v1.0.0

• Stable release. The public API and behaviour are those of 0.2.0, checked
  in the consumer pilot (cska-one, ecosystem-sandbox); from here on SemVer
  applies strictly.

v0.2.0

The pre-1.0 contract wave: every break below is deliberate, the API of this release is
meant to become 1.0 unchanged.

• BC: ExpandOptions is renamed ProcessingOptions (it configures every operation),
  and its constructor order is now base, documentUrl, expandContext, loader, maxDepth, maxBytes. Migration: rename the class; positional calls must switch to named arguments.
• Added ProcessingOptions::$maxDepth (default 512, the previous fixed cap) and
  $maxBytes (default no limit): guards for untrusted JSON strings, checked before and
  while decoding and during expansion (document too large, document nested too deeply).
  Depth now counts nested arrays and objects, as json_decode() does, not every value.
• BC: JsonLd::document() builds the node API from the node map (the structure
  flattening uses): embedded nodes without @id — a schema.org author — are now nodes,
  reachable through Node::reference(), and get blank node labels _:b0, _:b1, …;
  blank node labels of the input are relabeled the same way; nested @graphs become
  graphs of their own; a bare {"@id": …} reference is a node without properties.
  Node::properties()/toArray() hold references instead of embedded objects.
  Migration: follow reference() instead of reading embedded arrays; do not rely on
  input blank node labels.
• BC: Graph\Document's constructor is private; use Document::fromExpanded().
  Graph\Graph takes node objects in flattened form.
• Added Node::valueObjects() returning Graph\Value (value, language, datatype,
  direction): language-tagged values no longer lose their tag. Node::values() and
  friends accept single-valued members such as @index.
• BC: whatever a ContextLoaderInterface throws now surfaces as
  DocumentLoadingException with the loader's exception as the previous one:
  loading remote context failed for contexts (it was an InvalidContextException,
  and non-DocumentLoadingException throwables escaped unwrapped), loading document failed
  for a document URL (inside a term's scoped @context the W3C algorithm still reports
  invalid scoped context, now with the DocumentLoadingException as previous).
  Migration: catch DocumentLoadingException/JsonLdException
  instead of InvalidContextException or RuntimeException.
• BC: exceptions carry an error code: JsonLdException::errorCode(): ErrorCode (W3C
  JSON-LD 1.1 API codes plus recursive context inclusion and four processor codes:
  document nested too deeply, document too large, invalid N-Quads,
  canonicalization limit exceeded). The constructor is
  (ErrorCode $errorCode, ?string $message = null, ?Throwable $previous = null); the
  message defaults to the code. InvalidJsonException reports loading document failed
  with the JSON error as its message. Migration: branch on errorCode(), not on the
  message; custom loaders that throw DocumentLoadingException pass a code (or throw
  anything — it is wrapped).
• BC: Dataset::canonicalHash() now implements W3C RDF Dataset Canonicalization
  RDFC-1.0 (SHA-256 of the canonical N-Quads) instead of a home-grown algorithm, so hash
  values change; Dataset::equals() compares canonical N-Quads. Migration: recompute
  stored hashes.
• Added Dataset::canonicalNQuads(), Rdf\Canonicalizer (hash algorithm, work factor),
  Rdf\CanonicalForm (canonical N-Quads, issued identifiers, hash()),
  Rdf\HashAlgorithm (SHA-256, SHA-384) and Exception\CanonicalizationLimitException:
  poison graphs stop after workFactor × n Hash N-Degree Quads runs per blank node
  (default factor 4). The W3C rdf-canon suite passes 86 of 86 (64 evaluation, 21 map,
  1 negative), nothing skipped. Documented limitation of the specification: blank nodes that
  differ only in which named graph their relations live in tie, and the input quad order
  decides between equally valid canonical forms.
• BC: rdfDirection of JsonLd::toRdf()/fromRdf() is the enum RdfDirection
  (I18nDatatype, CompoundLiteral). Migration: 'i18n-datatype' →
  RdfDirection::I18nDatatype, 'compound-literal' → RdfDirection::CompoundLiteral.
• BC: the RDF vocabulary constants moved from Rdf\Literal to Rdf\Vocabulary
  (Literal::XSD_STRING → Vocabulary::XSD_STRING, likewise RDF_LANGSTRING, RDF_JSON,
  RDF_FIRST, RDF_REST, RDF_NIL, RDF_TYPE; new: XSD_BOOLEAN, XSD_INTEGER,
  XSD_DOUBLE, RDF_LIST, RDF_VALUE, RDF_LANGUAGE, RDF_DIRECTION, I18N).
• BC: Dataset::quads() is removed; read the Dataset::$quads property, as on the
  other RDF value objects.
• BC: internal methods left the public classes: Expander::expandRaw() and
  NQuads::formatTerm() are gone, Flattener no longer takes an Expander.
• Added Loader\MapContextLoader: a fixed URL → document map, exact match, no I/O.
• README states the scope (no compaction or framing in 1.0) and that the package reads
  JSON-LD rather than emitting schema.org markup.

v0.1.0

• JSON-LD 1.1 expansion, flattening, node API, RDF conversion (toRdf, fromRdf) and N-Quads.
• W3C JSON-LD API test suite: expand 368/368, flatten 54/55 (one compactArrays case skipped), toRdf 447/448 (one empty-object JSON literal case skipped), fromRdf 53/53. Compaction, framing and the 1.0 processing mode are not implemented.
• Context loading only through an injected ContextLoaderInterface; the default InlineContextLoader refuses URLs, PsrContextLoader takes a URL policy closure and a body size limit.
• Checked on 14 JSON-LD blocks collected from real pages (Wikipedia, YouTube, BBC, The New York Times, IMDb, Kommersant, Championat and others): expansion and flattening give the same RDF, N-Quads and fromRdf round-trip.
• Mutation score 98.3% (minMsi 92).