I view Celtix as you said: "as a product that provides implementations to several bindings." The same can be said for Axis with its support for different transports, and I would not be surprised if there are yet more (e.g. ServiceMix).

The spec has not "preserved" binding names - it is specifying a vendor-independent way of representing a binding using a common technology. It also expects there to be multiple implementations of that common binding - for example, by different SCA implementations. It fully expects that there will be other bindings defined and simply requires that such bindings have an identifying element from a different namespace (which is simply good XML). The use of "binding" in the name is good form but is not normative - you could have a binding with the name <foo:booBar> if you wanted.

IMO if someone uses the spec binding (e.g. <binding.ws>) they are explicitly saying that they do not care which implementation to use - and indeed it should not matter as all implementations should conform to the spec. They may do this for portability between different platforms, or between different configurations of a platform. Alternatively, if they want to us a specific implementation then they must express that intent and the way the spec supports that to be done is through implementation-specific bindings.

The spec also allows the user to specify multiple bindings at the same time - for example they could use <binding.ws>, <binding.ejb>, <axis:binding.axis> and <celtix:binding.jms> concurrently. The spec leaves it to the runtime to determine which of those are actually activated.

In terms of our implementation, we have a loading mechanism that loads implementations explicitly: the single component that registered with the LoaderRegistry for an element is the one that is used to handle that element. A component can register to handle multiple elements, and the loader can choose what happens when multiple components register for the same element (IIRC at the moment it goes with the last one, it could also ignore duplicate attempts i.e. go with the first one).

IMO this is simple, precise and flexible. It allows an administrator to easily configure the runtime using extensions by simply configuring in an implementation. That implementation can support a single element, or it may support multiple ones - it can even support none meaning it can only be used from non-XML based assemblies. The policy for determining what to do when multiple implementations register for the same element is determined by which implementation of LoaderRegistry is used. It uses the spec's method of choosing between implementations: using an implementation specific element.

I don't see why we need to change this but if you have a use case in mind please share.

--
Jeremy

On Aug 16, 2006, at 3:10 AM, Liu, Jervis wrote:

Hi,

One question has to be answered before we go any further is that we view Axis and Celtix (and others like Xfire) as different implementations of Web Service binding or different bindings.

If we view them as same binding but different implementations, we should stick to the schema of binding.ws and we have to solve the problem I raised previously: how to switch different impl for the same binding under the scene? This problem can be probably tackled by some kind of loading mechanisms, like the extension dir we are using at the moment. We may run into some limitations though, for example, if two implementations of same binding are both available to the loading mechanism, how to load a specific explicitly?

If we view Axis and Celtix as different bindings, e.g., we represent them using the schema of binding.axis and binding.celtix as described by the extension mode in spec, we can always load axis or Celtix explicitly, there is no need to invent a loading mechanism, all bindings can live happily together in the extension dir. However, I found the meaning of binding.axis or binding.celtix is confusing and does not following the concept in spec. SCA spec has preserved binding names for binding.ws, binding.jms, binding.eis etc, them are all names for different bindings/ protocols, but Axis or Celtix is really an product other than a binding or protocol type. Celtix as an ESB, supports various bindings/protocols besides soap/http, such as JMS and CORBA. Conceptually, I found its more clear to view Celtix as a product that provides implementations to several bindings (thus we can have a Celtix version loader/builder for each bindings, e.g., org.apache.tuscany.binding.celtix.WebServiceBindingLoader and org.apache.tuscany.binding.celtix.JMSBindingLoader) other than having a binding.celtix in Tuscany.

In summary, viewing axis or celtix as different implementations to the same binding is more clear in concept, but it needs us to invent a loading mechanism that can load the specific implementations explicitly. Not sure about other guys' opinion, any thoughts or comments?

Jeremy, I also have a question for you. What do you mean by saying "allow the extension directory to contain simple SCDL files (xml files) that define composites and which get their code through dependencies"? Do you mean adding a SCDL for extension dir to describe all extensions available under the extension dir? Then we take this scdl as a single composite that can be deployed into runtime? But this does not seem to address the loading problem (when multiple implementations available).

Thanks,
Jervis


-----Original Message-----
From: Meeraj Kunnumpurath [mailto:[EMAIL PROTECTED]
Sent: Tuesday, August 15, 2006 7:48 PM
To: [email protected]
Subject: RE: Create distribution for bindings, or how to
switch binding
implementations


Jervis,

Few thoughts ...

