You are viewing a plain text version of this content. The canonical link for it is here.
Posted to soap-dev@xml.apache.org by Gerd Aschemann <ge...@aschemann.net> on 2000/07/24 15:40:48 UTC
NASSL vs. CORBA/IDL
(My background are some years of experience with CORBA with ORBacus,
MICO, COPE ... please keep this in mind.)
I was looking for a high level description for SOAP services, ideally
I would like to have some sort of IDL. The only available tool I found
was NASSL and I gave it a try. Now I have some questions and comments
on it.
0. There are some bugs and very few documentation on NASSL. I would
really appreciate to have it in source in a short term ... How
about the efforts here?
1. I would like to distinguish between the interface and the
implementation in a better way. An IDL compiler usually generates
several files:
- an "interface" file
- a client stub which implements the interface (NASSL: proxy)
- a server skeleton which implements the interface as an abstract
class
- optionally some compilers generate some sort of an
implementation template which is a starting point for the
implementation class. The implementation is usually provided by
the developer as an extension of the abstract skeleton class.
NASSL currently generates me a proxy (IDL: Stub) and a "dummy
implementation" for the server side. I would prefer to have an
(Java) "interface" instead which can be implemented by me, and
maybe optionally the kind of template as a starting point for the
implementation.
I would like to have an NASSL-interface X and a generated Java
"interface X" and a self implemented "class XImpl implements X", or a
generated "abstract class _XImplBase implements X" plus a "class
XImpl extends _XImplBase" ... (the latter currently doesn't make
much sense, but maybe in the future).
2. How about Java namespaces? Is it possible to have the generated
classes/interfaces/proxies to be in a Java package, eg.,
"com.XXX.YYY..."?
3. For the proxy/to find the service it would be nice to have some
sort of IOR ... I understand two things:
- We don't (yet) have object references (see the spec and the mails
from Michi Henning for a discussion on this).
- But: We do have some sort of "service reference", a URL-URN pair,
derived from the providerDef (deployment descriptor).
This information is hard coded into the proxy class. Wouldn't it be
a good idea to have some sort of a "SOAPService" class which
contains such information and to have the proxy constructors (or
additional proxy constructors) which can be initialized with such
an object. See also the next point ...
4. On the first look it seems reasonable to encapsulate all
information about a service within one file. This is OK if one only
wants to provide a few implementations or to deploy the service a
few times. But consider a scenario where we have a single interface
and only one or two implementations, say one in Java and one in
Perl, but many many many deployments. This is the case for us (I am
consulting for a company): we describe and implement an interface
and sell the product several times where each customer deploys it
one or more times.
So maybe it would be a good idea to separate the specifications a
little bit more with some well defined relationships among them,
eg., the provider definition including or refering to an
implementation definition. The implementation definition could
refer to an interface definition. Think of it as some sort of
component description like CORBA components or EJB where you also
have different roles (developers, assemblers, system
administrators) with respective descriptions.
5. Finally there are some major bugs in the NASSL implementation, eg.,
if the .nsl contains an interface:
...
<nsl:interfaceDef name="DBQuery">
...
and an implementation descriptor:
...
<nsl:implements interface="DBXQuery"/>
...
The compiler will not generate Proxies and "dummy implementations".
It will not even report an error. But it will generate "wrong"
DeploymentDescriptors ...
6. It would be nice to extend NASSL to also generate Perl code.
So again: To get the source would be great!
Regards,
Gerd