Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!cs.utexas.edu!usc!elroy.jpl.nasa.gov!aero!obrien
From: obrien@aerospace.aero.org (Michael O'Brien)
Newsgroups: comp.lang.smalltalk
Subject: Re: C to Smalltalk
Message-ID: <82085@aerospace.AERO.ORG>
Date: 24 Aug 90 19:50:11 GMT
References: <1780@abvax.UUCP> <81923@aerospace.AERO.ORG> <381@scubed.SCUBED.COM>
Sender: news@aerospace.aero.org
Reply-To: obrien@aerospace.aero.org (Michael O'Brien)
Distribution: usa
Organization: The Aerospace Corporation
Lines: 51


|> >
|> >In article <1780@abvax.UUCP>, jaz@icd.ab.com (Jack A. Zucker) writes:
|> >|> I know this is a naive question, but is it possible to
|> >|> call smalltalk methods from C using either the Digitalk
|> >|> or Park Place products ?
|> >
|> In article <81923@aerospace.AERO.ORG> obrien@aerospace.aero.org
(Michael O'Brien) (That's me folks) replies:
|> >	Goodness knows what ParcPlace has planned for the future - they
|> >are enough of a business now not to let much out - but in the current
|> >release it's definitely impossible.  
|> 
In article <381@scubed.SCUBED.COM>, warner@scubed.com (Ken Warner) writes:
|> ------- THIS IS NOT A FLAME ----------
|> 
|> I am curious.  What sort of things did you (Jack) want to do that you
feel are
|> best done in this way.  And why could you not go the "other" way by
calling C
|> functions via primitives from Smalltalk?

	Obviously I'm not Jack, but on my part, I found early on that if you
could call a given package from C, you could usually call it from any damn
thing.  My first big Smalltalk project was to do a user interface to a giant
Prolog system.  Prolog could call C, or thought it could, and so I could
have had the interface bound into the Prolog system - except that Smalltalk
couldn't be called by C.

	In the end I wound up doing it in reverse: Quintus Prolog defined
a funky pseudo-RPC interface to allow itself to be called by C (i.e. going
the opposite way).  I put protocol stuff into the ParcPlace VM to do the
job and away we went.  Luckily UI things operate on a time scale that allowed
the RPC overhead.
|> 
|> One wild blue approach to calling Smalltalk from C (just off the top of my 
|> early morning head) would be to supply a set of entry points into a VM by
|> installing a set of of primitives.  Then link the VM into a C object.
Pretty
|> wild huh?  I wonder what it would take to do this. I have a real
strong feeling
|> that this can't be done with the current VM.  It's just a thought for
the sake 
|> of brain-storming.

	Well, I don't think this would really work, since for one thing the
load module would wind up with two "main()"s.  And Smalltalk wouldn't get
to initialize its image, and, and, and...
--
Mike O'Brien
obrien@aerospace.aero.org