I think flexibility and ease of use is important in terms of switching
between bindings, not just two implementations of the same transport
(for example axis and XFire for WS binding), but also between two
different transports (WS and JMS for example). With one of the ESBs I
have been using it is as simple as switching the protocol in the
endpoint URI for the target component. For example, you can switch
between Axis to XFire, by changing the endpoint URI of the component
from axis:http://localhost:81/services to
xfire:http://localhost:81/services, and if you want to switch
from WS to
JMS, it is as simple as changing the URI to jms://myQueue. I think the
gist is providing the flexibility to expose the services over
a variety
of transports, even more than one if that is deemed
necessary. According
to my understanding this issues are all addressed by the SCA spec.

I think the spec mandates support for SCA and web services
bindings and
talks about having extensibility mechanisms for other kinds
of bindings.
Though it should be possible to expose services across multiple
bindings, multiple implementations of the same transport
(XFire and Axis
for example) wouldn't make much sense, as the implementation
type itself
is transparent to the service consumer, unless different
ports are used
by the different bindings. I think, may be the issue is the mandatory
binding binding.ws from the spec perspective. You could look at
binding.ws as the default WS binding used by the SCA runtime
implementation (using axis for example) and to use a different WS
binding one could use the extension model and have a binding like
binding.ws.xfire.

Ta
Meeraj

-----Original Message-----
From: Liu, Jervis [mailto:[EMAIL PROTECTED]
Sent: 15 August 2006 03:56
To: [email protected]
Subject: Create distribution for bindings, or how to switch binding
implementations

Hi,

I can see from yesterday's IRC chat that we had a quick discussion on
how to create distribution for bindings. I think the main
problem we are
facing is how to switch binding implementations. For example, we may
want to switch the implementation of binding.ws between axis2
and celtix
for helloworldws sample, in an easy and user friendly way.
The approach
Jeremy proposed should work: package the binding as an extension and
somehow install it into the runtime(put it into the extension
dir?) . So
this is definitely one option. But I think we should also provide
alternatives to better address following scenarios:

1, Tuscany users installed a generic Tuscany distribution, and they
would like to be able to switch binding implementations without moving
libraries around or changing any directories. I think we can improve
usability if users only need to change a config file.

2. Some applications may want to use two binding
implementations at the
same time.

It seems to me that we will need a configuration somewhere to specify
the specific binding implementation. Can we have a
proprietary entry in
scdl file sth like <binging.ws.axis2/> or <binding.ws
bindingImplementation=axis/>? Any comments are welcome. To
better track
this thread, I have created jira 621.
https://issues.apache.org/jira/browse/TUSCANY-621. I also enclosed
relevant IRC chat log for your info.

Thanks,
Jervis

(05:27:47) cr22rc: jboynes : I'm ok about taking it out ..
but what are
we looking for doing for samples needing axis wsbservice binding
(05:28:44) jboynes: I want to package the axis binding as an extension
that can be installed in the runtime
(05:29:46) jboynes: so to run the sample you add the axis extension to
your installation
(05:30:43) ant: or we may have a distribution that includes
axis right?
(05:31:03) jboynes: sure
(05:32:32) jboynes: I want to make sure that the basic
concept (core + a
bunch of extensions) works
(05:33:19) kgoodson left the room (quit: Read error: 110 (Connection
timed out)).
(05:33:23) ant: could we talk about the ServletHost stuff now?
(05:33:33) jboynes: so we don't end up in a situation where
an extension
only works if it is packed into a distro in a special way
(05:33:40) jboynes: I need a couple more
(05:33:48) jboynes: I need to eat
(05:35:32) ant: ok, well ping when you're ready
(05:48:04) jboynes: ant: hi, better now I've had breakfast :)
(05:48:30) ant: yum
(05:48:31) jboynes: sorry for holding things up - I just reached the
pass out or get cranky phase
(05:48:48) dkulp left the room (quit: "using sirc version
2.211+KSIRC/1.3.12").
(05:49:19) ant: ok so there's an interface ServletHost
(05:50:02) jboynes: yep
(05:50:10) ant: so the WS binding should use that to register
a servlet
for each ws endpoint


This message has been checked for all email viruses by MessageLabs.



*****************************************************

    You can find us at www.voca.com

*****************************************************
This communication is confidential and intended for
the exclusive use of the addressee only. You should
not disclose its contents to any other person.
If you are not the intended recipient please notify
the sender named above immediately.

Registered in England, No 1023742,
Registered Office: Voca Limited
Drake House, Three Rivers Court,
Homestead Road, Rickmansworth,
Hertfordshire, WD3 1FX


This message has been checked for all email viruses by MessageLabs.

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to