Re: [RSS-DEV] Prism and RSS 1.0

Ron Daniel wrote:
> ... thinking about
> forbidding use of parseType when the value is anything other
> than 'Literal', thus requiring all the resources to be typed.
> (The assumption here is that by using the typed node abbreviation,
> and forcing alternating nested elements to follow the node, arc,
> node, arc, ... pattern, special-purpose parsers will be easier
> to implement. To be honest, this comes from my personal
> preferences and intuition, not from a serious examination of
> many different descriptions and attempts to implement dedicated
> parsers. So I'd be VERY interested in people's opinions on
> that notion.)

Ron - 

This XML "striping" appears a lot.  

Another place I have seen it is in deriving XML instance 
documents constrained by UML object models, where alternate 
XML elements are used for class-names and for object-labels 
(accessors, attribute labels, association role-names, whatever).  
I really don't like this because it is using the same 
syntax device to perform two very different semantic tasks. 
Pretty much the same problem as what you describe above.  

(Sometimes a lexical trick is used to provide a hint 
- e.g. class names start Upper case while object names 
start l.c. - but this is an obvious kludge.)   

From a syntax point-of-view, the problem appears to be 
that XML has a single tool - element nesting - while 
graphical languages like UML and RDF have a lot more 
(and in the case of UML an explicit recognition of the 
different meta-levels).  

From a conceptual point-of-view, I think that the problem is 
that an XML document can only express first-order-logic directly, 
so we have to cheat to capture higher-order ideas?  
(I'm getting way out of my depth here ...)

But there are some solutions.  
For example, you can move the information from the higher 
meta-level into another document.  
Using XSDL this is done by specifing the classes (types) to 
which the objects (elements) belong in the schema document, 
so the instance document can use the objects without repeating 
the information about what class (type, content-model) they are from.  
And the xsi:type attribute from XSDL provides a late-binding version when that is needed.  

But does separating the information into more than one document 
violate the design-goal that XML documents are self-describing? 
I don't think so - it is just that the higher meta-level 
information is provided by reference rather than by value.  

Do RDF + RDF/S work the same way?  
Can the "type" information be moved away from the value and relation information?

--    CSIRO Exploration & Mining
T:+61(8) 9284 8443 F:+61(8) 9389 1906 M:0403 302 672

Received on Monday, 2 October 2000 02:04:37 UTC