Re: GC method for objects

["Adam D. Moss" <adam@...>]

(following-up to myself...)

Adam D. Moss wrote:
> > A simpler way is to store the dummy created-on-demant heavyuserdata
> > directly in your {}, using a reserved key (e.g., __userdata).
>
> That's possible but pollutes the original table (but that
> might not be a problem for objects, where you wouldn't
> usually for example want to do a pairs() over the table).

Oh of course, silly me; most ways of actually implementing
objects and classes use a proxy table so pairs() doesn't
work properly anyway.

> > > You then need a way to map back from the userdata to the table
> >
> > Since this userdata is dummy anyway, you can store the table as the
> > userdata's metatable.
>
> That's a very nice idea (though then, surely, the __gc field has
> to go into the table itself and the same pollution occurs as above).

The table itself could go inside the userdata's metatable somewhere
(creating a loop, but oh well).  The penalty is that this removes
the ability to share this metatable between all such userdata.

--Adam
--
Adam D. Moss   -   adam@gimp.org