Am Dienstag, den 14.09.2010, 21:46 +0100 schrieb Richard Jones:
> On Tue, Sep 14, 2010 at 08:19:49PM +0000, Yoann Padioleau wrote:
> > Hi,
> >
> > I am trying to use your Ancient module to avoid having the garbage
> > collector spends lots of time iterating over huge data in memory. It
> > works quite well for arrays but for hashtbl I have some problems
> > where I am not able to find back keys that were clearly in the
> > original hashtbl (before Ancient.mark it).
> >
> > In the doc it says: 
> > 
> > (1) Ad-hoc polymorphic primitives (structural equality, marshalling
> > and hashing) do not work on ancient data structures, meaning that you
> > will need to provide your own comparison and hashing functions.  
> 
> The issue is described by Xavier Leroy:
> http://caml.inria.fr/pub/ml-archives/caml-list/2006/09/977818689f4ceb2178c592453df7a343.en.html
> 
> As far as my understanding goes, what happens is that the OCaml
> compare function (or some C equivalent in the runtime) looks at the
> two string pointers and decides that since both are out of the normal
> heap they are just opaque objects.  Thus it won't compare the content
> of the strings, but will just do pointer equality.  This massively
> breaks assumptions in some ordinary OCaml code, in this instance in
> Hashtbl.

There is now a way to change this. You can call caml_page_table_add
(since 3.11) to explicitly declare a memory region as containing Ocaml
values. The polymorphic comparison, the hash primitive, and marshalling
work then.

There is support for this in Ocamlnet-3:

http://projects.camlcity.org/projects/dl/ocamlnet-3.0.3/doc/html-main/Netsys_mem.html#VALvalue_area

Gerd


> 
> > which mean I have to transform my code using Hashtbl.xxx into one
> > using the functorized version of hashtbl ? I have hashtbl of strings
> > to complex data type.  What would be a good hash function for
> > strings ?
> 
> It may be that Map also has the same problems.  You wouldn't really
> know except by examining the code.
> 
> Later you wrote:
> > Actually it seems I have the problem only with Hashtbl from strings
> > to whatever.  I also have some Hashtbl from int to whatever and they
> > work fine after the Ancient.mark.
> 
> ints aren't compared in the same way.  They are always compared using
> pointer equality, so there's no issue.
> 
> I've only used ancient to store simple arrays, and when we needed to
> do string equality I remember writing a function which was aware of
> the above issue (you can compare them byte for byte just fine, even
> from OCaml code).
> 
> Rich.
> 


-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------