Re: Fresco

[Willem Rein Oudshoorn]

Björn Giesler <giesler@ira.uka.de> writes:

> On Wednesday 16 April 2003 16:56, Björn Gohla wrote:
> > On 2003-04-16 09:53, Björn Giesler wrote:
> > > while; when I started out, I wanted to eventually write a
> > > NSCORBAConnection. Right now, I don't think that is possible while
> > > still retaining our standard DO semantics.
> > could you explain, why you think it would be hard to do? anyway, having
> > corba connectivity with semantics as similar as possible to those of do
> > would still be quite useful.
> 
> I don't think that a CORBA-ObjC mapping wouldn't be too hard. CORBA has the 
> issue of the formal specification in IDL, though, which is OK if you need 
> formal specification anyway (C++) but isn't if you don't or already have 
> it some other way (ObjC, Protocols). So there would have to be an 
> IDL<->ObjC header file converter.

You mean to say you do *NOT* need a IDL<-->ObjC file converter right?
Because strictly speaking it is not needed.  You could easily
map the IDL interface types to protocols.  And even that could just
be a convention.  (Although you need to know the IDL type of of the object
in question.)

> The other, bigger can of worms is the ORB's run loop and event model, which 
> is rather incompatible to OpenStep's and not specified down to the last 
> iota: Some Orbs spawn a new thread for each incoming request, some don't, 
> some have that configurable, some don't. With a multithreaded Orb, the 
> entire of GNUstep must be thread-safe, etc.

Huh???
I do not have the reference handy, but about a year ago when I 
started using CORBA and studied it to quite some detail all
this stuff is/can be specified.  As far as I remember you
can set quite detailed invocation policies in the POA. 


All speaking from memory.

Wim Oudshoorn.