[redland-dev] implicit free

[Maxence Guesdon]

On Fri, 25 Nov 2011 12:48:36 +0000
Norman Gray <norman at astro.gla.ac.uk> wrote:

> 
> Maxence, hello.
> 
> On 25 Nov 2011, at 12:09, Maxence Guesdon wrote:
> 
> > My first guess was to make my own counter for each redland structure
> > pointer, then decrement this counter when the ocaml value embedding the
> > pointer was reclaimed by the Gc (when it becomes unreachable).
> > But since I'm not aware (from the ocaml side) of internal way of handling,
> > for example, nodes in a statement, I may free, for example, a node even if
> > it's still referenced internally by a statement.
> > 
> > Since there is a kind a recursive freeing, I may just don't care about C
> > structure pointers and let the developper use the free functions himself.
> 
> I think that part of the point of using a GC language is to free the developer from having to mess about with low-level rubbish like free functions.
> 
> My approach in the Racket library was to make a thin-as-possible Racket layer around the C library, which is in turn wrapped by a Racket layer which exposes the resulting objects in a more scheme-ish fashion.
> 
> So in the wrapper layer, I return scheme objects which encapsulate the librdf structs (librdf_uri, librdf_node, and so on) that come back from the functions called.  These structs are therefore conceptually 'owned' by the Racket layer, so where the librdf documentation notes that a returned object is shared, I make a copy of it using one of the librdf copy-create functions.  The only place where I recall that came unstuck was in the bug reported in <http://bugs.librdf.org/mantis/view.php?id=478>, where one of the copy constructors turned out to be a shallow copy rather than a deep one.
> 
> The Racket FFI allows me to associate a 'custodian' with each scheme object the library creates.  The custodian encapsulates code that is run when an object is freed by the GC, and that's where I can run librdf_free_* functions, without the user being troubled by them, and without me having to worry about reference counting.  I would imagine the OCaml FFI would have a similar capability.

Thanks for your explanation. If I understand correctly:
- there is smal additional cost (memory and cpu) due to copying,
- sharing is prevented.

Note that in a functional language, this may not be a problem since I
prefer returning copies rather than doing side effects.

Regards,

Maxence