Webref

Description

This repository contains machine-readable references of CSS properties, definitions, IDL, and other useful terms that can be automatically extracted from web browser specifications (see a list of projects known to use the data). The contents of the repository are updated automatically every 6 hours (although note information about published /TR/ versions of specifications are updated only once per day).

Specifications covered by this repository are technical Web specifications that appear in browser-specs.

The main branch of this repository contains automatically-generated raw extracts from web browser specifications. These extracts come with no guarantee on validity or consistency. For instance, if a specification defines invalid IDL snippets or uses an unknown IDL type, the corresponding IDL extract in this repository will be invalid as well.

The curated branch contains curated extracts. Curated extracts are generated from raw extracts in the ed folder by applying manually-maintained patches to fix invalid content and provide validity and consistency guarantees. The curated branch is updated automatically whenever the main branch is updated, unless patches need to be modified (which requires manual intervention). Curated extracts are published under https://w3c.github.io/webref/ed/.

Additionally, subsets of the curated content get manually reviewed and published as NPM packages on a weekly basis:

• @webref/idl contains a curated version of the ed/idl folder.
• @webref/css contains a consolidated and curated version of the ed/css folder.
• @webref/elements contains a curated version of the ed/elements folder.
• @webref/events contains a consolidated and curated version of the ed/events folder.

Important: The curated extracts only contain data for specifications that are in good standing (to keep the number of manually-maintained patches minimal and manageable). The NPM packages only contain curated extracts of specifications that are in good standing and that target web browsers.

Important: Unless you are ready to deal with invalid content, we strongly recommend that you process contents of the curated branch or NPM packages instead of raw content in the main branch.

Available extracts

This repository contains raw and curated information about nightly versions of Web specifications in the ed folder, as well as raw information about the released version (for /TR/ specifications) in the tr folder.

Note: The tr folder only contains information about released specifications. Specifications that have not been published as /TR/ documents (such as WHATWG specifications or Community Group reports) do not appear under the tr folder in particular.

More often than not, released versions of specifications are much older than their nightly version. Data in the tr folder is more invalid/inconsistent than data in the ed folder as a result. Additionally, no attempt is being made at curating data in the tr folder, use the tr folder at your own risk!

The following subfolders in the curated branch contain individual machine-readable JSON or text files generated from specifications:

• ed/backrefs: Back references to definitions of a specification from other specifications. One file per specification. This is a curated, derived view built from dfns and links extracts.
• ed/cddl: CDDL modules. If the specification defines a single CDDL module, one file gets created for the specification. If it defines multiple CDDL modules, one file gets created per CDDL module, plus one file named [shortname]-all.cddl with all CDDL definitions.
• ed/css: CSS terms (properties, descriptors, value spaces). One file per specification series.
• ed/dfns: <dfn> terms, along with metadata such as linking text, access level, namespace. One file per specification.
• ed/elements: Markup elements defined, along with the interface that they implement. One file per specification.
• ed/headings: Section headings. One file per specification.
• ed/idl: Raw WebIDL index. One file per specification series.
• ed/idlnames: WebIDL definitions per referenceable IDL name. One file per IDL name.
• ed/idlnamesparsed: Parsed WebIDL structure of definitions in the idlnames folder. One file per IDL name.
• ed/idlparsed: Parsed WebIDL structure of definitions in the idl folder. One file per specification.
• ed/ids: Fragments defined in the specification. One file per specification.
• ed/links: Links to other documents, along with targeted fragments. One file per specification.
• ed/refs: Normative and informative references to other specifications. One file per specification.

Individual files are named after the shortname of the specification, or after the shortname of the specification series for CSS definitions and raw IDL files. Individual files are only created when needed, meaning when the specification actually includes relevant terms.

The ed/index.json file contains the index of specifications that have been crawled, and relative links to individual files that have been created.

This repository uses Reffy, a Web spec exploration tool, to crawl the specifications and generate the data. In particular, the data it contains is the result of running Reffy. The repository does not contain any more data.

Raw WebIDL extracts are used in web-platform-tests, please see their interfaces/README.md for details.

Curation guarantees

Data curation brings the following guarantees.

