Chimezie’s posterous

Okay. I think it needs to be said that RDF / OWL tool-makers do not
do a good job of respecting the mechanics of xml:base . It seems
almost a given that most of them mangle the URI base resolution
conventions they find in existing RDF/OWL files, overwriting them with
their own. Often, this happens in a very destructive way that can be
even more annoying for a person who has much experience using XML and
thus has a better appreciation (perhaps) of the value of a base URI of
an OWL document for more than just owl:Ontology/@rdf:about.
 
This is the typical scenario: I create an OWL document like so:
 
<rdf:RDF>
 <owl:Ontology rdf:about="">
   <owl:imports rdf:resource=".. relative path .."/>
 </owl:Ontology>
 <owl:Class rdf:about="tag:info@example.com#Stuff"/>
</rdf:RDF>
 
First, let me describe the intent here. I think it is perfectly
reasonable to modularize large ontologies into fragments (if you will)
that are bundled together and linked via relative imports. Since they
are bundled together, I often like to use the power of URI resolution
to make relative references to imported OWL documents (as you can see
above) in a way that is completely independent from where the bundle
is being deployed (file system or on the web).
 
The downside of not using xml:base explicitly is that the URIs of my
classes need to be fully qualified, but as far as I'm concerned this
is outweighed by the advantage of knowing that I can deploy my bundled
import network on a website or on a filesystem and have any
self-respecting tool know how to resolve the relative URIs.
 
One of the more involved technical points during the development of
GRDDL was regarding the use of empty URI references in GRDDL results.
The primary importance of empty, relative URI references (in RDF serializations) is the
ability to refer to the containing document without necessarily having
the URI on hand. It was this particular dialogue that made me better
appreciate the power and simplicity of the URI base resolution process
that sits at the bottom of many of the important W3C specifications.
 
From RFC 3986 (the process of determining the Base URI to use when
resolving relative URIs):
 
* The base URI is embedded in the document's content.
* The base URI is that of the encapsulating entity (message, document, or none).
* The base URI is the URI used to retrieve the entity.
* The base URI is defined by the context of the application.
 
Getting back to my example above (this was supposed to be a short
rant). Too often what happens is that when I load an OWL document
such as the oneabove into an OWL editor and save it (even after not
making any changes), it results in:
 
<rdf:RDF xml:base="tag:info@example.com#">
 <owl:Ontology rdf:about="">
   <owl:imports rdf:resource="file:///path/to/OWL/document"/>
 </owl:Ontology>
 <owl:Class rdf:ID="Stuff"/>
</rdf:RDF>
 
<insert appropriate expletive>
 
Now, it's not so much the 'forced' use of rdf:ID that bothers me as
the fact that the relative paths used for the imports are now replaced
with absolute URIs. This sabotages the advantage I once had of being
able to rely on the URI resolution chain to be sufficient for clients
that need to resolve my relative URIs. Essentially, the OWL tool has
monopolized the opportunity to resolve relative URI references.
 
Now, in fairness, it seems much of the motivation of doing this is to
associate an explicit base URI with the ontology itself and often
ontology tools will complain if they are unable to determine an
absolute URI to use for this purpose. I think a better mechanism for
doing this would be explicit attributes or elements rather than
bastardization of the ability to give an explicit URI base in content.
 
It seems to me that if the author wanted to associate an explicit URI
with the ontology he or she would use one in the place of an empty,
relative URI reference. In fact, I would go as far as saying that it
is bad practice to forcibly insert an @xml:base in a document that
doesn't have one *and* uses empty, relative reference!
 
So, for any developers of RDF / OWL tools, please take care in trying
not to enforce a base resolution scheme despite the document author
providing one of their own for reasons that are orthogonal to
associating a URI to an ontology (and more important as far I'm
concerned): deployability.