Plugin store rdf4j cannot handle BNode

[thomaslcde]

Hello.

I use_rdflib[rdf4j]>=7.5.0_ downloaded yesterday. It seems when store=RDF4J we cannot handle Bnode.

  server = Dataset(store=RDF4JStore(base_url="http://localhost:7000/rdf4j-server/", repository_id=REPO_ID))
  NAME = Namespace("http://example.com/#")
  id = NAME.term("subject")
  graph_name = NAME.term("Agraph")
  server.bind("name", NAME)
  g = Graph(identifier=graph_name, store=server.store)
  g.add((id, RDF.type, Literal("Object")))
  bn = BNode()
  g.add((bn, RDF.type, RDF.Seq))
  g.add((id, NAME.alist, bn))
  print(server.serialize(format="trig"))

Here is a simple piece of code, the last line cannot serialize the dataset because there is a bnode in the graph.
Its works fine with store=Memory().
I tested with a docker compose and testcontainers, the image is eclipse/rdf4j-workbench:5.2.0-tomcat

The error is the follow:

  File "Somehere/.venv/lib/python3.13/site-packages/rdflib/contrib/rdf4j/util.py", line 179, in validate_no_bnodes
ValueError: Subject, predicate, and object must not be a BNode: Na739021968584714bc4992b050f7ab81, None, None

Why can't we use Bnode with rdf4j ?

[ticapix]

Manually checking using the rdf4j-workbench, the data is correctly inserted in the repository.

The issue is related to the dataset serialisation.

[edmondchuc]

Why can't we use Bnode with rdf4j ?

This fails to work with the SPARQL Store as well.

from rdflib.plugins.stores.sparqlstore import SPARQLUpdateStore

store = SPARQLUpdateStore(
    f"{ds.store.repo.http_client.base_url}repositories/test-repo",
    update_endpoint=f"{ds.store.repo.http_client.base_url}repositories/test-repo/statements",
    auth=("admin", "admin")
)
g = Graph(identifier=graph_name, store=store)

This is a general issue with serializers needing to query the store using blank node triple matching.

rdflib/rdflib/plugins/serializers/turtle.py

Lines 128 to 140 in 91c8cbc

	def buildPredicateHash(
	self, subject: _SubjectType
	) -> Mapping[_PredicateType, List[Node]]:
	"""
	Build a hash key by predicate to a list of objects for the given
	subject
	"""
	properties: Dict[_PredicateType, List[Node]] = {}
	for s, p, o in self.store.triples((subject, None, None)):
	oList = properties.get(p, [])
	oList.append(o)
	properties[p] = oList
	return properties

In your case, because subject is a blank node, it tries to fetch statements matching the pattern, but this is not possible over SPARQL-based protocols.

The good news is that there is a workaround by accessing the underlying store's repository client directly. Note the inline comment around adding blank node statements. See https://www.w3.org/TR/rdf12-concepts/#section-blank-nodes for more information.

server = Dataset(store=RDF4JStore(base_url="http://localhost:7000/rdf4j-server/", repository_id=REPO_ID))
NAME = Namespace("http://example.com/#")
id = NAME.term("subject")
graph_name = NAME.term("Agraph")
server.bind("name", NAME)

g = Graph(identifier=graph_name, store=server.store)

bn = BNode()
# Use addN when using blank nodes as the identifiers are locally scoped.
# See https://www.w3.org/TR/rdf12-concepts/#section-blank-nodes
g.addN([
    (id, RDF.type, Literal("Object"), g),
    (bn, RDF.type, RDF.Seq, g),
    (id, NAME.alist, bn, g)
])

repo: Repository = g.store.repo
# Retrieve the entire set of repository statements and load them into an in-memory RDFLib dataset.
memory_ds = repo.get(graph_name=graph_name)
print(memory_ds.serialize(format="trig"))

Output:

@prefix name: <http://example.com/#> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .

name:Agraph {
    name:subject a "Object" ;
        name:alist _:N1525f1a009414c3c887bf5f93077a7f7 .

    _:N1525f1a009414c3c887bf5f93077a7f7 a rdf:Seq .
}

[thomaslcde]

Hey thanks edmondchuc for the fix your code works well.
It seems to me, the behavior of the lib does not give much room as i expected.
Im gonna stick to before the 7.5.0 and directly call the rdf4j-server api with httpx.
I need to work with Bnode and not be constraints with it.

In another way, i do love the repo.transaction rdflib 7.5.0 provide, it could be possible to make both implementation live together. at the root all functions call the api. its should not be complicated.