Web IDL extracts

• All IDL files can be parsed by the version of webidl2.js referenced in package.json.
• WebIDL2.validate passes with the exception of the "no-nointerfaceobject" rule about [LegacyNoInterfaceObject], which is in wide use.
• All types are defined by some specification.
• All extended attributes are defined by some specification.
• No duplicate top-level definitions or members.
• No missing or mismatched types in inheritance chains.
• No conflicts when applying mixins and partials.

CSS extracts

Important: Starting with @webref/css version 7, the package contains the consolidated ed/css.json file, and no longer contains CSS extracts per spec. See the changelog for "How to upgrade" considerations.

The consolidated ed/css.json file, released in the @webref/css package, comes with the following guarantees:

• The file follows the relevant JSON schema in the version of Reffy that was used to crawl the specs.
• All syntax values (the syntax keys) can be parsed by the version of CSSTree set in peerDependencies in package.json.
• Feature names (the name keys) are unique per type provided that the for key is also taken into account for functions and types.
• CSS features targeted by for, legacyAliasOf, longhands and resetLonghands keys are guaranteed to exist in the package.
• All features have an href key that targets the CSS spec that defines the feature. When the feature is extended across CSS specs, this URL targets the base definition.
• All URLs (without fragment) that appear in href and extended keys are guaranteed to exist in the version of web-specs that was used to identify the specs to extract data from. The exact version that was used is not yet surfaced in the data though. In most cases, it should be the latest published version at the time when the data is released.

The consolidated file is generated from curated extracts in the ed/css folder. These extracts, released in the @webref/css package until version 7, come with the following guarantees:

• All extracts follow the relevant JSON schema in the version of Reffy that was used to crawl the specs.
• All values in CSS files can be parsed by the version of CSSTree used in peerDependencies in package.json.
• No duplicate definitions of entries in CSS files provided that CSS extracts of delta specs are not taken into account (such extracts end with -n.json, where n is a level number). The term "entries" includes CSS properties, at-rules, selectors, types and functions. Please note that specs may still extend entries defined elsewhere (to define new values for CSS properties, or new selectors for at-rules).
• CSS extracts contain a base definition of all CSS properties that get extended by other CSS property definitions (those for which newValues is set).
• All entries in CSS files that do not extend a base definition link back to their actual definition in the spec. In other words, all entries under properties[], properties[].values[], selectors[], atrules[] and values[] have an href key that contains an absolute URL with fragment, except properties that that have a newValues key, at-rules that neither have a prose nor a value key, and definitions of a delta spec that completely override a definition in a previous level.

Elements extracts

• All extracts follow the relevant JSON schema in the version of Reffy that was used to crawl the specs.
• All Web IDL interfaces referenced by elements exist in Web IDL extracts.
• All elements link back to their definition in the spec.

Events extracts

The consolidated ed/events.json file, released in the @webref/events package, comes with the following guarantees:

• The file follows the relevant JSON schema in the version of Reffy that was used to crawl the specs.
• All events have a type attribute that match the name of the event
• All events have a interface attribute to describe the interface used by the Event. The Web IDL interface exists in the latest version of the @webref/idl package at the time the @webref/events package is released, and represents an actual interface (i.e. not a mixin).
• All events have a targets attribute with a non-empty list of target interfaces on which the event may fire. All Web IDL interfaces in the list exist in the latest version of the @webref/idl package at the time the @webref/events package is released, and represent an actual interface (i.e. not a mixin).
• The bubbles attribute is always set to a boolean value for target interfaces that belong to a bubbling tree (DOM, IndexedDB, Serial API, Web Bluetooth).
• The bubbles attribute is only set for target interfaces that belong to a bubbling tree.
• The bubblingPath attribute is only set for target interfaces on which the event bubbles.
• The targets attribute contains the top most interfaces in an inheritance chain, unless bubbling conditions differ. For instance, the list may contain { "target": "Element", "bubbles": true } but not also { "target": "HTMLElement", "bubbles": true } since HTMLElement inherits from Element.
• For target interfaces that belong to a bubbling tree, the targets attribute only contains the deepest interface in the bubbling tree on which the event may fire and bubble. For instance, the list may contain { "target": "HTMLElement", "bubbles": true }, but not also { "target": "Document" } since event would de facto fire at Document through bubbling.

