cmem-plugin-graphsigning 0.5.0__py3-none-any.whl

cmem_plugin_graphsigning/__init__.py

@@ -0,0 +1 @@
+"""cmem-plugin-graphsigning"""

cmem_plugin_graphsigning/canonicalization.py

@@ -0,0 +1,173 @@
+"""RDFC-1.0 canonicalization, delegated to pyoxigraph rather than hand-rolled.
+
+Canonicalizing a blank-node-heavy graph correctly - assigning the one blank node
+labelling the algorithm defines as canonical, not merely *a* consistent one - is
+notoriously easy to get subtly wrong, and worst-case exponential in the number of
+blank nodes by the algorithm's own design. pyoxigraph implements RDFC-1.0 against
+the official W3C test suite, so this module leans on it rather than maintaining a
+from-scratch canonicalizer.
+
+Canonicalization runs in a throwaway subprocess so a graph shaped to blow up the
+algorithm's worst case can be killed outright once it exceeds the time budget - a
+plain thread or signal-based timeout cannot interrupt a blocking native call like
+this one.
+"""
+
+import hashlib
+import multiprocessing as mp
+import multiprocessing.connection
+from dataclasses import dataclass
+
+import pyoxigraph as ox
+import rdflib
+
+DEFAULT_TIMEOUT_SECONDS = 5.0
+
+
+class CanonicalizationTimeoutError(TimeoutError):
+    """Raised when canonicalizing a graph exceeds its time budget.
+
+    RDFC-1.0's blank node labelling algorithm is worst-case exponential in the
+    number of blank nodes by design, so a graph shaped adversarially - or just
+    unusually blank-node-heavy - can be made to run arbitrarily long; this is a
+    property of the algorithm itself; not a bug either here or in pyoxigraph.
+    """
+
+    def __init__(self, timeout: float) -> None:
+        super().__init__(
+            f"Canonicalization exceeded {timeout}s. A graph with many blank nodes "
+            f"that hash alike can make RDFC-1.0's worst case exponential; this is "
+            f"inherent to the algorithm, not a bug."
+        )
+
+
+class CanonicalizationError(ValueError):
+    """Raised when *nquads* could not be parsed and canonicalized."""
+
+
+@dataclass(frozen=True)
+class _WorkerFailure:
+    """A worker-side exception, carried back to the parent as data.
+
+    The original exception's class may not exist in the parent's import path in
+    every case (pyoxigraph's own exception types, for instance), so only its
+    message crosses the process boundary - re-raised here as a plain ValueError
+    rather than something that could fail to unpickle.
+    """
+
+    message: str
+
+
+def _canonicalize_worker(
+    nquads: bytes, connection: "mp.connection.Connection[bytes | _WorkerFailure, object]"
+) -> None:
+    """Parse, canonicalize and serialize *nquads* in a worker process.
+
+    Runs entirely inside the subprocess so the only thing crossing back to the
+    parent is the finished result - never a live pyoxigraph object, which would
+    not survive the process boundary regardless. Only the failures input can cause
+    are sent back with their reason; anything else ends the worker, and the parent
+    reports a worker that exits without a result.
+    """
+    try:
+        dataset = ox.Dataset()
+        for quad in ox.parse(input=nquads, format=ox.RdfFormat.N_QUADS):
+            dataset.add(quad)
+        dataset.canonicalize(ox.CanonicalizationAlgorithm.RDFC_1_0)
+        serialized = ox.serialize(list(dataset), format=ox.RdfFormat.N_QUADS) or b""
+        lines = sorted(serialized.decode("utf-8").splitlines())
+        result: bytes | _WorkerFailure = ("\n".join(lines) + "\n").encode("utf-8") if lines else b""
+    except (SyntaxError, ValueError) as error:
+        result = _WorkerFailure(f"{type(error).__name__}: {error}")
+    connection.send(result)
+    connection.close()
+
+
+def canonicalize_nquads(nquads: bytes, *, timeout: float = DEFAULT_TIMEOUT_SECONDS) -> bytes:
+    """Return *nquads* in RDFC-1.0 canonical form: sorted, newline-terminated N-Quads.
