- From: Bullard, Claude L (Len) <clbullar@ingr.com>
- Date: Thu, 9 Jan 2003 14:36:27 -0600
- To: "'noah_mendelsohn@us.ibm.com'" <noah_mendelsohn@us.ibm.com>, Tim Bray <tbray@textuality.com>
- Cc: Chris Lilley <chris@w3.org>, www-tag@w3.org
That's ok but a little over the top because it is easy to construe it as DTD/Schemas always hurt performance and should be avoided. That would be unacceptable as well. I don't think that is what you mean. SOAP is engineered to ensure a DTD or Schema is never needed. No demand. That's good. A more complex document with more opportunities for errors can create demand and it should be met with consistent, easy to apply features for validation. Otherwise, Goldfarb's fear that message oriented XML requirements will overwhelm document oriented XML requirements will become reality. The result would be abandoning XML for document technologies. That is a bad outcome. I'm not convinced we can do nothing. The position that some metainformation needs to be conveyed by simpler means, eg, that small set of decorations an XML processor needs to do ubiquitous tasks, is compelling. The XML namespace is the right place to do it. The DTD/Schema are the wrong place I am finally convinced. :-) When we did XML 1.0, we were in a chopshop mode of operations. That is not an insult; it was speed to market in a time when the survival of generalized markup was in doubt. We now have the time to reconsider at leisure, the parts we stacked over in the corner. I don't think we can go forward with this without seriously considering what XML SW has to offer. But be warned; this is not a way to get less conformance levels. We will have more. That may be ok. In SGML in the mid nineties, we talked about the Unicorn tests because it was self-evident that we lacked precisely this kind of modularization in SGML. It worked for the authors but the implementations were too expensive and too rare. The same thing happened in VRML97/2.0. So modularization per se doesn't bother me. Divesting a class of users of the tools they need would. If this isn't the time, ok. If it is, then we have the time this time to do it right. len From: noah_mendelsohn@us.ibm.com [mailto:noah_mendelsohn@us.ibm.com] Tim Bray writes: > Bullard, Claude L (Len) wrote: > > Someone please explain why this isn't a realistic > > option given an on-demand scenario. If the DTD/Schema > > were ALWAYS processed, I agree that is a bad thing. > > The one thing that jumps out at me is that requiring > > a DTD or Schema just to get at ID declarations is > > a heavy requirement given that the number of > > applications that use IDs is large, so this > > can in fact, become a virtual requirement to > > always get the DTD/Schema. > > > > Is that it? > > Pretty well. I lot of apps are just *not* gonna do it; but they still > might like to know about IDs. -Tim I'm in complete agreement with Tim. SOAP is an example of a system that uses xsd:ID [1] typed attributes and which goes to some lengths to NOT >require< validation of any sort, though partial or full validation of the message is permitted if useful to the consuming application. [2,3]. SOAP's ID attributes are in its own namespace. IMO, DTD or schema retrieval, as well as validation, is often impractical for performance and security reasons, among others. Interestingly, among the many performance risks we analysed in the schema WG were reports from implementors that failures to retrieve (timeouts) had bigger performance impact than successes. Furthermore, tf validation is required, then the document becomes useless in the case where an external DTD or schema is for some reason unavailable. I think this is unacceptable. I think I agree with Tim's other conclusion: do nothing is probably the least risky solution. We've got too many typing mechanisms already.
Received on Thursday, 9 January 2003 15:37:02 UTC