babel-walk

Lightweight AST traversal tools for Babel ASTs.

Babel supplies the wonderful babel-traverse module for walking Babel ASTs. Problem is, babel-traverse is very heavyweight, as it is designed to supply utilities to make all sorts of AST transformations possible. For simple AST walking without transformation, babel-traverse brings a lot of overhead.

This module loosely implements the API of Acorn parser's walk module, which is a lightweight AST walker for the ESTree AST format.

In my tests, babel-walk's ancestor walker (the most complex walker provided by this module) is about 8 times faster than babel-traverse, if the visitors are cached and the same AST is used for all runs. It is about 16 times faster if a fresh AST is used every run.

Installation

1$ npm install babel-walk

API

1var walk = require('babel-walk');

walk.simple(visitors)(node, state)

Do a simple walk over the AST. node should be the AST node to walk, and visitors an object containing Babel visitors. Each visitor function will be called as (node, state), where node is the AST node, and state is the same state passed to walk.simple.

When walk.simple is called with a fresh set of visitors, it will first "explode" the visitors (e.g. expanding Visitor(node, state) {} to Visitor() { enter(node, state) {} }). This exploding process can take some time, so it is recommended to cache the result of calling walk.simple(visitors) and communicate state leveraging the state parameter.

All babel-types aliases (e.g. Expression) work, but the union syntax (e.g. 'Identifier|AssignmentPattern'(node, state) {}) does not.

walk.ancestor(visitors)(node, state)

Do a simple walk over the AST, but memoizing the ancestors of the node and making them available to the visitors. node should be the AST node to walk, and visitors an object containing Babel visitors. Each visitor function will be called as (node, state, ancestors), where node is the AST node, state is the same state passed to walk.ancestor, and ancestors is an array of ancestors to the node (with the outermost node being [0] and the current node being [ancestors.length - 1]). If state is not specified in the call to walk.ancestor, the state parameter will be set to ancestors.

When walk.ancestor is called with a fresh set of visitors, it will first "explode" the visitors (e.g. expanding Visitor(node, state) {} to Visitor() { enter(node, state) {} }). This exploding process can take some time, so it is recommended to cache the result of calling walk.ancestor(visitors) and communicate state leveraging the state parameter.

All babel-types aliases (e.g. Expression) work, but the union syntax (e.g. 'Identifier|AssignmentPattern'(node, state) {}) does not.

walk.recursive(visitors)(node, state)

Do a recursive walk over the AST, where the visitors are responsible for continuing the walk on the child nodes of their target node. node should be the AST node to walk, and visitors an object containing Babel visitors. Each visitor function will be called as (node, state, c), where node is the AST node, state is the same state passed to walk.recursive, and c is a function that takes a single node as argument and continues walking that node. If no visitor for a node is provided, the default walker algorithm will still be used.

When walk.recursive is called with a fresh set of visitors, it will first "explode" the visitors (e.g. expanding Visitor(node, state) {} to Visitor() { enter(node, state) {} }). This exploding process can take some time, so it is recommended to cache the result of calling walk.recursive(visitors) and communicate state leveraging the state parameter.

Unlike other babel-walk walkers, walk.recursive does not call the exit visitor, only the enter (the default) visitor, of a specific node type.

All babel-types aliases (e.g. Expression) work, but the union syntax (e.g. 'Identifier|AssignmentPattern'(node, state) {}) does not.

In the following example, we are trying to count the number of functions in the outermost scope. This means, that we can simply walk all the statements and increment a counter if it is a function declaration or expression, and then stop walking. Note that we do not specify a visitor for the Program node, and the default algorithm for walking Program nodes is used (which is what we want). Also of note is how I bring the visitors object outside of countFunctions so that the object can be cached to improve performance.

1import * as t from 'babel-types';
2import {parse} from 'babel';
3import * as walk from 'babel-walk';
4
5const visitors = walk.recursive({
6  Statement(node, state, c) {
7    if (t.isVariableDeclaration(node)) {
8      for (let declarator of node.declarations) {
9        // Continue walking the declarator
10        c(declarator);
11      }
12    } else if (t.isFunctionDeclaration(node)) {
13      state.counter++;
14    }
15  },
16
17  VariableDeclarator(node, state) {
18    if (t.isFunction(node.init)) {
19      state.counter++;
20    }
21  },
22});
23
24function countFunctions(node) {
25  const state = {
26    counter: 0,
27  };
28  visitors(node, state);
29  return state.counter;
30}
31
32const ast = parse(`
33  // Counts
34  var a = () => {};
35
36  // Counts
37  function b() {
38    // Doesn't count
39    function c() {
40    }
41  }
42
43  // Counts
44  const c = function d() {};
45`);
46
47countFunctions(ast);
48// = 3

Caveat

For those of you migrating from Acorn to Babel, there are a few things to be aware of.

1. The visitor caching suggestions do not apply to Acorn's walk module, but do for babel-walk.

2. babel-walk does not provide any of the other functions Acorn's walk module provides (e.g. make, findNode*).

3. babel-walk does not use a base variable. The walker algorithm is the same as what babel-traverse uses.

   • That means certain nodes that are not walked by Acorn, such as the property property of a non-computed MemberExpression, are walked by babel-walk.

License

MIT