RE: Choice of response message in WSDL

Tony,

As I understand it, your example is of a service that returns either B or 
C but not both in response to a given input, i.e. it is a request-response 
message exchange pattern.

The syntax you propose is used in WSDL 2.0 to describe a more complex 
message exchange pattern, i.e. one that returns two output messages. The 
complete syntax requires that you reference a MEP URI that decribes the 
MEP, and that each output message element also includes a messageLabel 
attribute to identify the role played by each of the output messages.

In fact you could achieve your desired result by defining a new MEP that 
has 1 input role and 2 output roles, but the MEP specification would state 
that the two output roles are mutually exclusive. I am not actually 
suggesting this as a solution, just an illustration of the semantics of 
multiple output elements in WSDL 2.0.

There is a difference between messages and faults. Listing multiple faults 
means that any of them may occur. The MEP specifies when they may occur. 
Typically, a single fault occurs. To be completely consistent, we could 
have required that only a single fault be specified and that its content 
be specified by a wrapper type if more then one "real" type occurs.

The motivation for this design comes from programming languages which 
allow you to list Exception classes in a throws clause, e.g.

public Response operation(Request x) throws Exception1, Exception2, ..., 
ExceptionN

Arthur Ryman,
Rational Desktop Tools Development

phone: +1-905-413-3077, TL 969-3077
assistant: +1-905-413-2411, TL 969-2411
fax: +1-905-413-4920, TL 969-4920
mobile: +1-416-939-5063, text: 4169395063@fido.ca
intranet: http://labweb.torolab.ibm.com/DRY6/



"Tony Fletcher" <tony.fletcher@choreology.com> 
Sent by: www-ws-desc-request@w3.org
01/26/2005 08:48 AM

To
Arthur Ryman/Toronto/IBM@IBMCA
cc
<www-ws-desc@w3.org>, <www-ws-desc-request@w3.org>
Subject
RE: Choice of response message in WSDL






Dear Arthur,
 
Thank you very much for your response.
 
I regard this solution as a viable workaround, but I still view it as a 
'workaround'.  I understand that this solution of having a single wrapper 
with internal divisions, as in your example below, works for both WSDL 1.1 
and proposed 2.0.  Loosing the wrapper in the binding is a new twist that 
could be useful.
 
However I was looking for a way of expressing, in the *abstract* 
description of the Web Service that there are two, or more, regular (i.e. 
not fault) messages allowed as a response to a given request message (so 
that messages (elements) B and C appear directly in the Description and 
there is no requirement for the artificial element BorC).  This does not 
seem to be duplicating schema at all, but just allowing one to state that 
this request message has these possible defined responses and these 
possible defined fault responses, all of which seems to me to be 
appropriate for a description of the abstract services.  I understand that 
you do already support this for fault message responses so extending to 
response messages seems a logical step (and a concluding one in that it 
does not open any floodgates).
 
However, if you have already discussed this at length and the case has 
been lost I shall peacefully retire (disappointed!).
 
PS  to be clear using your example I am looking for the following (or 
something that means the same) to be legal:
<interface name="Responder">
    <operation name="abc">
        <input message="tns:A"/> 
         <output message="tns:B"/>
         <output message="tns:C"/>
       <outfault name="fault" message="tns:F"/>
    </operation>
</interface> 

 and similarly:
<interface name="Requestor">
<operation name="abc">
        <output message="tns:A"/> 
         <input message="tns:B"/>
         <input message="tns:C"/>
       <infault name="fault" message="tns:F"/>
    </operation>
</interface> 
 
Best Regards,
Tony 


Tony Fletcher
Technical Advisor 
Choreology Ltd.
68, Lombard Street, London EC3V 9L J   UK
Phone: 
+44 (0) 1473 729537
Mobile: 
+44 (0) 7801 948219
Fax: 
+44 (0) 870 7390077
Web:
www.choreology.com
Cohesions?
Business transaction management software for application coordination
Work: tony.fletcher@choreology.com 
Home: amfletcher@iee.org
-----Original Message-----
From: Arthur Ryman [mailto:ryman@ca.ibm.com] 
Sent: 24 January 2005 20:28
To: Tony Fletcher
Cc: www-ws-desc@w3.org; www-ws-desc-request@w3.org
Subject: Re: Choice of response message in WSDL


Tony, 

The WSDL 2.0 WG discussed the message issue at length, and the conclusion 
was that if we added support for constructs like "choice" then we would we 
slowly but surely reproduce a lot of XSD. There were many requests for 
more control over the parts of a message. The solution we adopted was to 
discard the WSDL 1.1 <message> and <part> elements altogether and simply 
use XSD Global Element Declarations (GEDs) directly. 

The key point about this design is that the GED is viewed as an abstract 
message definition. The concrete details come in at the binding level. The 
binding rules are simple in the case where the concrete message is the 
same as the abstract message, e.g. for SOAP, the simple case is to use the 
GED as the <body> content of the SOAP <envelope> 

In your case, the idea would be to define a GED, BorC, whose content model 
was a choice of the two response messages: 

<schema xmlns="http://www.w3.org/2001/XMLSchema" targetNamespace="
http://example.com/choicy" xmlns:tns="http://example.com/choicy"> 
        <element name="B" type="string"></element> 

        <element name="C" type="string"></element> 

        <element name="BorC"> 
                <complexType> 
                        <choice> 
                                <element ref="tns:B"></element> 
                                <element ref="tns:C"></element> 
                        </choice> 
                </complexType> 
        </element> 

</schema> 

The WSDL 2.0 interface (aka portType) is: 

<interface name="Responder">
    <operation name="abc">
        <input message="tns:A"/> 
         <output message="tns:BorC"/>
       <outfault name="fault" message="tns:F"/>
    </operation>
</interface> 


This introduces a top level wrapper element <BorC> that you probably don't 
want in the concrete SOAP message. The solution is to modify the SOAP 
binding rules, i.e. to copy the content of the GED into the <body> rather 
than the entire GED including that unwanted root element. 

