Move css-value-parsing.js, generic-property-descriptor.js, and shorthand-properties.js into cssom/helpers/. Convert all -impl.js files and multi-export modules from module.exports to exports.x style. Remove lazy-loading workarounds in -impl.js files and css-parser.js by reordering interfaces.js so CSSRuleList and MediaList load before their dependents.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

For cases where a getter just returns an underscored property, assign the public property directly in the constructor instead. The generated wrapper has its own getter to preserve encapsulation, so this removes a level of indirection. Trivial setters (that just assign) are also removed.

Getters are kept where needed: conditionText on CSSConditionRule (overridden by CSSMediaRule), style (non-trivial setter), selectorText on CSSPageRule (validation in setter), and CSSCounterStyleRule descriptors (delegated through _descriptors object).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The circular dependency (css-parser.js → generated wrappers → impl files → css-parser.js) is now broken by having the impl files keep a module reference to css-parser.js instead of destructuring. Property access on the module object is resolved at call time, by which point all modules have finished loading.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Remove over-defensive defaults (|| "", || null, || []) from privateData field access in all impl constructors except CSSStyleSheet and StyleSheet, which need defaults because CSSStyleSheet has an IDL constructor. All other impls are only created via css-parser.js which always provides all required values.

Also remove dead code: StyleSheet's href, title, and media fields were read from privateData but never passed by any caller. href and title are now initialized to null directly, and media was removed since CSSStyleSheet always creates its own MediaList.

In CSSMediaRule, read this._conditionText (set by super) instead of privateData.conditionText, which is cleaner.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Move the parsing of @container prelude into translateContainerRule() so the impl class receives containerName and containerQuery as separate privateData fields, matching the spec's architecture. The conditionText getter now derives its value from those two fields per the spec definition. This also removes the fallback heuristic parser and the css-tree dependency from CSSContainerRule-impl.js.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Rule impls were storing wrappers for parentRule and parentStyleSheet, requiring constant implForWrapper/wrapperForImpl conversions in internal code. Store impls instead, since the generated WebIDL wrappers already handle the impl-to-wrapper conversion via tryWrapperForImpl when exposing these to JavaScript.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

All csstree usage now goes through a single facade module that applies the @csstools/css-syntax-patches-for-csstree patches. An ESLint rule prevents direct imports of css-tree or the patches package from anywhere except the facade.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The spec's "insert a CSS rule" and "remove a CSS rule" algorithms operate on a CSSRuleList, so implement them as _insert()/_insertParsed() and _remove() on CSSRuleList-impl.js. CSSGroupingRule and CSSStyleSheet now delegate to these shared methods instead of duplicating the logic.

Also use isImpl() for rule type checks instead of comparing .type numeric constants, reject @import in constructed stylesheets per spec, and add TODOs for origin-clean and disallow-modification flag checks.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

replace() now sets the flag during its operation, preventing concurrent insertRule/deleteRule/replaceSync/replace calls per spec.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CSSImportRule.media now delegates to this.styleSheet.media per spec, instead of creating a separate MediaList. The @import media text is passed to the child CSSStyleSheet constructor.

Replace ad-hoc CSS string/URL escaping with csstree.string.encode(), which implements the CSSOM "serialize a string" algorithm. This affects CSSImportRule, CSSNamespaceRule, and css-value-parsing.js helpers.

Re-export csstree.string from the patched-csstree facade since fork() drops standalone utilities.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The getter normalizes "from" to "0%" and "to" to "100%". The setter validates the input is a valid keyframe selector (comma-separated list of "from", "to", or percentages) and throws SyntaxError for invalid values per spec.

CSSKeyframesRule's deleteRule/findRule now also normalize their select argument for correct matching.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Replace the ad-hoc regex with csstree.lexer.matchType("keyframe-selector"), which correctly enforces the [0,100] range and accepts all valid CSS number formats (.5%, 1e1%, etc.). Normalize numbers via parseFloat() so that e.g. .5% becomes 0.5% and 100.0% becomes 100%, matching Chrome's behavior.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

StyleSheetList._list was storing wrappers, requiring unnecessary implForWrapper conversions in style-rules.js every time computed styles were calculated. Switch to storing impls throughout: createStylesheet now returns an impl, addStylesheet/removeStylesheet pass impls, and scanForImportRules works directly with rule impl arrays.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Add serializeCustomIdent helper for <custom-ident> serialization, using csstree.ident.encode() normally and csstree.string.encode() for excluded keywords (CSS-wide keywords, "default", and production-specific exclusions). Apply it to CSSKeyframesRule, CSSCounterStyleRule, and CSSContainerRule.
- Fix CSSKeyframesRule indexed getter to return undefined instead of null for out-of-bounds access.
- Remove incorrect type getters from CSSLayerBlockRule (was 16) and CSSLayerStatementRule (was 17); the base CSSRule class already returns 0, which is correct for rules with no spec-defined constant.
- Update comments in CSSLayerStatementRule.webidl and CSSNestedDeclarations.webidl.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Extract evaluateMediaList() from the inline media-matching logic in style-rules.js, and use it both there and in a new matches getter. This also fixes @media rules with "all" or empty media lists not being applied during style computation.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…xt setter, and misc fixes

