Breaking Changes


  • Chevrotain is now ESM only. If you need to use Chevrotain 11.0.0+ in a CommonJS project, either:
  • UMD bundles (e.g: chevrotain.js/chevrotain.min.js) were removed.
  • ESM bundles have been moved from lib_esm to lib folder, so CDN urls will break, e.g:
    • New:
    • Old: (now broken).


  • Dropped support for legacy ES5.1 runtimes (e.g: IE11) The minimum ECMAScript version needed to run Chevrotain is now ES2015 (ES6). This should not affect anyone running on a modern engine,
    meaning modern NodeJS versions or popular evergreen browsers.

  • Various TypeScript signatures are now more accurate and strict which could potentially cause compilation errors with some grammars implemented in TypeScript, e.g:

    • OPTION methods now return OUT | undefined instead of just OUT.
    • A RULE implementation function in CstParser is now defined as returning void.
    • The ARGS for SubruleMethodOpts options type is now better enforced via generics in the subrule definition.


  • Custom APIs feature has been deprecated and removed. This means the following public APIs no longer exist:
    • resolveGrammar
    • validateGrammar
    • assignOccurrenceIndices
    • defaultGrammarValidatorErrorProvider
    • defaultGrammarResolverErrorProvider
    • IGrammarValidatorErrorMessageProvider
    • IGrammarResolverErrorMessageProvider
    • IParserDefinitionError
    • generateParserFactory
    • generateParserModule


  • Chevrotain now uses the package.json exports field, as specified in the documentationopen in new window:

    Warning: Introducing the "exports" field prevents consumers of a package from using any entry points that are not defined, including the package.json (e.g. require('your-package/package.json'). This will likely be a breaking change.


  • The Parser's default maxLookahead was reduced to 3. This could cause ambiguity issues in existing Parsers. The pre 7.0.0 behavior can be reproduced by passing an explicit maxLookahead

    class MoreLookaheadParser extends CstParser {
      constructor() {
        super([], {
          maxLookahead: 4,
        // ...

    It is also possible (and recommended) to increase the maxLookahead for a specific DSL method rather than globally for all. See relevant issueopen in new window.

  • The soft deprecated Parser class has been fully removed, use CstParser or EmbeddedActionsParser instead. The choice depends on if your parser outputs a CSTopen in new window or not. the outputCst property of the IParserConfig was also removed as this behavior is now controlled by base Parser class which is extended.

  • The soft deprecated static performSelfAnalysis method has been fully removed, use the instance method with the same name instead.

  • The IParserConfig's ignoredIssues property has been deprecated. Any Parser still using this property will throw an exception on initialization. If any ambiguities need be ignored, the IGNORE_AMBIGUITIES property should be used instead on specific DSL rules.

  • Nested / In-Lined rules via the NAME parameter for DSL rules have been deprecated, e.g:

    this.RULE("statements", () => {
          NAME: "$letStatement",
          ALT: () => {
            // ...
          NAME: "$selectStatement",
          ALT: () => {
            // ...

    This feature was not orthogonal with other features (e.g error recovery) and added quite a-lot of complexity for the small benefit of a little syntactic sugar. Instead of using "nested / in-lined" rules, simply extract the content of these rules to "regular" top level rules.

  • Reducing the usage of 'any' in the 'OR' method type signature may cause existing code to fail TypeScript compilation. In such a case an explicit usage of a generic any type will resolve the problem.

    this.OR<any>(/* ... */);
  • All methods of the Interface errorMessageProvider are now mandatory. To defer to the default error message template behavior, defer to chevrotain.defaultParserErrorProvider, e.g:

    import {
    } from "chevrotain";
    class myCustomErrorMsgProvider implements IParserErrorMessageProvider {
      buildNoViableAltMessage(options: {
        expectedPathsPerAlt: TokenType[][][];
        actual: IToken[];
        previous: IToken;
        customUserDescription: string;
        ruleName: string;
      }): string {
        // Custom user error message builder
        return "sad sad panda:" + options.actual[0].image;
      buildEarlyExitMessage(options: {
        expectedIterationPaths: TokenType[][];
        actual: IToken[];
        previous: IToken;
        customUserDescription: string;
        ruleName: string;
      }): string {
        // invoking the default error message string builder.
        return defaultParserErrorProvider.buildEarlyExitMessage(options);
      // Implementation of other properties from `IParserErrorMessageProvider`
      // ...
  • The TokenType's tokenName property has been deprecated (This actually happened in 6.3.1...) use the name property instead.

