RDF/XML and Turtle serializations broken for sub-namespaces involving special characters

[osma]

EasyRdf cannot properly serialize data from the EuroVoc thesaurus (see NatLibFi/Skosmos#421 for original issue report).

The problem seems to be that URI namespaces are shortened naively and the shortened URIs (qnames) may contain characters that are invalid in RDF/XML and/or Turtle. EuroVoc has two namespaces, http://eurovoc.europa.eu/ for thesaurus concepts and labels and http://eurovoc.europa.eu/schema# for schema elements such as custom classes and properties. Note that the latter is a sub-namespace of the former.

Code to demonstrate:

$graph = new EasyRdf_Graph();
EasyRdf_Namespace::set('euro', 'http://eurovoc.europa.eu/');

$types = array('http://eurovoc.europa.eu/schema#PreferredTerm');
$res = $graph->resource('http://eurovoc.europa.eu/212626', $types);

echo $graph->serialise('rdfxml');

This outputs:

<?xml version="1.0" encoding="utf-8" ?>
<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
         xmlns:euro="http://eurovoc.europa.eu/">

  <euro:schema#PreferredTerm rdf:about="http://eurovoc.europa.eu/212626">
  </euro:schema#PreferredTerm>

</rdf:RDF>

This is invalid XML, as # cannot be used in tag names.

If I change the output format to turtle instead I get this:

@prefix euro: <http://eurovoc.europa.eu/> .

euro:212626 a euro:schema#PreferredTerm .

In my reading of the RDF 1.1 Turtle grammar, # is not allowed unescaped in local names. In older Turtle versions it is certainly not allowed.

The safest would be to avoid shortening URIs if the result would not be a valid XML QName.

[osma]

I only now noticed that this has been discussed before, in #115 and PR #248.
Anyway, this bug affects EasyRdf 0.9.1.