z. Yang <ezyang at mit.edu>
>>
>> Hello Christian,
>>
>> Assuming that the foreign library is based off of mutation on matrices,
>> there's not an easy way to export a nice pure interface to them, because
>> they're not actually persistent: if another thread goes and modifies
>> some pointer which you also have in pure code, referential transparency
>> is broken.  So you'll have to think carefully about what sorts of
>> guarantees
>> you can (or want) to make to Haskell, and if you can't fulfill them, keep
>> the code in IO (or create your own monad for manipulating matrices).
>>
>> Edward
>
>
> _______________________________________________
> Beginners mailing list
> Beginners at haskell.org
> http://www.haskell.org/mailman/listinfo/beginners
>
>