Re: [ISSUE-37] New proposal on RIF interoperation with XML data and XML Schemas

Christian de Sainte Marie wrote:
> All,
> 
> </chair>
> 
> I have uploaded a new document [1] that contains:
> - my new proposal for dealing with XML data sources and for importing 
> data sources' data models as XML schema (closer, in the approach, to the 
> way SWC deals with combining RIF and RDF/OWL; and compatible with it, as 
> far as I can see);
> - Gary's strawman proposal for mapping XML Schema valid data to RIF 
> frames [2].
> 
> These are proposed resolutions for ISSUE-37 [3] (Interoperation with 
> Object-Oriented XML (using XML Schema, like JAXB)). Mine would also 
> resolve ISSUE-38 [4] (Interoperation with Arbitrary XML (like SAX/DOM)), 
> I think.

> Notice that the two proposals are orthogonal and not incompatible with 
> each other.

Well the detailed correspondences they set up between #/frames and XML 
schema are different and incompatible in detail.

It seems like there are two different aspects of the proposals and it 
might be worth separating them out.

(1) How is the mapping between the XML Schema and corresponding RIF 
representation defined?

Your proposal is that the RIF rule set declares it's use of a schema and 
instance document via Import and that import invokes a set of semantic 
rules which define an interpretation for # and for particular frame 
slots via XPATH.

I think Gary's proposal is silent on how the mapping is declared in the 
RIF rule set and the specification is given as a mapping rather than a 
set of semantic correspondence rules.

(2) What RIF constants correspond to particular XML Schema/XML instance 
document elements?

You proposal is to use qnames for this for most cases (presumably 
concatenating the qnames to iris for use in RIF but to also introduce a 
new syntax for attributes which modifies the RIF syntax rather than 
being part of the constant.

Gary's proposal defines a mapping from XML Schema to IRIs.

Is that a reasonable characterization?

On aspect (1) then the use of import and the style of specification seem 
reasonable to me.

On aspect (2) the the detailed mapping of XML Schema elements you 
propose has several problems.

Firstly there is more overloading of namespaces in XML Schema than just 
elements/attributes, the introduction section of [1] gives a list of 
reasons why qnames are insufficient to designate schema components.

Secondly, introducing new syntax just for attributes and having to 
change the semantics to cope with this seems like a huge and completely 
unnecessary change. It is perfectly possible to define an XML Schema 
element to IRI mapping which can designate attributes separate from 
elements so just do that. Gary's mapping is one such example.

Thirdly, your proposal talks about QNAMEs but RIF currently doesn't have 
QNAMEs. I'm unclear if you are taking the RDF-style approach of 
concatenating the components of a QNAME to form an IRI or proposing to 
add a QNAME datatype. I'm hoping it's the former.

Can you combine your approach to the mapping definition with Gary's 
actual mapping?

My previous comments on Gary's proposal still stand though. Given that 
designating XML Schema elements by IRIs is a common problem across W3C 
specs and is already being addressed by [1] is it not possible to just 
adopt that?

I also still like the idea that where SAWSDL [2] declarations exist in a 
schema one should use those to define the correspondence to classes and 
properties (sorry frame slots) and only fall back on XSCD for 
un-annotated schemas. However, that never seems to have gone down very 
well before and I don't care about it enough to press the point further.

Dave

[1] http://www.w3.org/TR/xmlschema-ref
[2] http://www.w3.org/TR/sawsdl/

-- 
Hewlett-Packard Limited
Registered Office: Cain Road, Bracknell, Berks RG12 1HN
Registered No: 690597 England

Received on Monday, 16 March 2009 19:09:27 UTC