Namespace shortening and invalid RDF outputs

[ianmillard]

In some instances the in-built turtle serialiser can output invalid documents due to incorrect namespace shortening.

Consider four types of animal

animals/dog
animals/cat
animals/bird
animals/reptiles/snake

Then take the following example, which parses some triples and outputs them as turtle.
Note we also have also defined a namespace prefix for animals.

    set_include_path(get_include_path() . PATH_SEPARATOR . 'lib/');
    require_once "EasyRdf.php";

    $triples  = '
      <http://example.com/id/1> <http://www.w3.org/2000/01/rdf-schema#type> <http://example.com/ns/animals/dog> .
      <http://example.com/id/2> <http://www.w3.org/2000/01/rdf-schema#type> <http://example.com/ns/animals/cat> .
      <http://example.com/id/3> <http://www.w3.org/2000/01/rdf-schema#type> <http://example.com/ns/animals/bird> .
      <http://example.com/id/4> <http://www.w3.org/2000/01/rdf-schema#type> <http://example.com/ns/animals/reptiles/snake> .
    ';

    EasyRdf_Namespace::set('id',      'http://example.com/id/');
    EasyRdf_Namespace::set('animals', 'http://example.com/ns/animals/');

    //  parse graph
    $graph = new EasyRdf_Graph();
    $graph->parse($triples, 'ntriples');

    //  dump as text/turtle
    print $graph->serialise('turtle');

This will produce the following output, which contains the invalid animals:reptiles/snake

@prefix id: <http://example.com/id/> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix animals: <http://example.com/ns/animals/> .

id:1 rdfs:type animals:dog .
id:2 rdfs:type animals:cat .
id:3 rdfs:type animals:bird .
id:4 rdfs:type animals:reptiles/snake .

In the last triple, the namespace shortening has been applied, but only matched against part of the URI.

There is a simple fix for the turtle parser, in changing line 62 and line 141 to be

if ($short && strpos($short, '/') === false) {

However, this problem still persists in the RDF/XML serialiser, as if I had say http://example.com/ns/property and http://example.com/ns/property/subproperty
then output such as the following would be produced

<ns0:property/subproperty rdf:resource="..." />

So probably a better fix is required within EasyRdf_Namespace ??

Also, in attempting to fix the "snake" issue, I discovered that the namespace shortenings are applied in the order in which they are declared. In the above example code, if you declare reptiles before animals then all is fine

EasyRdf_Namespace::set('reptile', 'http://example.com/ns/animals/reptiles/');
EasyRdf_Namespace::set('animal',  'http://example.com/ns/animals/');

however if you add reptiles after animals...

EasyRdf_Namespace::set('animal',  'http://example.com/ns/animals/');
EasyRdf_Namespace::set('reptile', 'http://example.com/ns/animals/reptiles/');

then you will never get reptile as animal will match first, causing the problem I discovered at the outset.

I suggest that probably each time a new namespace is declared/generated the internal map of prefixes and namespaces should be sorted by length of namespace?

thanks, and sorry this is so long!

Ian

[njh]

Yes, I like the idea of making sure the most specific namespace is matched.

I think I need to read the CURIE specification in detail:
http://www.w3.org/TR/curie/

And check that my implementation is compliant.

The serialiser implementations probably need different rules to ensure that the local part is valid for the serialisation.
For example the local part of a XML QNames isn't allowed to start with a number.

[indeyets]

any news on this one? I just got output with slashes in "local" part too.

[scor]

This issue is causing the RDFa test 0316 from the official RDFa test suite to fail. I initially filed #156 which is a duplicate of this one.

[indeyets]

I implemented simple fix in EasyRdf_Namespace. What do you think? can we close this issue?

[njh]

Looks good to me :)

Can you add a test to EasyRdf_NamespaceTest too please?

[indeyets]

@njh done

[scor]

Thanks @indeyets!

@njh let me know when you deploy that to http://www.easyrdf.org/converter and I'll test the RDFa test which was failing, it should work now...