Skip to content

Latest commit

 

History

History
119 lines (104 loc) · 7.48 KB

CHANGELOG.md

File metadata and controls

119 lines (104 loc) · 7.48 KB

Changelog

[0.3.0] - 2026-11-13

Added

  • The blocks of a document that share an identifier (a JSON-LD @id, a microdata itemid) describe one node for Google, whatever their format, with the union of their properties. The audit now merges them into one item, validated as a whole: a property one block omits is no longer reported missing when another block provides it, and the issues of the node are reported once. Audit::getItems() lists the items, each with its node and its source blocks, which keep their own audit; Audit::getTypes() still lists the blocks. Values the blocks disagree on are a warning of the merged item (validator Merge, AuditOptions::VALIDATOR_MERGE), since Google keeps them all. The merge is on by default; pass mergeNodesSharingAnIdentifier: false to Validator::audit(), or --no-merge to castor validate and castor check, to audit every block on its own.
  • castor qa:earl runs the W3C test suites and writes the EARL conformance report consolidated by the W3C JSON-LD implementation report. The W3C manifests are now kept with the downloaded test suites.
  • BENCHMARKS.md documents the benchmark figures and what each of them measures. castor qa:bench:validators now also runs the worker benchmark, which audits a stream of distinct documents and is the figure to quote for a crawler.

Fixed

  • Framing in JSON-LD 1.1 mode dropped a property whose only value was a node reduced to a pruned blank node identifier (W3C test frame#tp046): the empty node object is now kept, as {}.
  • Microdata: a top-level item (an itemscope without itemprop) placed inside another item's element was dropped, as was the content attribute of any element but <meta> (<link itemprop="name" content="...">, which has no href). Both are common on real pages, and Google reads them.
  • The W3C manifest entries that reuse the input of another entry with a different expected output (compact#ta038, frame#tp020, #tp021, #tp046 and #tp049) are now run. Every test that applies to a JSON-LD 1.1 processor passes.
  • Google validator: the address sub-properties of an event location (streetAddress, addressRegion...) are no longer reported when missing: the documentation only shows them in an example, and Google does not report them.
  • Google validator: the schema.org subtypes of a documented type are now validated with its rules, as Google does (a BusinessEvent is an Event, a Restaurant a LocalBusiness, a TechArticle an Article). The rule files can refuse subtypes (ItemList) or exclude some of them (Event excludes CourseInstance, PublicationEvent...).
  • Google validator: an entity nested under a property its parent does not document (an event in the about of an article, a video in a course) was never validated. It is now validated as a rich result of its own, as Google reports it, when its type is one Google documents on its own; the types Google only documents as parts of another one (Comment, Answer, AggregateRating, ImageObject...) are left alone.

[0.2.0] - 2026-11-12

Changed

  • Schema.org vocabulary upgraded from 30.0 to 30.1. This release promotes most of the pending.schema.org terms to the core vocabulary (jobTitle, Legislation, DefinedTerm, MemberProgram...): they are no longer reported as pending vocabulary usage.
  • Google validator: VideoObject now recommends a creator (or author), which must have a name or an alternateName, and accepts the LikeAction, CommentAction and ShareAction interaction types, following the Google documentation.
  • The FAQ rich result is marked as retired by Google. FAQPage is still validated, for backward compatibility.
  • Google validator: merchant listings now cover every property the Google documentation lists. The product-level recommendations (category, gtin, sku, size, color, hasCertification...), the offer-level ones (itemCondition, priceValidUntil, url, validFrom, validThrough) and the nested shipping, return, unit pricing, size, audience, certification and 3D model structures are checked. An offer price requires a priceCurrency, a finite return window requires merchantReturnDays, and return shipping fees require returnShippingFeesAmount.
  • Google validator: every structured data type is aligned with the current Google documentation. A recommended property is reported when missing, unless Google makes it conditional ("if applicable", "only recommended for..."): such a property is still checked when present, and enforced by a special rule when the document tells whether it applies. Notable changes: new recommendations (job posting experience, local business department, member program and tier URLs, product snippet notes and aggregate offers, recipe and step videos, vacation rental details), new nested structures (organization return and shipping policies, video key moments and live badge, beta carousel items, recipe sections), and requirements relaxed where Google relaxed them (course provider, event description and image, recipe ingredients and instructions). Most organization, profile page, discussion forum and Q&A recommendations are conditional and no longer reported when missing.

Fixed

  • The static schema.org context is now generated from the context schema.org publishes with each release, instead of being rebuilt from the vocabulary definition file. Documents expand as with "@context": "https://schema.org": schema: compact IRIs expand to http://schema.org/ (they expanded to https://schema.org/), and the definition file's own prefixes can no longer shadow a schema.org term (such as geo since 30.1).
  • Google validator: a property accepting either a data type or a structured type (such as "Text or CategoryCode") now validates the nested properties of an object value. The ImageObject of an article image, the HowToStep of recipe instructions and an object breadcrumb item were never checked so far.
  • JSON-LD script tags whose type attribute has no quotes (valid HTML, and what minifiers produce) or has whitespace around the = are no longer ignored.
  • A data-type="application/ld+json" attribute no longer makes a script tag count as JSON-LD.

[0.1.0] - 2026-08-17

Added

  • Schema.org validator, covering the schema.org vocabulary (version exposed at runtime as SchemaOrgValidator::VOCABULARY_VERSION).
  • Google validator, reporting whether a document is eligible for Google Rich Results.
  • Extraction of JSON-LD, microdata and RDFa structured data from HTML documents.
  • The four W3C JSON-LD 1.1 algorithms: expansion, compaction, flattening and framing, each validated against the official W3C test suites.
  • Deny-by-default remote @context loading through DocumentLoaderInterface, HttpDocumentLoader and RemoteContextPolicy. No outbound request is made unless the application opts in and allow-lists hosts explicitly.
  • Command-line interface (via Castor) to validate documents and run the algorithms.

Security

  • Hostile documents can no longer crash the process: input nested more deeply than a fixed limit is refused before the parsed structure graph is built, and a top-level scalar document is reported invalid instead of raising an uncaught error.
  • Remote-context loading enforces the policy on every redirect hop (not only on the final URL), refuses URLs carrying userinfo or a non-default port, and streams responses under a hard size cap.