pyshifty 0.0.5

Python bindings for the shifty validation engine

shifty Python bindings

This crate packages the shifty validator as a CPython extension using PyO3. It exposes the shifty module for RDFlib-centric workflows while reusing the same Rust engine that powers the CLI.

Installation

From PyPI (package name pyshifty, import name shifty):

pip install pyshifty

From the repository root you can develop locally with

cd python
uv sync  # or `pip install -r requirements.txt` if you prefer
uvx maturin develop --extras rdflib --release

The GitHub Actions workflow (.github/workflows/python.yml) publishes the package to PyPI as pyshifty and uses this README as the long description, so keep it up to date when the bindings change.

Usage

The extension exports one-shot helpers plus a CompiledShapeGraph cache that mirrors the CLI subcommands:

import shifty

# Validation. When you request diagnostics, a fourth element is returned.
conforms, results_graph, report_text, diag = shifty.validate(
    data_graph,
    shapes_graph,
    run_inference=True,
    inference={"min_iterations": 1, "max_iterations": 8},
    graphviz=True,
    heatmap=True,
    trace_events=True,
    return_inference_outcome=True,
)
print(diag["graphviz"])        # DOT for the shapes graph
print(diag["heatmap"])         # DOT for the execution heatmap
print(diag["trace_events"][0]) # First trace event (dict)
print(diag["inference_outcome"])

# Inference-only. Diagnostics are returned as a second element when requested.
inferred_graph, diag = shifty.infer(
    data_graph,
    shapes_graph,
    min_iterations=1,
    max_iterations=4,
    graphviz=True,
    return_inference_outcome=True,
)
print(diag["inference_outcome"]["triples_added"])

# Cache shapes once, then reuse them across datasets.
compiled = shifty.generate_ir(
    shapes_graph,
    skip_invalid_rules=True,
    warnings_are_errors=False,
    do_imports=True,
)
conforms, _, _, diag = compiled.validate(data_graph, run_inference=True)
print("Compiled cache conforms?", conforms)

Key options (mirroring the CLI flags):

• skip_invalid_rules (default: False), warnings_are_errors, do_imports
• Inference knobs: min_iterations, max_iterations, run_until_converged/no_converge, error_on_blank_nodes, debug; the inference={...} dict still works and aliases like inference_min_iterations remain.
• Diagnostics: graphviz (DOT for shapes), heatmap + heatmap_all (execution heatmap, triggers a validation pass), trace_events, trace_file, trace_jsonl, return_inference_outcome (adds iteration/insert counts).

If you omit all diagnostics, validate returns (conforms, results_graph, report_text) and infer returns the inferred rdflib.Graph just like before.

See example.py and brick.py in this directory for full RDFlib/OntoEnv examples.

Packaging notes

• Cargo.toml declares readme = "README.md", so this file must be checked into the repo.
• The sdist job in GitHub Actions uses the same path when stamping metadata via maturin.
• Any new runtime assets (fixtures, schemas, etc.) should be listed in pyproject.toml under tool.maturin.sdist.include so they reach PyPI.