CDDL extracts

• All CDDL files pass CDDL analysis by the version of Strudy referenced in package.json. Said differently, all CDDL files can be parsed by the version of cddlparser referenced by Strudy.

The CDDL extracts are not released in an NPM package for the time being.

Backrefs extracts

• All extracts follow the relevant JSON schema in the version of Reffy that was used to crawl the specs.
• The extracts contain both terms defined as private and public.
• Every backrefs[] entry corresponds to a definition that exists in the defining spec's dfns extract (matched by href).
• id, href, linkingText, type, for, and access are copied from that definition so the extract is usable on its own.
• referencedBy is a non-empty list of distinct referencing specifications. The defining specification itself never appears in referencedBy.
• Every referencedBy[].shortname identifies a specification present in the crawl (and thus listed in ed/index.json).

The backrefs extracts are not released in an NPM package for the time being.

Known consumers

The following projects are known to use webref data:

• MDN content management system, yari
• Web Platform Tests
• Open Web Docs MDN Browser Compat Data collector
• bikeshed, a specification authoring tool
• ReSpec, a specification authoring tool
• "DOM" Bindings for TypeScript
• browser API bindings for Dart
• the Pulsar text editor
• PostCSS preset env plugin
• strudy
• nodysseus
• WebIDLPedia
• Webdex
• Gost-DOM, a headless browser for Go (early prototype available).
• Gost-DOM Webref exposing parts of the data as native Go objects.

Using webref data in a project that is not yet in the list? Let us know!

Potential spec anomalies

Webref extracts are analyzed with Strudy to detect potential spec content anomalies such as broken links or invalid constructs, and report them as issues in the repository that hosts the spec.

Global analyses used to be published in the w3c/webref-analysis repository. That repository was archived in August 2024.

How to suggest changes or report an error

Feel free to raise issues in this repository as needed. Note that most issues likely more directly apply to underlying tools:

• Errors in the data are most likely caused by bugs or missing features in Reffy, which is the tool that crawls and parses specifications under the hoods. If you spot an error, please report it in Reffy's issue tracker.
• If you believe that a spec is missing from the list, please check browser-specs and report it there.

Development notes

GitHub Actions workflows are used to automate most of the tasks in this repo.

Data update

• Update ED report - crawls the nightly version of specifications and updates the contents of the ed folder. Workflow runs every 6 hours. Specifications that have not been modified since last crawl are skipped, unless the version of Reffy changed in the meantime. A typical crawl takes a few minutes to complete. A full crawl takes up to 12mn.
• Update TR report - crawls the released version of specifications and updates the contents of the tr folder. Workflow runs once per day. A typical crawl takes a few minutes to complete. A full crawl takes up to 8mn.
• Curate data & Prepare package PRs - runs whenever crawled data gets updated and updates the curated branch accordingly (provided all tests pass). The job also creates pull requests to release new versions of NPM packages when needed. Each pull request details the diff that would be released, and bumps the package version in the relevant packages/xxx/package.json file.
• Clean up abandoned files - checks the contents of repository to detect orphan crawl files that are no longer targeted by the latest crawl's result and creates a PR to delete these files from the repository. Runs once per week on Wednesday. The crawl workflows does not delete these files automatically because crawl sometimes fails on a spec due to transient network or spec errors.
• Test - runs tests on pull requests.
• Clean patches when issues/PR are closed - drops patches that no longer need to apply because underlying issues got fixed. Runs once per week.

Releases to NPM

• Publish @webref/xx package if needed - publishes a new version of the @webref/css, @webref/elements, @webref/events or @webref/idl package to NPM, tags the corresponding commits on the main and curated branches, and updates the relevant @webref/xxx@latest tag to point to the right commit on the curated branch. Runs whenever a pre-release PR is merged. Note that the released version is the version that appeared in packages/css/package.json, packages/elements/package.json, packages/events/package.json or packages/idl/package.json before the pre-release PR is merged.
• @webref/xx release: Request review of pre-release PR - assigns reviewers to NPM package pull requests. Runs once per week.