- From: Sanjiva Weerawarana <sanjiva@watson.ibm.com>
- Date: Wed, 1 Oct 2003 01:41:15 +0600
- To: "Amelia A. Lewis" <alewis@tibco.com>
- Cc: <jmarsh@microsoft.com>, <www-ws-desc@w3.org>
Hi Amy, > > Clearly I haven't understood this pattern stuff at all. I wonder > > how users will deal with it. > > The multicast solicit-response pattern was "rolled into" the > solicit-response pattern on the basis of the language that was > introduced to say that, basically, we only reflect information into a > pattern description if it is important to more than one participant. > The service cares that it is sending a message multicast. Each of the > receivers doesn't care; the rules that they follow are to send [no > response] or [one response] or [one fault]. Multicast solicit-response > does *not*, in fact, really work terribly well with the existing > output-input pattern; it would be better to change it by saying 1) it > uses message-triggers-fault, and 2) the input message is optional. > > If things are truly so confusing, then I will propose adoption of a > pattern that uses the above modifications, since *that's* what I'm after > (well, that and a notification pattern with message-triggers-fault). In > pub/sub, the *publisher* is the service, and output-first patterns are > the norm. +1 to adding this pattern. > Neither I nor anyone at TIBCO has ever advocated the serial unicast > "multi" patterns (which I, personally, thought of as simply clouding the > issues, but I probably shouldn't cast nasturtiums). Apparently they > confused lots of people, who associated them with the arguments I've > made for multicast-capable patterns. They are completely unrelated. > They are *not* my fault. That we have removed confusion, by removing > the serial unicast patterns (unless someone else steps forward to offer > defense), is probably an advantage to the readers of the spec, who won't So it looks like it was Roberto playing mind games with us ;-). I think he's running as far away from them as possible .. so I doubt anyone will defend 'em any more. Let's nuke 'em! > have all the backstory about multicast and the rest of it. I do not > regard this decrease of confusion as a good reason for removing the > material that *does* support multicast. I'm not proposing that any more. At one point I was advocating removing all patterns that we're not defining normative bindings for, but now that we are not following that rule I'm for defining all the patterns that have well-defined semantics and that someone deeply cares about. Since adopting these patterns is relatively low-cost (giving a URI and proper specese just to define the pattern) across the plethora of specs we have, I fully support adding the multicast patterns as well. Sanjiva.
Received on Tuesday, 30 September 2003 15:41:40 UTC