RE: On WSDL attributes

Steve:

> We continue to debate the merits for this work.  We have 
> constituencies in the Systems Management space (OASIS WSDM 
> TC) and Grid (OGSI) that clearly need the notion of 
> attributes in WSDL.  We do not believe the concept is limited 
> to those two domains, the concept is much broader than that.

But unlike message exchanges, it is not fundamental to _all_ Web services.
Therefore as a matter of good factorisation we would expect it to be in a
different spec. Sure, that spec might be referenced by OGSI, WSDM and
others, but it is not fundamental.

> We agree that a Web service fundamentally is about message 
> exchanges.  The attribute work defines additional information 
> that helps the requestor form these messages, by indicating 
> what sort of information can be the subject of the get and 
> set operations. Furthermore, it does this in such a way that 
> allows more powerful interfaces (such as OGSI service data 
> operations) to be defined to reuse the base information 
> provided by the attribute declaration mechanism.

This is a chicken-and-egg situation. You have to send messages to "get" your
attributes then you send messages. Either way you're dealing with message
exchanges, which is the important thign about Web services.

> You claim that attributes are not a concept in the WSA.  
> However it is true that they are not FORBIDDEN by WSA.  This 
> extension is needed for Grid and Systems management, and may 
> be quite useful in other spaces. 

Many other things are not forbidden by WSA. However WSA will be the document
that architects and developers will look to for guidance on building Web
Services based applications. It's going to surprise them if we add
technology that specifically isn't in WSA with the punchline that it's not
forbidden. We shouldn't put the cart before the horse: WSDL should be driven
by WSA more than it should drive WSA.

> BPEL has properties for 
> example, that appear to be a similar concept, would it not be 
> better if  we all used the same mechanisms for 
> properties/attributes etc? 

I don't understand this. My understanding of BPEL properties is that they
expose parts of a message as a unique "key." As such adding attributes to
WSDL complicates BPEL because now BPEL people have to have a way of
associating variables with attributes as well as bits of messages.

> We agree that attributes in WSDL 
> does not exist, we are proposing, based on our experience in 
> Grid and Systems management that it be added.

As outlined above, our position is that such an addition is poor in terms of
factorisation.

> The notion that the work in MEPs is sufficient and precludes 
> the work for attributes is misguided.  We proposed some 
> simple access mechanisms to suggest how bindings treat 
> attributes in an analogous way to operations. The MEPs work 
> does not reduce the need for attributes.

That may be a valid point, yet even if we concede it we maintain the poor
factorisation argument has priority.

> We continue to disagree about the relationship between 
> objects and SOA. Object orientation is an implementation 
> concept.  There is nothing in SOA that precludes an 
> object-orientation or a non-object orientation.

That seems to be a hot topic in the WS community at the moment. We wrote an
article on it for Novermber's WSJ, and Wener Vogel's excellent take on this
is also doing the rounds:
http://weblogs.cs.cornell.edu/AllThingsDistributed/archives/000120.html

Essentially it says that SOA and OO are chalk and cheese. Our take on this
is that OO is an implementation strategy and SOA is an integration strategy.
You disagree with our take, I know, but I think our view is beginning to be
widely shared. Thus when thinking about Web services architecture, we should
think purely in terms of message-level contracts, not in terms of shared
type systems.

> Of course, we could always use extensibility to support 
> attributes, in fact this what OGSI does today with 
> serviceData.  However, it was deemed appropriate, by several 
> constituencies to try to standardize attributes in the core language.

Certainly we don't disagree that the need exists in OGSI. Nonetheless it
remains our view that this is out-of-scope for core WSDL.

Jim & Savas

Received on Tuesday, 9 September 2003 10:25:32 UTC