- From: <keshlam@us.ibm.com>
- Date: Fri, 23 Jun 2000 14:03:16 -0400
- To: xml-uri@w3.org
For what it's worth:. I need to study it further, but I _think_ the DOM can in fact get away with saying "We just store and manipulate whatever string you tell us is the Namespace Identity. If there's processing that has to be done to derive that identity from the value in the namespace declaration attribute, that has to be done before you hit the DOM." It looks like this can support all of the primary proposals. It matches the Literal and Forbid modes directly. It's nowhere near as pretty in the Absolutize mode, but it's usable, and seems to allow the DOM to escape from its dependency on the ongoing debate... and convenience functions could be added later, if we _do_ go with Absolutize, to encapsulate some of the sharper corners. Absolutization would push significantly more work into the parser and/or application under this model. It also pushes a noticable amount of additional work into the namespace reconciliation process which may be needed before saving a DOM's contents as XML syntax. We already had to consider the case where a node's declared NSURI and prefix didn't match the declaration of that prefix in scope at this node (or where there was no declaration attribute for this prefix, or where the prefix must be deduced or synthesized). If absolutization is being used, best results require that that users _DO_ explicitly manage prefixes (so that if there are two distinct NS's in scope which absolutized to the same URI, you can tell which one the user intended to bind to --- they may _not_ yield the same URI when absolutized in a different context, eg if one is relative and the other isn't) and forces the comparisons to consider absolutization as well. Note that this seems to be workable only because the DOM has separate storage for the namespace declaration attribute, the namespace name actually bound to an individual Element or Attr, and the prefix which may (or may not) point from the latter to the former. I don't know enough about how the Infoset is thinking about namespaces to have any sense of whether it could get away with a similar solution for Absolutize. I still think Absolutize is the wrong answer, for reasons of: * computational and coding overhead (as mentioned above), * conceptual overhead (I really doubt that a novice user will understand it, given how many of the experienced folks on this discussion have gotten confused), * conceptual clash with the original intent of namespaces (relative namespace declaration can be defined in such a way that it is architecturally consistant,, but does not mesh well with the puposes namespaces were designed to serve), * and fairly weak motivation (the arguments seem to be either "we might want it in the future, if we can ever agree on what it means and how to use it" or "we don't know that it isn't in use and we're reluctant to admit to users that the W3C can make mistakes"). ______________________________________ Joe Kesselman / IBM Research
Received on Friday, 23 June 2000 14:03:31 UTC