rumil_tokens 0.10.0

Lossless source code tokenizer built on Rumil parser combinators. Classified token spans for syntax highlighting. Built-in grammars for Dart, Scala, YAML, JSON, and shell.

rumil_tokens

Lossless source code tokenizer built on Rumil parser combinators. Classifies source text into typed token spans: keywords, strings, comments, numbers, types, annotations, operators, variables, and punctuation. Token streams are lossless; concatenating token.text across a stream reconstructs the input exactly. Built-in grammars for Dart, Scala, YAML, JSON, and shell.

Usage

import 'package:rumil_tokens/rumil_tokens.dart';

final tokens = tokenize('final x = 42; // answer', dart);
for (final token in tokens) {
  print('${token.runtimeType}: ${token.text}');
}

Use a built-in grammar (dart, scala, yaml, json, shell) or define your own:

const rust = LangGrammar(
  name: 'rust',
  keywords: ['fn', 'let', 'mut', 'if', 'else', 'match', 'impl', 'struct'],
  types: ['i32', 'u64', 'String', 'Vec', 'Option', 'Result', 'bool'],
  lineComment: '//',
  blockComment: ('/*', '*/'),
  stringDelimiters: ['"'],
  annotationPrefix: '#',
);

final tokens = tokenize(source, rust);

Look up a grammar by name:

final grammar = grammarFor('dart'); // returns null for unknown languages

Enumerate the built-in grammars:

for (final g in builtinGrammars) {
  print(g.name);
}

For hot paths (REPL highlighting, large files), build the parser once and reuse it across calls:

final dartTokenizer = buildTokenizer(dart);
for (final source in sources) {
  final result = dartTokenizer.run(source);
  // ...
}

Lossless property

Concatenating token.text for every token reconstructs the original source:

assert(tokens.map((t) => t.text).join() == source);

Positions

For tooling that needs byte offsets, use tokenizeSpans:

final spans = tokenizeSpans(source, dart);
for (final s in spans) {
  print('[${s.start}, ${s.end}) ${s.token}');
  assert(source.substring(s.start, s.end) == s.token.text);
}

Spanned<Token> is an extension type over (Token, int, int). The [start, end) interval is half-open; spans are contiguous (spans[i].end == spans[i+1].start) and anchored (spans.first.start == 0, spans.last.end == source.length).

Grammar coverage

Known limitations (see CHANGELOG.md):

• YAML block scalars (|, >) tokenize the indented body as regular YAML content rather than one string literal.
• Dart string interpolation ("$x", "${expr}") remains one StringLit.
• Shell braced expansions do not balance nested braces.
• Heredoc body is one StringLit.
• Nested generic close renders the outer >> as right-shift.

Part of the rumil-dart monorepo.