Looking deeper into the origins of representing namespaced elements in braced syntax, it looks like the issue of a canonical mapping from QName to a URI has a long history at the W3C:

The RDF Specification contains the following language:

Within RDF/XML, XML QNames are transformed into RDF URI references by appending the XML local name to the namespace name (URI reference). For example, if the XML namespace prefix foo has namespace name (URI reference) http://example.org/somewhere/ then the QName foo:bar would correspond to the RDF URI reference http://example.org/somewhere/bar. Note that this restricts which RDF URI references can be made and the same URI can be given in multiple ways.

– RDF/XML Syntax Specification (Revised) §5.1 Identifiers

This method of concatenation introduces flexibility:

  • abstract is a term in the DublinCore http://purl.org/dc/terms/ namespace. Its canonical URI is http://purl.org/dc/terms/abstract, which the expanded form of dcterms:abstract using the concatenation method.
  • type is a property name in the RDF http://www.w3.org/1999/02/22-rdf-syntax-ns# namespace. Its canonical URI uses an URI fragment to reference an XML ID. Here again, the expanded form of rdf:type matches the canoncial URI.

However:

  • int is a datatype defined by the XML-Schema http://www.w3.org/2001/XMLSchema namespace. Its canonical URI is http://www.w3.org/2001/XMLSchema#int. In this case, applying the RDF QName-to-URI transformation would result in http://www.w3.org/2001/XMLSchemaint, which does not match the canonical URI for that type.

Interestingly, the difference comes down to the fragment character being part of the RDF namespace, but not so for the XML Schema namespace, possibly in order to support the Dublin Core case.

This gave rise to a proposal to W3C in 2001, which was turned into RDF Tracking issue rdfms-qname-uri-mapping.

The “resolution” for that issue was to reject the enclosed proposal to make a case-distinction in the RDF mapping algorithm based on the last character of the namespace URI.

When is a QName Mapping valid?

There are at least two papers by W3C TAP on the matter:

The former claims that

The QName “x:p” is a concise, unambiguous name for the {URI, local-name} pair {“http://example.com/ns/foo”, “p”}.

The important takeaway is that QNames are contextual, it’s just a pair of Namespace URI and local part. It does not have the same uniqueness guarantee that would follow from a URI with or without fragment identifier because that can, by definition, only ever refer to a single resource.

An RDF ontology will always talk about terms, so it’s reasonable to assume they have an identity that can be mapped to a full URI, but this isn’t the case for XML Schema, where the same local part may refer to an element, attribute, or datatype.

The braced mapping seems to originate either from XSV which is mentioned in this mail or this memo by James Clark.

For a list of related e-Mail threads, see also this summary.