  • The GAST Flat class was renamed to Alternative.


  • Due to re-implementation of the grammar analysis via "grammar recording", certain semantics action will now need to be wrapped in the new ACTIONopen in new window Parsing DSL method. This will not affect Parsers that output a CST and only affect some of the Parsers which employ embedded semantic actions. The Missing ACTION wrappers will be automatically detected and throw a descriptive error message to ease migration.

  • Grammar de-serialization support has been removed as it is now redundant as Chevrotain no longer relies on Function.prototype.toString. And the de-serialization feature was a workaround to issues caused by Function.prototype.toString. This means the serializedGrammar propertyopen in new window of the Parser's configuration was removed, and using it will cause an error to be thrown during initialization.

  • The Parser's getGAstProductions methodopen in new window now returns a plain JavaScript object representing a Map/Dictionary rather than the Chevrotain's internal HashTable implementation.


  • Setting the Parser's input before this.performSelfAnalysis is called will now throw an error. This limitation is necessary in order to enable to enable the automatic detection of missing this.performSelfAnalysis calls. To avoid this issue do not pass the input token vector input to the Parser's constructor and instead always set the input after the Parser instance was created, This pattern is demonstrated in the tutorial:


  • The Parser constructor no longer accepts a token vector as an argument. The "input" setter should be used instead, for example:

    // Old API
    class MyOldParser extends Parser {
      constructor(input, config) {
        super(input, allTokens, config);
    const oldInstance = new MyOldParser(
        /* token vector */
    // New API
    class MyNewParser extends Parser {
      constructor(config) {
        super(allTokens, config);
    const newInstance = new MyNewParser({});
    newInstance.input = [
      /* token vector */
    • Note that the input setter has existed for a while and has been used in the official examples and documentation, therefore it is likely that only the constructor need be modified in existing parsers.
  • Automatic Concrete Syntax Treeopen in new window output is now enabled by default. This means that parser which rely on embedded actions must explicitly disable the CST output, for example:

    class MyNewParser extends Parser {
      constructor() {
        // we have to explicitly disable the CST building for embedded actions to work.
        super(allTokens, { outputCst: false });
    • If a parser already uses CST output no change is needed in 4.0
  • DSL repetitions no longer return any values in embedded actions mode:

    • MANY / AT_LEAST_ONE no longer return an array of the iteration results. The iterations results should be collected manually instead:

      // Before 4.0.0
      const stmts = $.MANY(() => {
        return $.SUBRULE(Statement);
      // After 4.0.0
      const stmts = [];
      $.MANY(() => {
    • Similarly MANY_SEP / AT_LEAST_ONE_SEP also no longer return any results. These used to return both the repetition result array and an array of separators Tokens consumed. It is still possible to manually collect the repetition results, but not the separator tokens.

    • This change has no effect when using automatic CST creation.


  • A CST Node's children dictionary no longer contains empty arrays for unmatched terminals and non-terminals. This means that some existence checks conditions in the CST visitor must be refactored, for example:

    class MyVisitor extends SomeBaseVisitor {
      atomicExpression(ctx) {
        // BAD - will fail due to "TypeError: Cannot read property '0' of undefined"
        if (ctx.Integer[0]) {
          return ctx.Integer[0].image;
        // GOOD - safe check
        if (ctx.Integer) {
          // if a property exists it's value is guaranteed to have at least one element.
          return ctx.Identifier[0].image;


  • The creation of TokenTypes using the class keyword syntax has been soft deprecated. and is no longer officially supported. e.g:

    // No longer officially supported
    class Identifier {
      static pattern = /[a-zA-Z_]\w+/;
    // Use the createToken API instead
    const Identifier = createToken({
      name: "Identifier",
      pattern: /[a-zA-Z_]\w+/,

    See the reasoning in this issueopen in new window.