opencog-dev team mailing list archive

[Gustavo Machado Campagnani Gama <gama@xxxxxxxxxxxxx>]

Hi Ari,

On Wed, 2008-10-22 at 10:28 +0300, Ari Heljakka wrote:
> Hi Gama,
> 
> > * Started investigating another suggestion from the Petaverse port:
> >   replace the Handle typedef with a real class. This should help us find
> >   out several semantic bugs in the code that rely on the assumption that
> >   Handle == unsigned long.
> 
> I remind that the vtree system relies on the assumption that Handles can 
> be converted to (long) integers. There is a natural way to get rid of 
> this assumption by extending the Vertex variant to contain also your new 
> Handle class (instead of containing the Handle-integer) and then rewrite 
> the code that relies on the integer assumption, correspondingly.
> 
> The code that relies on this assumption includes most significantly PLN, 
> and less significantly the space-time schemata used by the reasoning 
> module, and maybe some of the older MOSES code.

The current Vertex typedef is a boost::variant<Handle, ...>, and not
boost::variant<unsigned long, ...>. Apparently this is enough to handle
all of the unit tests that use the vtree system (which, unfortunately,
aren't many).

However, it's likely that we'll have to make some adjustments when the
PLN and/or MOSES code is merged. But it didn't take too long for Welter
and I to migrate the current opencog tree, so I'm hoping that the same
will apply to PLN and MOSES.

Anyway, thanks for the heads up.

--
Gustavo