Trig serialisation handling prefixes incorrectly

[dschallis]

In version 4.1.2 (under python 2.7), running the Trig test detailed at #317 produces invalid Trig output.

Running the code:

from rdflib import ConjunctiveGraph

data = """
@prefix ns1: <http://chartex.org/chartex-schema#> .
<http://chartex.org/graphid/document1> = {
    ns1:Person_A a ns1:Person ;
        ns1:TextSpan "Simon" .
    ns1:Person_B a ns1:Person ;
        ns1:TextSpan "Walter" .
}
<http://chartex.org/graphid/document2> = {
    ns1:Person_C a ns1:Person ;
        ns1:TextSpan "Agnes" .
    ns1:Person_D a ns1:Person ;
        ns1:TextSpan "Emma" .
}
<http://chartex.org/graphid/Person_Atenant_ofPerson_B> = {
    ns1:Person_A ns1:tenant_of ns1:Person_B .
}
<http://chartex.org/graphid/Person_Ctenant_ofPerson_D> = {
    ns1:Person_C ns1:tenant_of ns1:Person_D .
}
<http://chartex.org/graphid/ConfidenceMetrics> = {
    <http://chartex.org/graphid/Person_Atenant_ofPerson_B> ns1:confidence 0.9265 .
    <http://chartex.org/graphid/Person_Ctenant_ofPerson_D> ns1:confidence 0.8765 .
}
"""

cg = ConjunctiveGraph()
cg.parse(data=data, format='trig')
print cg.serialize(format='trig')

produces the invalid Trig output:

@prefix ns1: <http://chartex.org/chartex-schema#> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix xml: <http://www.w3.org/XML/1998/namespace> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

<ns2:Person_Atenant_ofPerson_B> = {
    ns1:Person_A ns1:tenant_of ns1:Person_B .
}

<ns2:ConfidenceMetrics> = {
    ns2:Person_Atenant_ofPerson_B ns1:confidence 0.9265 .

    ns2:Person_Ctenant_ofPerson_D ns1:confidence 0.8765 .
}

<ns2:Person_Ctenant_ofPerson_D> = {
    ns1:Person_C ns1:tenant_of ns1:Person_D .
}

<ns2:document2> = {
    ns1:Person_C a ns1:Person ;
        ns1:TextSpan "Agnes" .

    ns1:Person_D a ns1:Person ;
        ns1:TextSpan "Emma" .
}

<ns2:document1> = {
    ns1:Person_A a ns1:Person ;
        ns1:TextSpan "Simon" .

    ns1:Person_B a ns1:Person ;
        ns1:TextSpan "Walter" .
}

The Trig output contains the prefix "ns2" for graphs, but this prefix isn't defined anywhere. In addition, the prefixed graph name shouldn't be enclosed within angle brackets, e.g. the correct output for each named graph should be either:

@prefix ns2: <http://chartex.org/graphid/> .

ns2:document1 = {
    ...
}

Or:

<http://chartex.org/graphid/document1> = {
    ...
}

[jjon]

hi @dschallis,

Inasmuch as that test was from my own data, I got curious. It seems like back then I didn't notice the problem you found. I tried it again with my development clone and got the same effect you did. But then things got weird and I don't have sufficient experience to interpret what I'm seeing.

With an updated version of 4.2.dev (and Python 2.7.1), I did exactly the experiment you did and got the same result, the list of namespaces was this:

>>> print cg.serialize(format='trig')
@prefix ns1: <http://chartex.org/chartex-schema#> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix xml: <http://www.w3.org/XML/1998/namespace> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

so I tried this:

>>> list(cg.namespaces())
[('xml', rdflib.term.URIRef(u'http://www.w3.org/XML/1998/namespace')), ('rdfs', rdflib.term.URIRef(u'http://www.w3.org/2000/01/rdf-schema#')), ('rdf', rdflib.term.URIRef(u'http://www.w3.org/1999/02/22-rdf-syntax-ns#')), ('xsd', rdflib.term.URIRef(u'http://www.w3.org/2001/XMLSchema#')), (u'ns1', rdflib.term.URIRef(u'http://chartex.org/chartex-schema#')), ('ns2', rdflib.term.URIRef(u'http://chartex.org/graphid/'))]

and thought, "that's odd, there's ns2 right there, and it is defined properly." Baffled, I did this again:

print cg.serialize(format='trig')

and got this:

>>> print cg.serialize(format='trig')
@prefix ns1: <http://chartex.org/chartex-schema#> .
@prefix ns2: <http://chartex.org/graphid/> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix xml: <http://www.w3.org/XML/1998/namespace> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

I'm able to reproduce this effect reliably, but I have no idea what might cause this behavior. It appears that cg.namespaces() lacks ('ns2', rdflib.term.URIRef(u'http://chartex.org/graphid/')) until the graph is first serialized as trig, only then does cg.namespaces() contain the generated namespace.

I hope someone with more experience than I will be able to explain why this happens.

[dschallis]

Hi @jjon,
I've found the cause and have been working on a fix, hopefully I'll get the chance to write some tests and send a pull request sometime today/tomorrow.

[satra]

@dschallis - are there any updates on this? we have run into this with trying to serialize/deserialize the w3c prov model as trig.

[dschallis]

@satra unfortunately not, I looked into the issue, and found some deeper problems regarding rdflib's trig handling (#432 and #433), which make it hard to use/test the fix I'd originally intended. I'm still planning to look into these when I get a chance, but I'll need to spend some time understanding more about some of rdflib's internals first.

[RinkeHoekstra]

I just stumbled upon the same problem, and the problem occurs when the TurtleSerializer serializes a IRI that does not belong to a known namespace. This namespace is then created based on some guesswork, but because the TrigSerializer has by that point already serialized the namespace declaration (the turtle serializer is essentially called for every graph in the store), the newly added namespace is not picked up properly.

For now my workaround is just to serialize the graph twice in succession.

[drewp]

I was running into something like this. These patches may help:
drewp@5d9a7b0
drewp@bd57db7
drewp@f167ecf

I don't remember why I was doing these or why I stopped trying to get them merged back.

[white-gecko]

Regarding the workaround of @RinkeHoekstra still the issue is, that qnames are generated, but printed in <…>

Edit: I've checked this on the current master branch again and patches of @drewp seem to be integrated/merged. Can this issue be closed?

[gromgull]

this works fine in master now.