XML serialization with base=namespace uri Results in Empty vann:preferredNamespaceUri for Vocab Description

[typinator]

I've created a vocab description and use both the XML and Turtle serializations.

TRAK = Namespace('https://purl.org/trak/metamodel/vocab#')

The Turtle one (format='turtle', base = TRAK) is correct starting with (excerpt) :

@prefix owl: <http://www.w3.org/2002/07/owl#> .
@prefix trak: <https://purl.org/trak/metamodel/vocab#> .
@prefix vann: <http://purl.org/vocab/vann/> .
 .

trak: a owl:Ontology ;
    vann:preferredNamespacePrefix "trak"^^xsd:string ;
    vann:preferredNamespaceUri trak: .

having sorted using the otsrdflib package to create a systematic and repeatable order.

The problem occurs with the XML version (format="pretty-xml", base=TRAK). This produces:

<?xml version='1.0' encoding='us-ascii'?>
<rdf:RDF  xmlns:ns1="http://www.w3.org/2002/07/owl#" xmlns:ns2="http://purl.org/vocab/vann/"  xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xml:base="https://purl.org/trak/metamodel/vocab#">
  <ns1:Ontology rdf:about="">
    <ns2:preferredNamespacePrefix rdf:datatype="http://www.w3.org/2001/XMLSchema#string">trak</ns2:preferredNamespacePrefix>
    <ns2:preferredNamespaceUri rdf:resource="" />
    ...
  </ns1:Class> 

The problem is that the base namespace uri that formed the value for ns2:preferredNamespaceUri has been replaced leaving a null / empty value which isn't equivalent to the base uri.

How do I get around this to produce the equivalent of the Turtle vann:preferredNamespaceUri statement? I can obviously manually correct after output but this doesn't seem like the right approach.

[gjhiggins]

Tricky.

I fed the RDF/XML Trak Vocab source to Jena’s riot command-line parser and it produced:

<>      vann:preferredNamespacePrefix  "trak";
        vann:preferredNamespaceUri     <#>;

Using rdflib to parse the above-mentioned RDF/XML source, I couldn’t reproduce your example:

 vann:preferredNamespacePrefix "trak"^^xsd:string ;
 vann:preferredNamespaceUri trak: .

I could only get:

    vann:preferredNamespacePrefix "trak"^^xsd:string ;
    vann:preferredNamespaceUri <> ;

There’s some discussion of how to handle VANN’s preferredNamespaceUri on SemanticArt’s gist issues list :

Note: The rdfs:range of the VANN preferredNamespaceUri predicate (although not explicitly stated in the VANN documentation) is accepted to be an xsd:string, and not an IRI, or xsd:anyURI.

(although, judging by their SHACL declarations, they eventually settled on xsd:anyURI)

And the SWI-Prolog implementation takes a similar approach:

Note that the range of vann:preferredNamespaceUri is a literal.

So, to address your question:

How do I get around this to produce the equivalent of the Turtle vann:preferredNamespaceUri statement? I can obviously manually correct after output but this doesn't seem like the right approach.

If it’s not feasible to modify the source, then I suspect that you are stuck with manual correction.

In case it proves useful, here’s a programmatic helper to use before serialization:

def retype_preferredNamespaceUri(g, ns):
    import rdflib
    prefuri = rdfib.URIRef("http://purl.org/vocab/vann/preferredNamespaceUri")
    if len(ontref := list(g.subject_objects(prefuri))) == 1:  # should only be one
        ontrefs, ontrefo = ontref[0]
        g.remove((ontrefs, prefuri, ontrefo))
        g.add((ontrefs, prefuri, rdflib.Literal(str(ns), datatype=rdflib.XSD.anyURI)))
    return g

[typinator]

Thanks for this. It raises a number of interesting points. I take your point about the conflicting definitions of the datatype applied to the range.

The graph and the Turtle + RDF serializations and produced by querying a Neo4J directed graph describing the TRAK metamodel and architecture viewpoints (short description on trakmetamodel and trakviewpoints websites.

In terms of rdflib it's the same source graph used for both the Turtle and RDF serializations. The problem is that the XML form of base = 'https://purl.org/trak/vocab#' + <vann:preferredNamespaceUri rdf:resource="" /> is not equivalent to the Turtle serialized vann:preferredNamespaceUri trak:.

I'll give that def a try.

The other thought was a quick and dirty treat the XML file as text and find and replace which is the analog of my manual correction.

My one wish is that I had control over the output sort order of elements so that I don't have to use additional packages.