RE: On WSDL attributes

Savas, we agree. General Web services do not need a first-class notion
of persistent state associated with a service.
 
Adding functionality to the Web service model necessarily implies
additional constraints on underlying implementations. To facilitate
cross-platform interoperability, we must be careful to minimize the
architectural constraints on implementations. The current
message-oriented definition of Web services appears to be an excellent
tradeoff between function and minimal architectural commitment.

There are communities who wish to associate state with service
instances; our WG should ensure that they can do so through
extensibility.

--Jeff

> -----Original Message-----
> From: www-ws-desc-request@w3.org [mailto:www-ws-desc-request@w3.org]
On
> Behalf Of Savas Parastatidis
> Sent: Tuesday, September 09, 2003 4:44 AM
> To: WS Description List
> Subject: On WSDL attributes
> 
> 
> All,
> 
> Here are some thoughts by Jim Webber and myself on the introduction of
> attributes in WSDL...
> 
> 
> The proposal on the introduction of attributes in the WSDL
specification
> has come a long way due to the work that the relevant Task Force has
> produced. The most recent update to the proposal is, we believe,
closer
> to the WSDL way of describing message exchanges. However, it is so
close
> that we believe attributes are not necessary for WSDL.
> 
> 1. Attributes represent a concept that it is not part of the Web
> Services Architecture. Nowhere in the WSA document, to our knowledge,
is
> it suggested that a Web Service has attributes. Web Services send and
> receive messages. They do not have operations, functions, methods, or
> attributes as it is the case with object-based component models. WSDL
is
> used to describe messages that can be sent and received. The notion of
> an "attribute" attempts to add a characteristic to Web Services that
> simply does not exist.
> 
> 2. The current version of the proposal defines particular message
> exchange patterns. Since there is already work being carried out in
this
> area, we feel that there is no need for the introduction of attributes
> in WSDL. For instance, the notion of a solicit-response MEP is
somewhat
> analogous to "getting" an attribute, while a request-only MEP is
> analogous to "setting" an attribute. Given these abilities which are
> already an accepted part of WSDL, this undermines the need for "read",
> "read-write", and "write" qualifiers for attributes.
> 
> To summarise: We believe that attributes are a fundamental property of
> object-based systems, and do not have a corresponding use in SOA. We
are
> supported by WSA in this thinking. Furthermore, we believe that the
> benefits of an attribute style interaction in terms of being able to
> "set" or "get" structured XML data from a Web service is already
> supported with WSDL operations.
> 
> While we think that the work of the ATF is correct in itself, we would
> oppose the inclusion of such work into WSDL, and would instead
> anticipate that it would form part of some other specification which
> leverages WSDL extensibility. Those communities which have a
> demonstrated need to deploy Web services in a distributed object-like
> scenario (e.g. Grid), can then utilise the separate attribute
> specification to support their needs, without adding non-WSA features
to
> WSDL.
> 
> Regards,
> Jim Webber
> Savas Parastatidis
>

Received on Wednesday, 10 September 2003 15:19:37 UTC