Re: ISSUE-137 (including XML includes)

On 15 Sep 2008, at 11:42, Jim Hendler wrote:

> Rather than inlining below - let me answer Bijan and Ivan
>  My concern is that if I have an OWL 2 document in some format and  
> I move it to an RDF/XML document, or vice versa, then we'll have  
> interoperability problem with a "pure" Xinclude solution.  What I  
> am suggesting is that we have some sort of directive (I am of mixed  
> mind as to whether it would be suggested or required) that is used  
> when Xinclude is used so there is something in the "OWL" (not in  
> the serialization) that would help pick up the interoperability in  
> this case
>  for example, I could imagine that in the Ontology directory, at  
> the level of owl:imports we could create owl:Xincludes or something  
> like that (or perhaps to say that when an Xinclude is used an  
> imports SHOULD (MUST?) also be used
>  something like that

Let me try to tease out the various options. Corrections welcome.

1) (Alan's) We put a *triple* (i.e., change the graph) in.  
Advantages: "works"[1] for all RDF serializations. Disadvantages:  
Breaks syntactic layering; contaminates the graph and the structural  
model; *requires* a bespoke solution (thus precludes using standards  
like XInclude).

2) (Peter's) Raw Xinclude. We use Xinclude where it works (RDF/XML  
and OWL/XML) and let the other serializations introduce their own  
(equivalent) means of handling this. Advantages: Purely syntactic;  
uses standard mechanism; no ramifications for OWL Full, etc..  
Disadvantages: Interop lag as other serializations and their  
implementation catch up.

3) (Jim's) We use XInclude put also provide a marker that XInclude  
was used.

Ok, I'm not sure how to evaluate this design. If we have a triple per  
inclusion statement then we have Alan's solution *except* we also  
have a default implementation in the RDF/XML? But then we'd have the  
"same" information in two places. Jim could you elaborate your  
proposed design sketch a little more?

BTW, I presume that the only alternative RDF serialization that  
matters is Turtle. Obviously, Manchester syntax and functional syntax  
could easily accommodate another directive.


[1] It is tendentious of me to put "works" in scare quotes here. Alan  
thinks it works because it appears in all serializations without  
change and defers semantic handling to another layer. I think that if  
semantic handling is at another layer then, functionally speaking, it  
doesn't work.

Received on Monday, 15 September 2008 11:05:47 UTC