> In message <B546CA81.308B%Henrik.Gedenryd at lucs.lu.se> you wrote:
>
> > >> Yes, exactly! Craig's Correspondents is a very complete
> implementation
> > >> of this approach.
> > >
> > > Indeed. It really should go into the system as soon as feasible. I've
> >
> > While I'd hate to reopen that can of worms, I recall a noisy
> debate about
> > this some months ago. IIRC, the conclusion was that Craig's
> stuff, for all
> > its merits, requires rather substantial VM changes as it
> stands, which will
> > require a fair amount of rework on each VM platform, all the while the
> > actual new primitives only had one substantial difference wrt
> the existing
> > stuff.
> That's probably true (can't remember the details anymore) but the nice
> thing is that even if there were  significant differences needed for the
> socket support code it can now be handled by writing a new plugin and so
> on. In fact you could swap at runtime between plugins in order to try
> out one and then the other.
> tim

Since the set of primitives is slightly different for correspondents,
requiring changes in the socket class, you could create a separate socket
class to work with the correspondents framework that utilizes a
correspondents socket plugin.  Then all the existing socket stuff could work
simultaneously with stuff based on correspondents.

This is precisely what we did with XSocket (a socket implemenation built
specifically for windows NT) since it required a different set of
primitives...at the time it required a couple of modifications to the VM
proxy stuff (and thus requiring a new VM to be compiled) to get XSockets
working as a plugin.

Perhaps correspondents could be modified and distributed in this way.

The VM pluginization stuff is great!!!

- Stephen