owlrl-turbo

Part of the Legra RDF stack: pyshacl-turbo, rdflib-turbo and owlrl-turbo.

Functional alpha of native RDFS and OWL-RL materialization for Python. Uses the shared exact-term dataset runtime provided by rdflib-turbo, with native inference extracted from pyshacl-turbo. There is no runtime dependency on Python owlrl or pyshacl-turbo and no Python inference fallback.

pip install owlrl-turbo==0.1.0a1

from rdflib_turbo import Dataset
from owlrl_turbo import expand

source = Dataset().parse(data='''
    @prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
    <urn:Person> rdfs:subClassOf <urn:Agent> .
    <urn:alice> a <urn:Person> .
''')
expanded = expand(source, inference="rdfs")
assert bool(expanded.query('ASK { <urn:alice> a <urn:Agent> }'))
assert len(source) == 2  # the default leaves the source unchanged

# Reuse the existing dataset handle and commit only on successful inference.
expand(source, inference="owlrl", inplace=True)

expand(graph, inference="owlrl", inplace=False, destination=None) accepts rdflib_turbo.Dataset or RDFLib Graph/Dataset. Modes are rdfs, owlrl and both. The return value has the input kind. Native inputs remain native between calls; RDFLib inputs require a copy into the native runtime and conversion back. RDFLib namespaces and blank-node identities are preserved. Named input contexts are retained. RDFLib Dataset inference defaults to urn:owlrl-turbo:inference; use destination to explicitly name the output graph for Dataset inputs.

Native operations release the GIL and hold a dataset lock throughout execution. Inference commits only after successful native execution. Errors leave the source unchanged. RDFLib inplace commits occur after native success, using RDFLib's normal mutation API; external RDFLib stores do not acquire a new transaction guarantee.

This is not a complete drop-in owlrl replacement. DeductiveClosure classes, custom Python rule hooks, owl:imports loading, configurable axiom/datatype options, OWL-RL extensions and complete inconsistency diagnostics are not exposed. Inference initialization follows pyshacl-turbo's pinned modes, rather than configurable owlrl constructor flags. Full OWL-RL conformance and output parity are not claimed.

Output stays within standard RDF: literal subjects and non-IRI predicates are excluded. Valid literal-valued consequences remain part of the contract. Native owl:hasKey checks every property, intentionally differing from the pinned Python reasoner's incomplete comparison. See ADR 0001.

For SHACL, pass expanded.to_rdflib() to pyshacl_turbo.validate(). This conversion copies; pyshacl-turbo does not yet share the new dataset handle. No speedup claims are made for this initial release. Requires Python 3.11+; rdflib-turbo supplies the native platform wheel or builds its extension from source.

MIT OR Apache-2.0. Copyright 2026 DataRoad Inc, Delaware, USA, trading as Legra.

Development

Use make dev and make check. The initial alpha is published only by the verified GitHub Actions workflow; see release instructions.

Install rdflib-turbo first, then install this project with its test extra and run pytest. Tests compare complete closures on bounded fixtures and exercise source mutation, malformed inputs, standard RDF output and a pyshacl-turbo pipeline.

CI requires 100% measured Python package line and branch coverage. This does not measure Rust coverage, full API compatibility or standards conformance.