+
+    Runs in a fresh ``spawn``-started subprocess, killed outright if it does not
+    finish within *timeout* seconds - the only way to bound a blocking native call
+    that does not check back in with Python for a cooperative cancellation.
+
+    The parent waits for the result and for the worker's exit together. Waiting for
+    the exit alone would never end for a result larger than the pipe holds, since the
+    worker cannot finish sending it until the parent starts reading; waiting for the
+    result alone would never end for a worker that died without sending one.
+    """
+    ctx = mp.get_context("spawn")
+    receiver, sender = ctx.Pipe(duplex=False)
+    process = ctx.Process(target=_canonicalize_worker, args=(nquads, sender))
+    process.start()
+    sender.close()
+    try:
+        ready = mp.connection.wait([receiver, process.sentinel], timeout)
+        if not ready:
+            raise CanonicalizationTimeoutError(timeout)
+        try:
+            result = receiver.recv()
+        except EOFError:
+            process.join()
+            raise CanonicalizationError(
+                f"Canonicalization ended without a result (exit code {process.exitcode})"
+            ) from None
+    finally:
+        receiver.close()
+        if process.is_alive():
+            process.kill()
+        process.join()
+    if isinstance(result, _WorkerFailure):
+        raise CanonicalizationError(f"Failed to canonicalize RDF: {result.message}")
+    return bytes(result)
+
+
+def canonical_digest_bytes(nquads: bytes, *, timeout: float = DEFAULT_TIMEOUT_SECONDS) -> bytes:
+    """Return the SHA-256 digest of *nquads* in its RDFC-1.0 canonical form.
+
+    **This hash is a security use**: every digest a signature is made over goes
+    through here.
+    """
+    return hashlib.sha256(canonicalize_nquads(nquads, timeout=timeout)).digest()
+
+
+def _as_rdflib_term(term: ox.NamedNode | ox.BlankNode | ox.Literal) -> rdflib.term.Node:
+    """Translate one pyoxigraph term into the equivalent rdflib term, without normalizing it."""
+    if isinstance(term, ox.Literal):
+        language = term.language
+        return rdflib.Literal(
+            term.value,
+            lang=language,
+            datatype=None if language else rdflib.URIRef(term.datatype.value),
+            normalize=False,
+        )
+    if isinstance(term, ox.BlankNode):
+        return rdflib.BNode(term.value)
+    return rdflib.URIRef(term.value)
+
+
+def parse_without_normalizing(
+    data: bytes, rdf_format: ox.RdfFormat = ox.RdfFormat.N_QUADS
+) -> rdflib.Graph:
+    """Parse *data* into an rdflib.Graph without rewriting any literal's lexical form.
+
+    rdflib's own parser rewrites a recognized datatype's lexical form the moment it
+    constructs a Literal - an ``xsd:dateTime`` written ``...Z`` comes back ``...+00:00`` -
+    governed by a process-global flag (``rdflib.NORMALIZE_LITERALS``) this plugin must
+    never touch: Corporate Memory's DataIntegration runs every installed plugin in one
+    shared, concurrently-executing Python interpreter, so a global flip here would leak
+    into every other plugin sharing that process, or race against one running at the
+    same time. Parsing through pyoxigraph instead - which never rewrites a literal - and
+    translating its terms into rdflib's own term types (each ``Literal`` built with
+    ``normalize=False``, a per-instance argument rather than shared state) reads the same
+    bytes into the same graph structure without that side effect.
+    """
+    graph = rdflib.Graph()
+    for quad in ox.parse(input=data, format=rdf_format):
+        graph.add(
+            (
+                _as_rdflib_term(quad.subject),
+                _as_rdflib_term(quad.predicate),
+                _as_rdflib_term(quad.object),
+            )
+        )
+    return graph