Re: IDL?

Jayakumar Muthukumarasamy ([email protected])
Wed, 14 Dec 1994 12:03:01 -0500 (EST)

>
> JK,
> >
> > Are we planning to generate code using the IDL's we define? The
> > problem I have with it, is that we will tie all of our implementation to
> > the version of CORBA that we are using. Different CORBA impl's use different
> > styles for generating C++ code and the styles are generally incompatible.
> > Also, generation typically produces more code than is necessary. Moreover,
> > we are not immune to changes in code generation between versions of CORBA.
> > I am defining the interface for Application-IDL using IDL, but, I think the
> > implementation can be done by hand. It would be OK as long as it conforms
> > to the IDL specs. Your comments are solicited.
> >
>
> Given that we want to build a model server as a separate process that controls
> the model, the IDL for the models should be used to generate code. Otherwise
> we cannot build the model server. It doesn't make sense for us to write the
> model code as if it was going to be standalone, and then not be able to wrap
> it with CORBA. I think the implementation of the model should be done to make
> the CORBA wrapping easy.
>
> I was assuming that the .h files that CORBA generates from the idl are pretty
> standard: for each method and attribute there will be a method in the class.
> Is that not so?
>
>

There are parts of the translation that are different. But these
things affect the way in which one would write his/her code. e.g. in the
orbeline implemenation, the string type changes to CORBA::String (a class)
in the c++ implementation, while in the orbix implementation, it changes to
"char *". Recently, OMG has standardized, the conversion from IDL to C++,
but, I haven't been able to get my hands on the documentation, and I am
not sure, to what extent, the standard goes towards specifying the
translation. The people at orbeline said they will release a version of
CORBA that conforms to this standard sometime around Mid-Jan (this
incidently will work will g++ also) and I am not sure, how different this
will be with respect to the current implementaion. It would help to clear
our doubts if we can get hold on the standard somehow. But, till then I was
hoping to use IDL to specify the interface and write the C++ code that
conforms to the interface. The only place, where we would need IDL, is
to specify the interface to the model server as a whole, But, this as you
have mentioned, has its problems. As it looks, we might not get to coding
until mid-Jan anyway. Moreover, since we will be using IDL to specify the
design in anycase, we could keep this going and by the time, we get to
coding we will know one way or the other.

-jk

PS: I am still trying to get the documentation, I will let you know, if I
get hold of something.