trig export of multiple graphs assigns wrong prefixes to prefixedNames

[engsterhold]

It seems, that trig export does not handle prefixes and named graphs very well (or at all). Trying to export a dataset with multiple graphs overwrites already assigned prefixes, produces different prefix assignments, wrong graph prefixes or prefixNames on different runs.

Testcase:

from rdflib import URIRef, Literal,  Graph, Dataset
def trig_export_test():
    graphs = [(URIRef("urn:tg1"),"A"), (URIRef("urn:tg2"), "B")]
    ds = Dataset()
    for i, n in graphs:
        g = Graph(identifier=i)
        a = URIRef("urn:{}#S".format(n))
        b = URIRef("urn:{}#p".format(n))
        c = Literal(chr(0xf23f1))
        d = Literal(chr(0x66))
        e = Literal(chr(0x23f2))
        g.add((a,b,c))
        g.add((a,b,d))
        g.add((a,b,e))
        ds.graph(g)

    for n,k in [ ("json-ld","jsonld"),
                 ("nquads", "nq"), ("trix", "trix"), ("trig", "trig")]:
        ds.serialize("ds.{}".format(k), format=n)

trig_export_test()

Can produce the following

@prefix ns1: <urn:> .
@prefix ns2: <urn:B#> .
@prefix pns1: <urn:x-rdflib:> .
@prefix pns2: <urn:A#> .
@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#> .

pns1:tg2 {
    pns2:S pns2:p "f",
            "⏲",
            "󲏱" .
}

pns1:tg1 {
    pns2:S pns2:p "f",
            "⏲",
            "󲏱" .
}

{}

or

@prefix ns1: <urn:x-rdflib:> .
@prefix ns2: <urn:A#> .
@prefix pns1: <urn:> .
@prefix pns2: <urn:B#> .
@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#> .

{}

pns1:tg1 {
    pns2:S pns2:p "f",
            "⏲",
            "󲏱" .
}

pns1:tg2 {
    pns2:S pns2:p "f",
            "⏲",
            "󲏱" .
}

both are created with the same function

[jbaublitz]

I wanted to add that I see this behavior even in the case of parsing a dataset with a single graph in nquads format and serializing as trig with no special characters. This seems to be an issue serializing named graphs as trig even in cases with one named graph in the dataset. I can also verify that the same nquads parsing allows proper nquads serialization so it does not appear to be a data corruption issue in the Graph data structure.

[gromgull]

I've fixed the broken trig output, but it remains ugly.

The underlying issue is #698 - since you do:

g = Graph(identifier=i)
[ ... add data ... ]
ds.graph(g)

you end up with a Graph instance with it's own namespace manager. So each subgraph generates it's own independent space of prefix mappings during trig serialisation.

It's actually much worse than this - these graphs do not share a store. So if you tried to persist the dataset, nothing would work.

It's our fault, passing a graph instance to DataSet.graph should probably raise an exception.

Change the code to :

g = ds.graph(i)
[... add data ...]

and all is good.