rdfms-not-id-and-resource-attr

It was proposed at the Dec 14 telecon that the current syntax
working draft resolved the issue:

 http://www.w3.org/2000/03/rdf-tracking/#rdfms-not-id-and-resource-attr


It was noted that agreeing to publish the working draft *does
not* implicitly agree this.

I oppose this proposed resolution.
I currently intend to vote against it at the next telecon.

I find two possibilities acceptable:
+ dropping reification in its entirity
+ having a plausible syntax for reification

I am concerned about the proposed meaning of the rdf:ID attribute
in certain empty property element productions (those with one or more
propertyAttr).
http://ilrt.org/discovery/2001/07/rdf-syntax-grammar/#emptyPropertyElt

  gives rdf:ID in this case the meaning of identifying the object node of
the property element triple.

In all other cases, an rdf:ID on a propertyElt identifies the reification of
the property element triple.

The phrase "property element triple" above means the triple created as a
result of the property element with predicate given from the qname of the
property element start tag.


As I have previously argued M&S is inconsistent and has been
inconsistenly implemented and can be clarified in a number of
different ways. Dave's clarification has the side effect of
being to continue to apply the death of a thousand cuts to
reification.

Dave's position is that in the document:

<rdf:RDF xmlns:rdf="..."
         xmlns:x="http://example.org/">
  <rdf:Description>
     <x:prop rdf:ID="r1">val</x:prop>
     <x:prop rdf:ID="r2" rdf:parseType="Resource">
         <x:prop>val2</x:prop>
     </x:prop>
     <x:prop rdf:ID="r3" rdf:parseType="Resource"/>
     <x:prop rdf:ID="r4"/>
     <x:prop rdf:ID="x5" x:prop="val2" />

  </rdf:Description>
</rdf:RDF>


That the property attribute rdf:ID within a property element
maps to reification in the graph for almost always (e.g. cases
r1, r2, r3, r4) and just occassionally, according to some baroque
rules that are still incorrectly articulated, it has some other
semantics. (See following posting for feedback on the mistake - this of
course can be fixed, but is symptomatic of a bad rule).

The result is that anyone who isn't an RDF syntax expert cannot use
reification, because every so often this corner case will come back
and bite them.

If that is what we want, we should kill reification completely.


Hence,

given that:
-  M&S gives two inconsistent readings for the relevant productions
-  implementations have varied in their misreadings of M&S on this issue
-  it was raised very early in RDF comments that M&S is inconsistent
  (and this comment has not yet been satisfactorily addressed)


I propose that we
+  decide that rdf:ID on property elements productions uniformly
   signals reification
+  action the syntax WD editor to make the necessary changes


Moreover, but independently,

I propose that we allow rdf:ID and rdf:resource on empty property
element productions.

(This is a change, but seems to me to be only a simple tidying up.
If no one else agrees with this last point, I would not oppose
the opposite resolution).


Links:
  A previous thread:
http://lists.w3.org/Archives/Public/w3c-rdfcore-wg/2001Oct/thread.html#332
  The first posting to rdf comments on this issue:
http://lists.w3.org/Archives/Public/www-rdf-comments/1999JanMar/0026.html

On killing reification:

That is a serious suggestion, I have prepared a proposal to that
effect but I am seeking feedback from HP colleagues before making it.
i.e. as HP rep, I should try to check whether any HP people use, or
intend to use, reification before finalising my position on it.
(It would be kind of embarrassing if I had to oppose my own proposal because
my colleagues felt strongly the other way!).,


Jeremy

Received on Sunday, 16 December 2001 10:45:02 UTC