Re: Potential new issue: PSVI considered harmful

Jonathan Borden writes:

>> Agreed. This underscores why an umambiguous 
>> mapping of type names e.g. QNames to URI 
>> references is so important. This mapping is 
>> perhaps the only reasonable way a type 
>> indicated in such an XML instance document can be
>> connected to the schema fragment that defines the type.

Agreed, but with some need for care in deciding what is being referenced 
by those URI references. 

Note that, at least in the case of W3C XML Schemas, not all schemas need 
exist in the form of schema documents.  As we've said repeatedly, the 
recommendation explicitly supports the case where you invent an API for 
creating schemas; use that API to establish declarations, types, etc,; 
validate using that "in memory" schema.   The schema need not ever exist 
in a schema document --- or, the base types may be in a documents but not 
the subtypes -- or the other way around.   Mix n' match according to your 
needs.   These should be moderately common cases in server-side scenarios 
when schemas are synthesized dynamically to describe data from non-XML 
databases, etc. 

W3C XML schema provides that the QNames provide uniform reference to all 
components, whether dynamic (as above) or not.  Indeed, even in the case 
of a declaration that came from documents,QNames reference the abstract 
result.  For example:

        Schema doc 1: defines type A
        Schema doc 2: defines subtype B
        Schema doc 3: defines subsubtype C in namespace NS

A QName reference to "NS:C" is a reference to the net, effective type 
resulting from the combination of the constraints.  It is not a reference 
to the serialized definition of C in a schema document.   Indeed, were one 
to switch to a different version of document 1, the effective definition 
of C might change, even though doc 3 is untouched.

The need to put everything of interest into URI space is well-understood. 
The Schema WG has been for sometime working on approaches that will 
achieve this.  It is important to cover all the interesting cases, and not 
to fall into the trap of letting a serialization of a definition be 
confused with the definition itself.

>> Perhaps each language specific "PSVI" ought derive 
>> from a general "TAI" which is what we ought focus on.

Exactly the direction I was suggesting one one would consider.   Thank 

Noah Mendelsohn                              Voice: 1-617-693-4036
IBM Corporation                                Fax: 1-617-693-8676
One Rogers Street
Cambridge, MA 02142

Received on Friday, 14 June 2002 10:10:04 UTC