FYI, we have had a related request to allow more flexibility in our SOAP 
binding to permit muliptle children in the <body>. The binding rule: "copy 
the element content into the <body>" would work for your case and the 
multiple children case. 

Does this binding approach satisfy your requirement? 

Note that the fault issue is handled at the MEP level. We have several 
predefined MEPs already that specify how and when fault elements occur. 
See part 2. [1] 

[1] http://www.w3.org/TR/2004/WD-wsdl20-extensions-20040803/ 

Arthur Ryman,
Rational Desktop Tools Development

phone: +1-905-413-3077, TL 969-3077
assistant: +1-905-413-2411, TL 969-2411
fax: +1-905-413-4920, TL 969-4920
mobile: +1-416-939-5063, text: 4169395063@fido.ca
intranet: http://labweb.torolab.ibm.com/DRY6/ 


"Tony Fletcher" <tony_fletcher@btopenworld.com> 
Sent by: www-ws-desc-request@w3.org 
01/24/2005 01:30 PM 


To
<www-ws-desc@w3.org> 
cc

Subject
Choice of response message in WSDL








To the W3C Web Service Description Group, 
  
[Please copy me directly on responses as while I am on the WS-Choreography 
mailing lists I am not on the Web Service Description mailing lists - 
Thank you.] 


In the Last Call version of the WS-Choreography specification several 
exchange elements are allowed in an interaction element.  One is the 
request going in one direction and the others must be in the reverse 
direction.  Only one of these is allowed to be the 'normal' response 
message, all the others must be fault messages. 
  
The case I am particularly interested in seems to be supported by neither 
WS-Choreography at present nor WSDL 2.0  and I wonder if it should be. (I 
understand that WSDL 2.0 could support what I propose as an extension, 
though I make  this comment to the WSD group with the aim of making it a 
standardised feature.) 
  
 Suppose I have request - response protocol pair but there can be several 
distinct response messages.  So I want to say the request message is A and 
the response is B or C (or possibly fault message X or Fault message Y). 
  
I realise that of course you can write it as five (in this case) one way 
interactions, but that looses the request response semantic.  You could 
also re-write the protocol to only use a single response message and 
internally to the response message have different parameter values that 
give the semantics of B or C - and likewise one can re-write the Fault 
message to combine X and Y, but why should one have to change the protocol 
to suit WSDL? 
  
 In WS-Choreography I would like to be able to write, for example, 
something like: 
<interaction name="ABCF" channelVariable="tns:aChannel" operation="a"> 
      <participate relationshipType="SuperiorInferior" 
fromRole="tns:Superior" toRole="Inferior"/> 
      <exchange name="A" informationType="Atype" action="request"> 
              <send variable="tns:A"/> 
              <receive variable="tns:A"/> 
       </exchange> 
       <exchange name="B" informationType="BType" action="respond"> 
              <send variable="tns:B"/> 
              <receive variable="tns:B"/> 
        </exchange> 
        <exchange name="C" informationType="CType" action="respond"> 
              <send variable="tns:C"/> 
              <receive variable="tns:C"/> 
        </exchange> 
        <exchange name="F" informationType="FType" action="respond"> 
              <send variable="tns:F" causeException="true"/> 
              <receive variable="tns:F" causeException="true"/> 
        </exchange> 
</interaction> 
  
and in the corresponding Web Service description  I would like to be able 
to write something like: 
<portType name="Requester">
    <operation name="abc">
         <output message="tns:A"/>
         <input message="tns:B"/>
         <input message="tns:C"/>
         <fault name="fault" message="tns:F"/>
    </operation>
</portType>
<portType name="Responder">
    <operation name="abc">
         <input message="tns:A"/>
         <output message="tns:B"/>
         <output message="tns:C"/>
         <fault name="fault" message="tns:F"/>
    </operation>
</portType> 
  
or with explicit choice construct: 
<portType name="Requester">
    <operation name="abc">
         <output message="tns:A"/>
         <choice> 
              <input message="tns:B"/>
             <input message="tns:C"/> 
         </choice>
         <fault name="fault" message="tns:F"/>
    </operation>
</portType>
<portType name="Responder">
    <operation name="abc">
         <input message="tns:A"/>
         <choice> 
              <output message="tns:B"/>
             <output message="tns:C"/>
          </choice>
       <fault name="fault" message="tns:F"/>
    </operation>
</portType> 
  
 I would be quite happy to have either some sort of explicit 'choice' 
construct around the multiple responds that are regular permitted 
responses and therefore do not have cause exception set, or an implicit 
choice as we currently have for multiple exception causing responses. 
 
Best Regards, 
Tony 


Tony Fletcher 
Technical Advisor 
Choreology Ltd.
68, Lombard Street, London EC3V 9L J   UK 
Phone:   
+44 (0) 1473 729537 
Mobile: 
+44 (0) 7801 948219 
Fax:     
+44 (0) 870 7390077 
Web: 
www.choreology.com 
Cohesions? 
Business transaction management software for application coordination 
Work: tony.fletcher@choreology.com 
Home: amfletcher@iee.org