- From: David Brownell <david-b@pacbell.net>
- Date: Fri, 26 May 2000 12:16:04 +0100
- To: "Steve DeRose" <Steven_DeRose@brown.edu>, <xml-editor@w3.org>, <xml-dev@xml.org>
> I do not see any reason to rule out fragment identifiers in system identifiers. > There are lots of potential uses for them. Consider: Compatibility is such a reason. Removing that contraint would create a non-interoperable class of "XML 1.0" (not!) documents: > * Grabbing a piece of an XML document to embed in another (a whole object > that is text/xml or application/xml, cannot generally be referenced as a > non-NDATA entity (since it includes the DOCTYPE, at least if it's valid). Making XML 1.0 dependent on something analagous to XPath, which creates a circular dependency (yeech). > * A URI could point to a zip or tar archive, and the fragment identifier > may specify a particular XMl file out of the archive. Having that level of intelligence about URI processing seems better suited to something like XInclude. > * A URI could point to a big XML document that serves only to collect a lot > of modular fragments for re-use: such as the tables of FAA-mandated warning > text used in aircraft manuals. Or, each warning could have a unique URI. Makes it a lot easier to maintain the warnings, under many development models, as well as to process them (just read one little file, not a monstrous compendium). > * system identifiers can be used for lots of other things, like DTDs (later > presumably schemas), and data in all kinds of notations. XML 1.0 doesn't address schemas, they're out of scope. I could see notations having fragment IDs though, since they're never interpreted directly by the XML processor (application issue). > What motivation could there be for absolutely prohibiting fragment > identifiers? More to the point, what motivation _was_ there? I think simplicity (for external entities) is a big win. At this point, removing this constraint seems to me quite unwise. > It seems to me it's none of XML's business what the syntax of > URI references is, or whether fragment identifiers are needed. What of a > media type which defines the fragment identifier (they are media type > specific, after all) in such a way that it ends up being *required* for > proper interpretation None of the text/xml or application/xml media types is so defined. So that argument doesn't work with SYSTEM identifiers used for external entities. - Dave
Received on Friday, 26 May 2000 07:15:01 UTC