RE: Section 5 vs Schema

Don Box suggests:

>> I believe that one could express all section 5-isms (typed
>> references, arrays) using idioms currently supported in schema + a
>> handful of NS-qualified attributes.

Yes, one can certainly do that, but I don't think it addresses the 
question of a schema language for the resulting graph model.   It seems to 
me that a good schema language captures equivalence classes as well as 
validity.  So the XML schema language can tell you that there's a sense in 
which the following are equivalent:

        <a xsi:type="xsd:integer">3</a>

and


        <a xsi:type="xsd:integer">003</a>

In the case of the encodings, I think you want a schema language that 
treats the following as equivalent:


        <b id="someid">hello</b>
        <a href="someid">

and 
 
        <b id="someotherid">hello</b>
        <a href="someotherid">

and that understands that the graph edge named "a" is equivalent not only 
in the two examples above, but also for:

        <a>hello</a>

In other words, these are all three cases in which the graph has an 
inbound edge named "a" with a value of "hello".  I think that XML schema 
is the wrong level of description language in which to capture that 
equivalance. 

I would expect a description language for SOAP encoding to have primitives 
like "struct", "array", and perhaps some notion of typed node or edge (as 
opposed to element.)  In short, I think that schema is constraining the 
serialization of the graph, not the graph itself, and that seems 
problematic to me. 


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

Received on Thursday, 31 January 2002 16:45:22 UTC