[cap-talk] OO interoperation via OCap paradigm, vats (was: Re: Security and languages talk)

[Jed Donnelley]

On 5/5/2008 11:42 AM, Rob Meijer wrote:
> On Mon, May 5, 2008 11:20, Matej Kosik wrote:
...
> A very annoying problem with regular languages today is with
> interoperability.There really isn't a good way to couple OO components
> written in different languages, not to mention trying to do the same
> using ocap languages.
> This field of interoperability might be a good way to potentially break
> the deadlock. It seems that creating an ocap based IPC mechanism that
> focuses
> on coupling the different languages as much as local or distributed
> components written in the same language, may make each of the languages
> much more useful, just by creating a way to combine for example perly with
> java-ish code in a natural and ocap compliant way.

I find a great deal of appeal in the above notion.  While
I'm far from an expert in the language area, it does seem
to me that if our OCap model/paradigm is as effective as
we argue it is, then we should be able to use it as a
means of communication between components written in
different OCap languages (E, Joe-E, Caja, ...?).

Part of the appeal for me is that this approach seems
to me similar to the way I've used and argued for further
use of network communication as a means of insuring
what I refer to as "network discipline" on capability
operating systems.  I suppose you could also call this
"membrane discipline", but I believe it is the serialization
of capabilities and their protection properties for the
network that really forces the clarification of what might
otherwise seem minor issues.

For me the fundamental issue is that both operating systems
and languages are in a position of great "power" and
therefore flexibility that can (and in my experience usually
does) lead to designs that differer from the OCap model
in perhaps subtle or otherwise not easy to notice ways.
By setting up such languages and/or operating systems to
communicate through "vat"s (serialized over network
communication), any deviations from the nominal
OCap model should quickly show up.

Has this approach been tried with, say, E, Joe-E,
and Caja?  I.e. can the above languages communicate
across a vat network interface?

--Jed  http://www.webstart.com/jed/