RE: MEP and Operations

Comments interspersed.

FABLET Youenn writes:
"A few words about MEP and operations.

"Let's have an operation that takes as input a date and a country name. 
As output, you get the size of the population of this country at that
date.
IMO, the definition of this operation  should not change whether:
    - you receive the date and country name as one message or two 
separate messages
    - you send the output to the requester A, to another node B or both 
A and B."

Dale> +1 to that!

FY> The relationship between MEP and operations is somehow confusing
because 
of the use of the term "message"  in the operation definition.
At now we have a syntax like:
        <input message="dateAndCountryName"/>.
I would much prefer to have an operation defined with a different term 
like "dataset" for instance,
        <input dataset="dateAndCountryName"/>
or allow to have multiple input elements in an operation:
         <input name="date" type="date_type"/>
         <input name="countryName" type="countryName_type"/>

Dale> Hmmh +0.5. I guess I was already reading  "message" as really
amounting to "dataset". Since real messages can often be individuated
by things like Message-ids, and have a lot of meta information
in headers and what not, they are not what wsdl:message defines
anyway. Real messages are still probably not quite the "bits on the
wire"
but are not far removed. So the element tag is misleading. Does
that matter? (Deducting .5 for the combination of previous uncertainty 
combined with empathy for existing early implementers.) 

FY> An operation can then be mapped to one or more MEP (Get, 
Request-Response, Request-ResponseAndForward...). It will define the 
"real" messages to be received/sent and what you will find in each
message.

Dale> +1

FY>Each (operation+MEP) can then be related to one or more protocols
(wire 
format).

Dale> Are SOAP participants still utilizing the TMEP concept (and
clarifying
it if they are)?

FY> IMO, we should be able to define in WSDL the MEP we want to use for
a 
particular operation.
Because MEP are protocol-generic, we should uncouple them from
protocols.
Uncouple operations from MEP is also a good thing because an operation 
should not have different behaviours whether implemented with MEP A or 
MEP B.

Dale> Possibly +1 here also. Should we "define" in WSDL or instead
"refer"
or "declare" within WSDL, with MEPs defined elsewhere? I assume wherever
this
occurs it would be toward the binding side of things, right? Bindings
then
are encodings plus MEP(s) plus transfer protocol details 
plus eventually other stuff? 

Comments welcome,

   Youenn

Received on Thursday, 13 June 2002 13:04:48 UTC