CSS Parser API

1. Introduction

Common data-interchange / parsing formats are very valuable for reducing the learning curve of new languages, as users get to lean on their existing knowledge of the format when authoring and only have to newly learn the specifics of the language. This is why generic parsing formats like XML or JSON have become so popular.

The CSS language could benefit from this same treatment; a number of languages and tools rely on CSS-like syntax to express themselves, but they usually rely on ad-hoc parsing (often regex-based) which can be relatively fragile, and might break with CSS practices in interesting syntax corner cases. Similarly, CSS syntax is increasingly used in places like attribute values (such as the sizes attribute, or most of the SVG presentation attributes), and custom elements wanting to do the same thing similarly have to rely on ad-hoc parsing right now.

To help with these sorts of cases, this spec exposes the [css-syntax-3] parsing algorithms, and represents their results in a mildly-typed representation, simpler and more abstract than what [css-typed-om-1] does for CSS properties.

2. Parsing API

typedef (DOMString or ReadableStream) CSSStringSource;
typedef (DOMString or CSSStyleValue or CSSParserValue) CSSToken;

partial namespace CSS {
  Promise<sequence<CSSParserRule>> parseStylesheet(CSSStringSource css, optional CSSParserOptions options = {});
  Promise<sequence<CSSParserRule>> parseRuleList(CSSStringSource css, optional CSSParserOptions options = {});
  Promise<CSSParserRule> parseRule(CSSStringSource css, optional CSSParserOptions options = {});
  Promise<sequence<CSSParserRule>> parseDeclarationList(CSSStringSource css, optional CSSParserOptions options = {});
  CSSParserDeclaration parseDeclaration(DOMString css, optional CSSParserOptions options = {});
  CSSToken parseValue(DOMString css);
  sequence<CSSToken> parseValueList(DOMString css);
  sequence<sequence<CSSToken>> parseCommaValueList(DOMString css);
};

dictionary CSSParserOptions {
  object atRules;
  /* dict of at-rule name => at-rule type
     (contains decls or contains qualified rules) */
};

parseCommaValueList() is in Syntax, and thus here, because it’s actually a very common operation. It’s trivial to do yourself (just call parseValueList() and then split into an array on top-level commas), but comma-separated lists are so common that it was worthwhile to improve spec ergonomics by providing a shortcut for that functionality. Is it worth it to provide this to JS as well?

Do we handle comments? Currently I don’t; Syntax by default just drops comments, but allows an impl to preserve information about them if they want. Maybe add an option to preserve comments? If so, they can appear *anywhere*, in any API that returns a sequence.

What do we do if an unknown at-rule (not appearing in the atRules option) shows up in the results? Default to decls or rules? Or treat it more simply as just a token sequence?

Parsing stylesheets/rule lists should definitely be async, because stylesheets can be quite large. Parsing individual properties/value lists should definitely be sync, because they’re small and it would be really annoying. Parsing a single rule, tho, is unclear—is it large enough to be worth making async, or is it too annoying to be worth it?

3. Parser Values

[Exposed=Window]
interface CSSParserRule {
  /* Just a superclass. */
};

[Exposed=Window]
interface CSSParserAtRule : CSSParserRule {
  constructor(DOMString name, sequence<CSSToken> prelude, optional sequence<CSSParserRule>? body);
  readonly attribute DOMString name;
  readonly attribute FrozenArray<CSSParserValue> prelude;
  readonly attribute FrozenArray<CSSParserRule>? body;
  /* nullable to handle at-statements */
  stringifier;
};

[Exposed=Window]
interface CSSParserQualifiedRule : CSSParserRule {
  constructor(sequence<CSSToken> prelude, optional sequence<CSSParserRule>? body);
  readonly attribute FrozenArray<CSSParserValue> prelude;
  readonly attribute FrozenArray<CSSParserRule> body;
  stringifier;
};

[Exposed=Window]
interface CSSParserDeclaration : CSSParserRule {
  constructor(DOMString name, optional sequence<CSSParserRule> body);
  readonly attribute DOMString name;
  readonly attribute FrozenArray<CSSParserValue> body;
  stringifier;
};

[Exposed=Window]
interface CSSParserValue {
  /* Just a superclass. */
};

[Exposed=Window]
interface CSSParserBlock : CSSParserValue {
  constructor(DOMString name, sequence<CSSParserValue> body);
  readonly attribute DOMString name; /* "[]", "{}", or "()" */
  readonly attribute FrozenArray<CSSParserValue> body;
  stringifier;
};

[Exposed=Window]
interface CSSParserFunction : CSSParserValue {
  constructor(DOMString name, sequence<sequence<CSSParserValue>> args);
  readonly attribute DOMString name;
  readonly attribute FrozenArray<FrozenArray<CSSParserValue>> args;
  stringifier;
};

Trying to be as useful as possible, without exposing so many details that we’re unable to change tokenization in the future. In particular, whitespace and delims all get parsed into DOMStrings. Am I succeeding at this goal?