- From: <Noah_Mendelsohn@lotus.com>
- Date: Sat, 14 Jul 2001 22:22:22 -0400
- To: Martin Duerst <duerst@w3.org>
- Cc: "'Julian Reschke'" <julian.reschke@gmx.de>, Michael Brennan <Michael_Brennan@Allegis.com>, Pradyumna_Siddhartha <Pradyumna_Siddhartha@infy.com>, xmlschema-dev@w3.org
This is drifting into a rehash of the oft-visited question of what, if anything should be at "the end" of a namespace URI reference. I suspect we don't want to reopen the whole debate again, but for what it's worth, one of the main problems that was discussed when we considered the issue during design of schemas had to do with the timeouts that can occur when a namespace does not, in fact, have a schema at all. If you require that schema-aware processors always attempt to dereference a NS URI to get anything (RDDL, schema, etc.), then if there is any chance that such a resource is not there, most network systems incur quite a delay. This problem was reported by a vendor who had operational experience with deployed systems that used earlier schema systems and a convention of derferenceing the URI; they found the performance problems and timeouts to be unacceptable. Of course, in situations where you are reasonably sure that a resource will be available (cached or otherwise), doing a dereference to get it can be a wonderful strategy. Those are among the reaons why we (a) decided to build an architecture that would support the development of conventions, such as RDDL, for locating schemas from a NS URI but (b) stopped short of requiring that processors attempt the dereference. Also, there are situations in which the application consuming a document cannot trust the instance author to supply an appropriate schema. For these circumstances, processors can be built to accept schemas from the consuming application or other sources -- of course, it is possible also to build processors that derefence the NS URI, check the schema retrieved to make sure it is indeed the one intended, and thus ensure that producer and consumer had a consistent presumption regarding the schema. The schema recommendation does not standardize such behavior, but does not preclude its implementation either in particular processors, or in some layered standard to be issued in the future. Such a standard could mandate, for example, the use of RDDL. So, while we know from past debates that not everyone agrees on what is desireable or whether the decisions made by the schema WG were indeed what one person or another would have preferred, I think the above does summarize some of the issues we considered and the intended behavior of the schema recommendation. Hope this is helpful in the current discussion of WSDL. Thanks. ------------------------------------------------------------------------ Noah Mendelsohn Voice: 1-617-693-4036 Lotus Development Corp. Fax: 1-617-693-8676 One Rogers Street Cambridge, MA 02142 ------------------------------------------------------------------------
Received on Saturday, 14 July 2001 22:24:49 UTC