- Add CSSSupportsRule's matches getter, evaluating @supports conditions against csstree's lexer (handles and/or/not and nested conditions).
- Add CSSStyleRule's selectorText setter with validation via csstree — invalid selectors are silently ignored per spec.
- Fix parseInt to Number in CSSStyleDeclaration's item().
- Update comments in CSSPageRule.webidl, CSSStyleRule.webidl.
- Bump @domenic/eslint-config to ^5.1.0.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Use csstree to validate individual media queries instead of naive comma-splitting. Invalid queries become "not all" per spec. appendMedium/deleteMedium compare by canonical serialization.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Split `parseStylesheet` into `parseStyleSheet` (creates new sheet) and `parseIntoStyleSheet` (populates existing sheet), with a shared `parseRules` helper.
- Replace `createStylesheet` with `createStyleSheetForElement` (for `<style>` and `<link>`) and `parseImportedStyleSheet` (for `@import`), each deriving its own parameters instead of accepting an opaque options bag.
- Rename `addStylesheet`/`removeStylesheet`/`fetchStylesheet` to camelCase `addStyleSheet`/`removeStyleSheet`/`fetchStyleSheet`.
- Move `media`, `href`, `title`, and `ownerNode` from CSSStyleSheet to the StyleSheet base class, and uncomment `ownerNode` in StyleSheet.webidl.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- parseKeyframeRule: reject input with trailing garbage, extra at-rules, or multiple keyframes by checking total children count instead of only counting Rule nodes
- translateImportRule/translateNamespaceRule: return null for invalid preludes missing a URL, instead of manufacturing rules with empty href/namespaceURI
- populateGroupingRuleChildren: filter out CSSNestedDeclarations in non-nesting context, since bare declarations inside top-level grouping rules are invalid

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

csstree already rejects preludes without a valid URL/string by throwing, so the regex-based href extraction in the catch branch could only manufacture rules from inputs the real parser rejected. With the !href validation guard, it was also dead code in practice.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The loop over csstree children was assigning prefix for every Identifier, so trailing junk after the URL (e.g. `@namespace foo url("a") bar;`) got silently absorbed. Validate the expected grammar (<prefix>? <url>) and reject preludes with extra children.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Instead of creating CSSNestedDeclarations in non-nesting contexts and filtering them out in the caller, have parseRawAsNestedRules skip the parseRawAsDeclarations path entirely when the caller doesn't want declarations.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…observable

The spec says replace() should set the disallow modification flag, then do the parsing "in parallel" before unsetting it and resolving the promise. The previous implementation was an async function that did everything synchronously, so the flag was set and unset in the same microtask and no code could ever observe it as true.

Now replace() sets the flag synchronously and defers the actual work (clear rules, parse, unset flag, resolve) to a queueMicrotask callback. This makes insertRule(), deleteRule(), replaceSync(), and another replace() correctly throw/reject NotAllowedError while a replace() is in progress.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Verifies that garbage CSS input clears existing rules without throwing, mixed valid/invalid input keeps valid rules, and the sheet remains usable after parse errors. Both methods behave identically (as expected by spec).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Drop @import and @namespace inside grouping rules (they're only valid at stylesheet top level)
- Detach old rules (nullify parentRule/parentStyleSheet recursively) when replace()/replaceSync() clears the sheet, via a new CSSRuleList _clear() method
- Exempt @layer statement rules from the "other rules exist" check when deleting @namespace, matching the exemption already present for insertion
- Detach removed keyframe rules in CSSKeyframesRule deleteRule()

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

A holdover from cssstyle

…test files

Add tests for style.sheet inside shadow roots (known failing, #3645) and programmatically setting title on <style> elements (#3634). Rename tuwpt files to use properly-cased class names (e.g. CSSImportRule- instead of cssimportrule-).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Convert _constructed, _disallowModification, _keyText, and _selectorText to #private fields since they are only accessed within their own class. Cross-file-accessed properties like _style, _conditionText, and _list remain as underscore-prefixed.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add the disabled IDL attribute that delegates to the associated stylesheet's disabled flag per the HTML spec. Update WPT expectations with descriptive reasons for tests we can't fix yet.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>