RE: MEPs: Hardcoded or not? (was: Re: Minutes of teleconference02-05-23)

I agree with Jacek that SOAP MEPs are
different from WSDL "MEPs".

But I do not think calling the WSDL patterns
"MEPs" contributes to clean boundaries
between web service description layers.

Consider the "RequestResponse" label used in WSDL
to refer to a single-connection, synchronous,
half-duplex ("one-side-reads-and-waits-to-write-until
other-is-done-writing").
This XML messaging pattern does not correctly 
capture what is semantically essential
to the RequestResponse MEP.
Instead, it includes accidental features of
one flavor of "binding." A RequestResponse MEP can
just as well make use of two TCP
connections, sequenced so that
one follows the other, with only
one input defined on each node, and
each with a URL for connecting. RequestResponse
can be done either way, so why confuse
the audience by suggesting that
the WSDL operation with an input,
output, and fault _is_ RequestResponse?

To avoid related confusions, I support
deprecating the "Notification", "One-Way"
"SolicitResponse" and "RequestResponse"
terminology, which drags the [MEP|choreography|flow|fsa|etc]
layer just inside WSDL. The definitions of types, messages,parts,
operations, port-types, etc form a clump of functionality.
Adding the other operation group labels 
might have had some initial
heuristic value, but now should be handled 
in greater depth elsewhere. 

[BTW, I also don't think the SOAP group 
should necessarily be the one that
takes on the MEP analysis problem. MEPs
involving concepts like forks, joins,
guard conditions, and so on--it
should be dealt with by some other group.
Maybe the architecture group first.]

-----Original Message-----
From: Jacek Kopecky [mailto:jacek@systinet.com]
Sent: Friday, May 31, 2002 3:39 AM
To: Jean-Jacques Moreau
Cc: www-ws-desc@w3.org; Glen Daniels
Subject: Re: MEPs: Hardcoded or not? (was: Re: Minutes of
teleconference02-05-23)



 Jean-Jacques, 
 my point was that SOAP MEPs are a different kind of beast from 
WSDL MEPs, like apples and oranges.
 In WSDL, MEPs are built from the point of view of one node -
what messages come and go through it.
 In SOAP, MEPs are built from the point of view of the message 
exchange - what message(s) go through what nodes.
 For instance the Simple Request Response can be translated into
two (not necessarily different) MEPs in WSDL because there are
two nodes involved. A generic SOAP MEP will generate one or 
multiple WSDL MEPs (or even multiple usages of one WSDL MEP on 
one node). 
 Let's take a hypothetical Circular Path SOAP MEP where node A
sends a message to node B, that sends a message to node C and
that sends a message to node A. In WSDL, this probably maps to
notification followed by one-way for nodes B and C, and to 
request with an independent response (probably just 
request/response only with different binding information) for 
node A.
 So I still think we have a finite set of MEPs in WSDL and that 
it is limited to (multi)request/(multi)response and one-way.
 Best regards,

                   Jacek Kopecky

                   Senior Architect, Systinet Corporation
                   http://www.systinet.com/



On Thu, 30 May 2002, Jean-Jacques Moreau wrote:

 > 
 > So far, the only two SOAP MEPs we have (well, one and a half, really)
 > are between two nodes only. IMO, we should be able to model such
simple
 > MEPs in WSDL.
 > 
 > Jean-Jacques.
 > 
 > Jacek Kopecky wrote:
 > 
 > >  Jean-Jacques,
 > >  IMO in WSDL the term MEP is a bit different from SOAP MEP. The
 > > difference is that in SOAP an MEP may span multiple nodes and is
 > > defined from the point of view of a message, whereas WSDL
 > > describes one node and all MEPs used in WSDL must be defined from
 > > the point of view of that one described node.
 > >  In WSDL, other MEPs than one-way and request/response should
 > > IMHO be viewed as orchestration, out of scope of WSDL. Therefore
 > > we can hardcode these two.
 > >  Best regards,
 > >
 > >                    Jacek Kopecky
 > >
 > >                    Senior Architect, Systinet Corporation
 > >                    http://www.systinet.com/
 > >
 > > On Fri, 24 May 2002, Jean-Jacques Moreau wrote:
 > >
 > >  > I was not able to attend yesterday, and I apologize in advance
if
 > >  > I am reiterating a discussion that has occured already, but I
 > >  > wanted to point out that I agree with Glen and that, if
possible,
 > >  > MEPs should not be hardcoded into the spec.
 > >  >
 > >  > Specifically, SOAP 1.2 currently defines one MEP. It is expected
 > >  > that specification for other MEPs will be produced in the
future.
 > >  > I think it would be desirable that these other MEPs can be
 > >  > described in WSDL, otherwise there will be services out there
 > >  > that cannot be described (and hence used).  It would be
desirable
 > >  > if these other MEPs can be described without us reopening the
 > >  > whole spec every time.
 > >  >
 > >  > Thank you,
 > >  >
 > >  > Jean-Jacques.
 > >  >
 >

Received on Friday, 31 May 2002 17:12:00 UTC