- From: Joseph Kesselman <keshlam@us.ibm.com>
- Date: Mon, 16 Jul 2001 11:56:31 -0400
- To: "'www-dom@w3.org'" <www-dom@w3.org>
The downsides of passing in a string: 1) You have to generate the string, then reparse it. Wasted work if the data was already in an easier-to-retrieve form. 2) You have to build a complete list; you can't defer resolution and compute it on demand. This too may impose additional costs. In use case B, for example, you may have to walk the context-node's parents to gather namespaces; a string requies you scan them all, whereas an API has the option of looking back only as far as necessary for this query, and looking farther at a later time if another query has not yet been resolved. Abstracting this into a NamespaceResolver API leaves us room to consider those issues. If someone wants to use a single string internally, that's fine. And perhaps we should consider a factory that accepts such a string and returns a NamespaceResolver with that content. But I think nailing it down as only a string would excessively constrain implementations. ______________________________________ Joe Kesselman / IBM Research
Received on Monday, 16 July 2001 11:57:05 UTC