10 releases (5 breaking)

0.5.7 Mar 12, 2024
0.5.6 Mar 12, 2024
0.4.0 Jan 9, 2024
0.3.1 Nov 26, 2023
0.0.2 Sep 28, 2023

#1183 in Development tools

Download history 1607/week @ 2024-01-28 1566/week @ 2024-02-04 1117/week @ 2024-02-11 890/week @ 2024-02-18 1268/week @ 2024-02-25 1691/week @ 2024-03-03 1469/week @ 2024-03-10 1037/week @ 2024-03-17 557/week @ 2024-03-24 959/week @ 2024-03-31 483/week @ 2024-04-07 1443/week @ 2024-04-14 818/week @ 2024-04-21 525/week @ 2024-04-28 831/week @ 2024-05-05 908/week @ 2024-05-12

3,086 downloads per month
Used in 13 crates (4 directly)


16K SLoC

Authoring Parse Rules

This is a short, or not so short, guide to implement parse rules using the Biome parser infrastructure.


The convention is to prefix your parse rule with parse_ and then use the name defined in the grammar file.

For example, parse_for_statement or parse_expression.


Most parse rules take a &mut reference to the parser as their only parameter and return a ParsedSyntax.

fn parse_rule_name(&mut: Parser) -> ParsedSyntax {}

You're free to add additional parameters to your function if needed. There are rare cases where you want to consider returning ConditionalParsedSyntax as explained in conditional syntax

Parsing a single node

Let's assume you want to parse the JS if statement:

JsIfStatement =
 test: JsAnyExpression
 consequent: JsBlockStatement
 else_clause: JsElseClause?

Presence Test

Now, the parsing function must first test if the parser is positioned at an if statement and return Absent if that's not the case.

if !p.at(T![if]) {
 return ParsedSyntax::Absent;

Why return ParsedSyntax::Absent? The function must return ParsedSyntax::Absent if the rule can't predict by the next token(s) if they form the expected node or not. Doing so allows the calling rule to decide if this is an error and perform an error recovery if necessary. The second reason is to ensure that the rule doesn't return a node where all children are missing.

Your rule implementation may want to consider more than just the first child to determine if it can parse at least some of the expected children. For example, the if statement rule could test if the parser is located at an else clause and then create an if statement where all children are missing except the else clause:

if !p.at(T![if]) && !p.at(T![else]){
  return Absent

Your implementation can also call into another parsing rule if the first child is a node and not a token.

let assignment_target = parse_assignment_target(p);

if assignment_target.is_absent() {
  return Absent;

let my_node = assignment_target.precede_or_missing();

But be careful with calling other rules. Your rule mustn't progress the parser - meaning that it can't advance in the parsing process and consume tokens - if it returns Absent.

Parse children

The parse rules will guide you in how to write your implementation and the parser infrastructure provides the following convenience APIs:

  • Optional token 'ident'?: Use p.eat(token). It eats the next token if it matches the passed-in token.
  • Required token 'ident': Usep.expect(token). It eats the next token if it matches the passed-in token. It adds an Expected 'x' but found 'y' instead error and a missing marker if the token isn't present in the source code.
  • Optional node body: JsBlockStatement?: Useparse_block_statement(p).or_missing(p). It parses the block if it is present in the source code and adds a missing marker if it isn't.
  • Required node body: JsBlockStatement: Use parse_block_statement(p).or_missing_with_error(p, error_builder): it parses the block statement if it is present in the source code and adds a missing marker and an error if not.

Using the above-described rules result in the following implementation for the if statement rule.

fn parse_if_statement(p: &mut Parser) -> ParsedSyntax {
 if !p.at(T![if]) {
  return Absent;

 let m = p.start();

 parse_any_expression(p).or_add_diagnostic(p, js_parse_errors::expeced_if_statement);
 parse_block_statement(p).or_add_diagnostic(p, js_parse_errors::expected_block_statement);
// the else block is optional, handle the marker by using `ok`

 Present(m.complete(p, JS_IF_STATEMENT));

Hold on, what are these missing markers? Biome's AST facade uses fixed offsets to retrieve a particular child from a node. For example, the 3rd child of the if statement is the condition. However, the condition would become the second element if the opening parentheses ( isn't present in the source text. That's where missing elements come into play.

Parsing Lists & Error Recovery

Parsing lists is different from parsing single elements with a fixed set of children because it requires looping until the parser reaches a terminal token (or the end of the file).

You may remember that parse_* methods shouldn't progress parsing if they return Absent. Not progressing the parser is problematic inside while loops because it inevitably results in an infinite loop.

