- From: Francisco Curbera <curbera@us.ibm.com>
- Date: Thu, 9 Dec 2004 16:45:42 -0500
- To: Mark Baker <distobj@acm.org>
- Cc: public-ws-addressing@w3.org, public-ws-addressing-request@w3.org
Hi Mark, It is a pity you missed the discussion we had yesterday at the f2f in Redmond. The case was made (by Jeff M.) that to have identifier semantics you need to provide a mechanism that will allow you to decide if two such identifiers identify the same entity or not; that is a Boolean valued function essentially which will return true or false given a pair of those identifiers. This is old news for those familiar with distributed system design of course. The spec does not have such a thing because it follows the semantics you agreed with below: it allows you to return "true" (in fact, only wrt the metadata that applies to the interaction, but let's not get into that one now) if you get a byte per byte match of selected fields, but you can never get a false answer. This is why EPRs do not have identifier semantics. Paco Mark Baker <distobj@acm.org> To: Francisco Curbera/Watson/IBM@IBMUS Sent by: cc: public-ws-addressing@w3.org public-ws-addressing-req Subject: Re: Are EPRs identifiers? uest@w3.org 12/09/2004 04:31 PM Paco, On Thu, Dec 09, 2004 at 10:01:30AM -0500, Francisco Curbera wrote: > So I think by now everyone agrees on this: when the address or reference > props. are different then: > > - the metadata may or may not be different, > > - they EPRs may or may not "point to/reference" different > entities/endpoints/resources/thingies/whatever. > > That is, out of a negative byte by byte comparison you cannot infer > anything at all (so you may be better off retrieving fresh copy of the > metadata, for example.) I think we should at least make this crystal clear > in the spec, since it has taken a bunch of smart guys like us a while to > figure it out. +1 > If after this realization the group still wants to call EPRs identifiers > then need to understand we will be endorsing a very ad-hoc notion of > "identifier". This would be a gross mistake IMO: attaching new meaning to > concepts already in use, can only result in confusion - of the general > public as well much as the TAG itself. Whoa, really? How does the above justify such a conclusion? I can't see it. If you're assuming that identifiers don't demonstrate those characteristics you describe, I think you're mistaken, as the evidence shows otherwise. Phone numbers - nope, as a business or household can have multiple numbers, even multiple numbers in different area codes. URIs; nope, clearly. CORBA IORs, nope (I still remember those discussions circa 1994). I don't doubt that examples exist, but I think they're few and far between, and it seems to me that the larger the system, the less chance there is of being able to enforce such a rigid rule. FWIW, I think EPRs with RefParams are *not* identifiers, but those without them are. Mark. -- Mark Baker. Ottawa, Ontario, CANADA. http://www.markbaker.ca
Received on Thursday, 9 December 2004 21:46:18 UTC