Proposal to drop support for pointing to a specific DR

We have a slight problem, but only a slight one.

In the current published version of the DR doc there's a section on 
"Multiple DRs With Different Attribution" [1]. This suggests that you 
can do this:

<dr xml:id="red">
   <iriset>
     <includehosts>example.com</includehosts>
   </iriset>

   <descriptorset>
     <palette:color>red</palette:color>
   </descriptorset>
</dr>

<dr ref="http://example.com/powder2.xml#square" />

i.e. link from a POWDER doc to a specific DR in another doc.

Well, you can in POWDER. The semantics here being that after you've 
finished parsing the first DR, you might want to go and take a look at 
http://example.com/powder2.xml#square. Fine... but it doesn't translate 
exactly into POWDER-S. At least, not as we're now writing it following 
the discussion over how to express the sub class relationship [2]. The 
simpler way to assert the sub class, and in my view the more natural 
way, is to do this:

<owl:Class rdf:about="#resourceset_1">
   <rdfs:subClassOf rdf:ID="description_1"/>
</owl:Class>

This has particular benefits when it comes to expressing ordered lists 
in POWDER-S where it becomes a very elegant and simple way of excluding 
IRI sets that should be excluded 'cos they're higher up the list (see 
the Stasinos paper on this).

But... notice that the descriptor set and IRI set have their identifiers 
and the bit of RDF/OWL here just adds to the information about 
resourceset_1 - it has no identity of its own. Therefore, there is 
nowhere to use the 'square' identifier that we had in the original 
POWDER XML.

Does this matter?

What exactly should a processor do with the information that 'there's 
another DR at http://example.com/powder2.xml#square ? Shouldn't it first 
decide whether to trust it or not? If you go straight to the DR you 
might skip the attribution information - which goes against the ethos 
somewhat (and means a different processing model depending whether you 
arrive at the DR with or without a fragment identifier). I think it 
could be argued... OK, I will argue... that the better thing to do is to 
link to the external document as a whole. One might think of it like this:

<attribution>
   <maker ref="http://www.example.org/foaf.rdf#me" />
</attribution>

<dr xml:id="red">
   <iriset>
     <includehosts>example.com</includehosts>
   </iriset>
   <descriptorset>
     <palette:color>red</palette:color>
   </descriptorset>
</dr>

<seealso ref="http://example.com/powder2.xml" />

In POWDER-S that last element would become:

<rdf:Description rdf:about="">
   <rdfs:seeAlso rdf:resource="http://example.com/powder2.xml
</rdf:Description>

Which means, in POWDER-S, you may as well put it in what we think of as 
the attribution block since the subject of the triple is the document 
itself, as is the case for the foaf stuff etc.

I hope I'm making this clear although I fear I may not be :-(.

Basically, I'm arguing that we should just use an element called 
<seealso /> to link from one POWDER doc to another and not worry about 
linking to a specific DR.

Phil.


[2] http://lists.w3.org/Archives/Public/public-powderwg/2008Apr/0034.html
[1] http://www.w3.org/TR/2008/WD-powder-dr-20080317/#multiDRatt

Received on Wednesday, 7 May 2008 16:36:25 UTC