That's why you must do error recovery when parsing lists. Luckily, the parser comes with the infrastructure to make error recovery a piece of cake. The general structure for parsing a list is (yes, that's something the parser infrastructure should provide for you):

Let's try to parse an array:

[ 1, 3, 6 ]

We will use ParseSeparatedList in order to achieve that

struct ArrayElementsList;

impl ParseSeparatedList for ArrayElementsList {
    type ParsedElement = CompletedMarker;

    fn parse_element(&mut self, p: &mut Parser) -> ParsedSyntax<Self::ParsedElement> {

    fn is_at_list_end(&self, p: &mut Parser) -> bool {
        p.at_ts(token_set![T![default], T![case], T!['}']])

    fn recover(
        &mut self,
        p: &mut Parser,
        parsed_element: ParsedSyntax<Self::ParsedElement>,
    ) -> parser::RecoveryResult {
            &ParseRecoveryTokenSet::new(JS_BOGUS_STATEMENT, STMT_RECOVERY_SET),

Let's run through this step by step:

    &ParseRecoveryTokenSet::new(JS_BOGUS_STATEMENT, STMT_RECOVERY_SET),

The or_recover performs an error recovery if the parse_array_element method returns Absent; there's no array element in the source text.

The recovery eats all tokens until it finds one of the tokens specified in the token_set, a line break (if you called enable_recovery_on_line_break) or the end of the file.

The recovery doesn't throw the tokens away but instead wraps them inside a JS_BOGUS_EXPRESSION node (first parameter). There exist multiple BOGUS_* nodes. You must consult the grammar to understand which BOGUS* node is supported in your case.

You usually want to include the terminal token ending your list, the element separator token, and the token terminating a statement in your recovery set.

Now, the problem with recovery is that it can fail, and there are two reasons:

  • the parser reached the end of the file;
  • the next token is one of the tokens specified in the recovery set, meaning there is nothing to recover from;

In these cases the ParseSeparatedList and ParseNodeList will recover the parser for you.

Conditional Syntax

The conditional syntax allows you to express that some syntax may not be valid in all source files. Some use cases are:

  • syntax that is only supported in strict or sloppy mode: for example, with statements is not valid when a JavaScript file uses "use strict" or is a module;
  • syntax that is only supported in certain file types: Typescript, JSX, modules;
  • syntax that is only available in specific language versions: experimental features, different versions of the language e.g. (ECMA versions for JavaScript);

The idea is that the parser always parses the syntax regardless of whatever it is supported in this specific file or context. The main motivation behind doing so is that this gives us perfect error recovery and allows us to use the same code regardless of whether the syntax is supported.

However, conditional syntax must be handled because we want to add a diagnostic if the syntax isn't supported for the current file, and the parsed tokens must be attached somewhere.

Let's have a look at the with statement that is only allowed in loose mode/sloppy mode:

fn parse_with_statement(p: &mut Parser) -> ParsedSyntax {
 if !p.at(T![with]) {
  return Absent;

 let m = p.start();
 p.bump(T![with]); // with
 parenthesized_expression(p).or_add_diagnostic(p, js_errors::expected_parenthesized_expression);
 parse_statement(p).or_add_diagnostic(p, js_error::expected_statement);
 let with_stmt = m.complete(p, JS_WITH_STATEMENT);

 let conditional = StrictMode.excluding_syntax(p, with_stmt, |p, marker| {
  p.err_builder("`with` statements are not allowed in strict mode", marker.range(p))


The start of the rule is the same as for any other rule. The exciting bits start with

let conditional = StrictMode.excluding_syntax(p, with_stmt, |p, marker| {
 p.err_builder("`with` statements are not allowed in strict mode", marker.range(p))

The StrictMode.excluding_syntax converts the parsed syntax to a bogus node and uses the diagnostic builder to create a diagnostic if the feature is not supported.

You can convert the ConditionalParsedSyntax to a regular ParsedSyntax by calling or_invalid_to_bogus, which wraps the whole parsed with statement in an BOGUS node if the parser is in strict mode and otherwise returns the unchanged with statement.

What if there's no BOGUS node matching the node of your parse rule? You must then return the ConditionalParsedSyntax without making the or_invalid_to_bogus recovery. It's then up to the caller to recover the potentially invalid syntax.


  • Parse rules are named parse_rule_name
  • The parse rules should return a ParsedSyntax
  • The rule must return Present if it consumes any token and, therefore, can parse the node with at least some of its children.
  • It returns Absent otherwise and must not progress parsing nor add any errors.
  • Lists must perform error recovery to avoid infinite loops.
  • Consult the grammar to identify the BOGUS node that is valid in the context of your rule.


~251K SLoC