RE: Minority objection to requiring unique GEDs or required feature to distinguish operations

Hi Jacek!

unfortunately as Amy has pointed out, there is there nothing in the 
status quo to prevent a service from declaring that the dispatching 
mechanism is "out of band", "random" or even "undisclosed" by simply
conjuring up an appropriate URI.

the outcome of all this is that the consumer of a web service will 
indeed have to treat operations that accept the same input as a 
single event driven operation (for non-RPC styles) and i think 
that is the architectural goal of some members of this WG.

where i'm still unclear is why that scenario wouldn't be better 
modelled, as you outlined, by a single operation returning a choice 
of the possible response messages - in my mind all this just a 
work-round for operation overloading:

    foo('piff') returns 'paff'
    bar('piff') returns 'poff'

sending 'piff' to 'foo' could return 'poff', so
foo and bar are the same operation!

Paul


-----Original Message-----
From: Jacek Kopecky [mailto:jacek.kopecky@deri.org]
Sent: 12 August 2004 10:02
To: Downey,PS,Paul,XSJ67A C
Cc: WS-Description WG
Subject: RE: Minority objection to requiring unique GEDs or required
feature to distinguish operations


Paul, objecting minority,

The problem I see with dispatching upon policy is that the client has no
idea which of the multiple operations with the same on the wire message
contents is in effect and also which of the potential different result
messages it can expect back (this can be appropriately modified for
non-req-resp patterns).

As a client, I'd like to be sure that it's operation A (and none
different) that is in effect, especially if I choose to ascribe
different semantics to different operations at the endpoint. Of course
one could view the situation that from the initiator's point of view,
both (or all) ambiguous operations could be viewed as a single one with
an out-of-bound switch between the various result formats, but then I
believe it should also be modeled as a single WSDL operation.

On this issue, I agree with the status quo that describing the dispatch
mechanism is good and mandatory.

Jacek

On Thu, 2004-08-12 at 10:33, paul.downey@bt.com wrote:
> Mark wrote:
> 
> > I'm confused.  There's always an operation, no?  
> > Sometimes what's in the message isn't the "name" per se, 
> > but there's always some token in the message envelope 
> > that can be mapped to some spec which describes the
> > operation's semantics.  
> 
> my understanding of this issue is that it is possible to 
> dispatch messages to two different operations that accept 
> exactly the same message contents based upon something 
> completely out of band, e.g. i've credit left in my account, 
> it's raining, we're now at war. That's dispatching based
> upon /policy/ rather than /description/.
> 
> Paul
> 
> 
>

Received on Thursday, 12 August 2004 12:10:14 UTC