W3C home > Mailing lists > Public > www-ws-desc@w3.org > April 2002

Re: issue: remove solicit-response and output-only operations?

From: Roberto Chinnici <roberto.chinnici@sun.com>
Date: Tue, 16 Apr 2002 18:29:01 -0700
Message-ID: <3CBCCFDD.203B4175@sun.com>
To: Prasad Yendluri <pyendluri@webMethods.com>
CC: www-ws-desc@w3c.org
Prasad Yendluri wrote:

> Hi Roberto,
>
> Roberto Chinnici wrote:
>
> > It seems to me that if we remove solicit-response and notification operations
> > from interfaces (portTypes), we'll remove a major source of confusion in WSDL.
> >
> > Both mechanisms described in Prasad's email can be recovered by putting what
> > used to be solicit-response and notification operations in their own port type
> > and reversing them, making them request-response and one-way operations.
>
> <PY>
> I am not sure if I understand what is being proposed above. How would this portType
> be different from a request-response/oneway port type?

<RRC>
It would not be different at all, that's exactly my point (see the rest of my proposal).
</RRC>

> What I am looking for is exactly what Pallavi is looking for also. That is, how do we
> capture outbound requests (from the request initiator perspective) in addition to the
> current service provider (request-receiver / responder) perspective, so that we can
> capture both client and server views of the exchanges (as in a business process).
>
> </PY>
>
> > A future event-notification description language or an upper-layer business
> > process language could then use this new port type for its own purposes.
>
> > If we need to capture the fact that a WSDL 1.next service defines some
> > outbound operations, which I'm guessing is what Pallavi is hinting at in
> > her message, we can allow adding "outbound ports" to a service (and
> > later to a service type). It could be as simple as adding a new attribute
> > direction="inbound|outbound" (with inbound as default value) to <port/>.
> > The full specification of an event/callback registration mechanism can
> > then be done later. Notice also that it's much easier for a tool to match
> > up a client and a server if they both declare ports of the same type
> > but with opposite directions, as opposed to identifying conjugate
> > operations in different portTypes.
>
> <PY> This I think would  be an alternate way of capturing what we are looking for.
> So, we will have two ports with the same (type and) binding but, for the port
> representing request-initiating (client) end-point have the direction attribute set
> to "outbound" (and "inbound" for the server side). Seems reasonable to me. However I
> see some difficulty in representing an exchange that goes back and forth where the
> client and server switch roles back and forth in realizing a business process. That
> is one side is sometimes a client to the other and vice-versa.

<RRC>
This should be easy to do. If two peers need to play both the client and the server
role of the exchange, they would each declare two ports of the same type, one
inbound and one outbound. The exact description of the conditions under which
one would act as client and the other as server (and viceversa) would be left to
the business process description language.

Admittedly, there are some issues with the exact meaning of a binding for an
outbound port, but they don't seem to be any worse than those already in
WSDL 1.1 wrt solicit-response and notification operations, and this hasn't
prevented other languages from building on WSDL 1.1.
</RRC>

Roberto

--
Roberto Chinnici
Java and XML Software
Sun Microsystems, Inc.


> </PY>
>
> Regards, Prasad
>
> > Roberto
> >
> > --
> > Roberto Chinnici
> > Java and XML Software
> > Sun Microsystems, Inc.
> >
> > Prasad Yendluri wrote:
> >
> > > We discussed this briefly during F2F last week also but to expand a little more
> > > on this, I see a need for both types of mechanisms. That is,
> > >
> > > 1) A solid event-notification mechanism (that is service initiated) that
> > > Sanjiva proposes as the replacement. This would however call for the
> > > specification of the complete details including the event subscription
> > > mechanism etc. as we had discussed in the f2f. This is to facilitate a server
> > > to generate events (subsequent to a request).
> > >
> > > 2) A way to describe things from a request initiator perspective. Revised and
> > > completed versions of Solict-Response and Notification that "fix" the abstract
> > > definitions and provide full details of the concrete definitions.
> > >
> > > I personally would be willing to go with the removal of these patterns if there
> > > is a clear way to capture both sides of an exchange (e.g. in a business
> > > process) say by an upper layer mechanism (analogous to WSFL or XLANG). However
> > > both seem to (XLANG I know for sure) rely on Solicit-Response / Notification
> > > type patterns to accomplish this. It is not clear to me how else this could be
> > > accomplished...
> > >
> > > Regards, Prasad
> > >
> > > -------- Original Message --------
> > > Subject: RE: issue: remove solicit-response and output-only operations?
> > > Resent-Date: Mon, 15 Apr 2002 20:19:41 -0400 (EDT)
> > > Resent-From: www-ws-desc@w3.org
> > > Date: Mon, 15 Apr 2002 17:18:28 -0700
> > > From: "Malu, Pallavi G" <pallavi.g.malu@intel.com>
> > > To: "'Sanjiva Weerawarana'" <sanjiva@watson.ibm.com>, www-ws-desc@w3c.org
> > > CC: "'pyendluri@webmethods.com'" <pyendluri@webMethods.com>
> > >
> > > Sanjiva,
> > >
> > > For representing Rosettanet PIPs we need solicit-response operations.
> > > e.g. PIP3A4 - is a PurchaseOrderRequest ans PurchaseOrderConfirmation
> > > scenario between the buyer and the seller.
> > >
> > > buyer:
> > >    <operation name="submitPO">
> > >      <output message="PORequest"/>
> > >      <input message="POResponse"/>
> > >   </operation>
> > >
> > > and corresponding seller:
> > >   <operation name="processPO">
> > >      <input message="PORequest"/>
> > >      <output message="POResponse"/>
> > >   </operation>
> > >
> > > So unless we have some first-class description of an event mechanism in
> > > place, I suggest we leave the "solicit-response" and "output-only" as is in
> > > WSDL1.2.
> > >
> > > -Pallavi
> > >
> > > -----Original Message-----
> > > From: Sanjiva Weerawarana [mailto:sanjiva@watson.ibm.com]
> > > Sent: Friday, April 12, 2002 9:14 PM
> > > To: www-ws-desc@w3c.org
> > > Subject: issue: remove solicit-response and output-only operations?
> > >
> > > The WG would like to solicit your comments on whether we should
> > > eliminate WSDL 1.1's "solicit-response" and "output-only"
> > > operations as we produce WSDL 1.2.
> > >
> > > Here are the two issues from the latest part1 document. Note that
> > > I have posted these together as the decisions obviously need to
> > > be coupled.
> > >
> > >   <issue id="issue-remove-solicit-response-operations" status="open">
> > >     <head>Should we remove solicit-response operations?</head>
> > >     Solicit-response operations are not fully defined in WSDL
> > >     1.1. There are multiple interpretations of these in the community:
> > >     event, callback etc.. Also, there is little evidence that anyone
> > >     is actually using them.  We could consider replacing this with
> > >     a first-class description of an event mechanism.
> > >     <source>Sanjiva Weerawarana</source>
> > >   </issue>
> > >
> > >   <issue id="issue-remove-notification-operations" status="open">
> > >     <head>Should we remove notification operations?</head>
> > >     Notification operations are also not fully defined in WSDL
> > >     1.1. There are multiple interpretations of these in the community:
> > >     event, callback etc.. Also, there is little evidence that anyone
> > >     is actually using them. We could consider replacing this with
> > >     a first-class description of an event mechanism.
> > >     <source>Sanjiva Weerawarana</source>
> > >   </issue>
> > >
> > > Thanks,
> > >
> > > Sanjiva.
Received on Tuesday, 16 April 2002 21:32:56 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:58:19 GMT