RE: Correlation.MessageRef [was AM corrections]

Hi Jean-Jacques,

> -----Original Message-----
> From: Jean-Jacques Moreau [mailto:moreau@crf.canon.fr]
> Sent: 03 April 2001 13:45
> To: Williams Stuart
> Cc: xml-dist-app@w3.org; Mark Jones; chris.ferris@east.sun.com
> Subject: Re: Correlation.MessageRef [was AM corrections]
> 
> Hi Stuart,
> 
> I think both Chris Ferris and I are somewhat unhappy with
Correlation.MessageRef
> being a local/relative reference.

I'm not sure I detected unhappyness in Chris's comments... but maybe that's
just optimisim on my part :-).

> IMO, the value of MessageRef (which I assume is something that would be
embedded
> in the message itself/binding) should not change as the message progresses
> through intermediaries.

The Correlation (or Causality) concept in the AM document does *not* make
the assumption that it leverages off of some syntactic element in the
message - I wanted to avoid the AM implying that such a syntactic element
was a required in the design of XMLP. It is certainly an option that the
design of XMLP may wish to consider and would give rise to a more globally
significant means of referencing a message.

However, the important thing that I believe I was trying to capture is the
potential for such correlations (or causalities) to be leveraged off of the
*behavioural* elements of underlying protocols (as in SOAP 1.1 over HTTP) as
well as possibly syntactic elements in definition of bindings to underlying
protocols and possibly syntactic elements within the definition of the core
of XMLP itself.

> It makes it hard otherwise to track messages, or to refer to messages one
has 
> seen in the past.
> 
> For example, a given intermediary might send an ACK every 5 messages, and
list
> the corresponding MessageRef's. How would the sender know which messages
are
> being ACKed, if it used a different numbering scheme than the
intermediary?

Firstly, if I were going to be 'obtuse' I'd ask what is this 'ACK'ing
business - is that application semantic? Are these 'ACK' messages messages
that the application sees or are they something that happens within the XMLP
layer. If the former, then its an application semantic and 'opaque' to XMLP,
if the latter it happens within the XMLP layer and would not be evident at
the interface at the top of XMLP.

Slightly more generously, I might try to adapt you question a bit to a
situation where a recipient generated a message in response to very 5th
message carrying a list of references to each of the 5 messages it has
received. In that case I would agree that Correlation as defined in the AM
doesn't cover this. However, I would also argue that you have constructed an
application semantic here and that you would have to generate more globally
significant message identifiers and the syntactic constructs (blocks) to
label the messages and denote the ACK list.

I think that you presenting a problem that Correlation.MessageRef was not
intended to solve.

> 
> Jean-Jacques.

Regards

Stuart

<snip/>

Received on Tuesday, 3 April 2001 10:31:38 UTC