in favor of allowing a fragment in a URI for metadata for an attribute type

[Mark Wahl]

Section 4.3 of
http://openid.net/specs/openid-attribute-types-1_0-02.html
suggests that in URIs defined for attributes for OpenID AX,
"URI fragment specifiers should not be used."

Now I'm no RDF expert, but I'm in favor of allowing fragments,
and perhaps even encouraging them. I'd prefer this statement
be removed from subsequent versions of the OpenID AX, in order
to not dissuade other schema developers from using fragments.
Here are some points for discussion on that topic, I'd be
interested in hearing feedback esp. from other RDF implementors.

0. Some servers will have but a single, small, fixed schema.  I'd
rather those servers be able to reference and serve a single RDF
file with their complete schema, instead of needing to break that
schema into a bunch of little schemas.

For example, suppose a server only supports FOAF.  Now FOAF does not
use fragments for the property definitions for its attribute types,
but the attribute types defined in FOAF are not currently resolvable
to an RDF document that describes those attribute types.

If xmlns.com where the FOAF RDF is hosted were to implement having these
attribute type URIs used in FOAF be resolvable, they
would either need to
  - create a few dozen little RDF files that together mirror the content of
    foaf.rdf, or
  - implement a URI rule that http://xmlns.com/foaf/0.1/*
    returns foaf.rdf

If I were redefining FOAF, I'd have its attributes be defined as fragments,
so that there is only one base URI for the FOAF schema.  (Also to give
them RDF class definitions, see below).

1. It appears to be current practice for RDF representations of metadata
for attributes in Higgins to use fragments.

In OWL-based systems, the RDF object at the base URI of the document
is an OWL Ontology.

In Higgins, which uses OWL, the object at the base URI is an OWL
Ontology that 'imports' the Higgins Ontology.  The RDF file for
an attribute contains an OWL Class for the attribute named by a
fragment,e.g., #Firstname, and several related OWL properties and
RDF instances in that same file that add structure to that class.

2. In our 'schemat' implementations which attempt to generate RDF for
existing schemas of 'legacy'/'installed base' protocols, it is desirable to
be able to have additional, named OWL classes, RDF objects, and other
modelling and descriptive data definitions that are shared across
multiple attributes of a single schema. For example, a schema may
define its own value syntax and matching rules, and wish to share
those syntaxes and matching rules across the attributes of that schema.
It would be desirable if there could be a single RDF file which contains
the attribute type metadata, the syntax definition and matching rule
definition, rather than needing to have the attribute type metadata
in a set of files that are separate from the syntax definitions and
matching rule definitions, or are duplicated in those files.

3. I find that in our implementation 'schemat' of identity metaschema
attribute metadata retrieval that is a definite performance gain if
all the schema elements for a particular schema can be retrieved in
a single HTTP GET.  It is likely that an implementation interested
in an attribute Firstname of a particular schema would also be
seeing a few other attributes Lastname, Middlename etc of the same
schema, and it would be good if a GET that retrieves the data for
Firstname also gives the implementation the rest of the schema so
that it does not need to keep going back and GETing for each
attribute type.

4. Requiring that each be in a separate document would likely lead to
duplication of metadata, particularly Dublin Core metadata that
describes "the schema as a whole".  I feel it would be better if the
RDF object at the base URI has the Dublin Core metadata for the
schema as a whole, and that the Attribute Type Metadata is a class
named by a fragment below that base URI.

5. (appeal to authority) http://www.w3.org/DesignIssues/Fragment.html
"This means that identifiers for arbitrary RDF concepts should have
fragment identifiers. "


Mark Wahl
Informed Control Inc.