The # character is erased from the URI of a NAMED graph while querying it

[anatoly-scherbakov]

Please consider the following Python snippet.

import rdflib

print(f'RDFLib version: {rdflib.__version__}')

response = rdflib.ConjunctiveGraph().query('''
SELECT DISTINCT ?g
FROM NAMED <http://www.w3.org/2000/01/rdf-schema#>
WHERE {
  GRAPH ?g {
    ?s ?p ?o .
  }
}
''')

print(list(response))

On my machine, this outputs:

RDFLib version: 5.0.0
[(rdflib.term.URIRef('http://www.w3.org/2000/01/rdf-schema'),)]

Essentially, what's happening is that we create one named graph from http://www.w3.org/2000/01/rdf-schema# and then query the RDF dataset for all named graphs which have at least one triple. Which yields http://www.w3.org/2000/01/rdf-schema - without the # character.

I found this when debugging a larger SPARQL query, where the statement like

GRAPH rdfs: {
  ...
}

did not yield anything.

1. Is this expected behavior?
2. If it is, could you please point me to the place in SPARQL spec which describes it?
3. If not, maybe you could point me to a direction where this might happen? I digged into the code, but I am not familiar with rdflib well enough yet; I could not find where the IRI could have been corrupted.

By the way, when constructing a ConjunctiveGraph with parse() calls which specify IRIs for the named graphs — all of this works perfectly.

[anatoly-scherbakov]

The source of the problem appears to be here:

https://github.com/RDFLib/rdflib/blob/master/rdflib/parser.py#L262

absolute_location = URIRef(location, base=base).defrag()
...
input_source = URLInputSource(absolute_location, format)

The URIRef.defrag() method does exactly this – it removes the URL fragment:

rdflib/rdflib/term.py

Line 262 in 5e06430

def defrag(self):

    def defrag(self):
        if "#" in self:
            url, frag = urldefrag(self)
            return URIRef(url)
        else:
            return self

The problematic line 262 in parser.py looks to be from 08.03.2009 according to git blame – this is quite old 🙂

Preparing a PR for this...

[anatoly-scherbakov]

After adding this, I realized that rdflib traverses redirects and changes the IRI of the named graph in case redirects took place. This means that the name of a named graph is generally not predictable in a SPARQL query. This would have been solved if SPARQL would support something like

FROM NAMED <iri> AS <another_iri>

But this construct arguably does not work?