Recursively walk and transform EcmaScript AST


Recursively walk and add extra information/helpers to Esprima / Mozilla SpiderMonkey Parser API compatible AST.

The main difference between other tools is that it also keeps information about tokens and white spaces and it is meant to be used to transform the tokens and not the string values itself.

This library is specially useful for non-destructive AST manipulation.

This module was heavily inspired by node-falafel and node-burrito but I needed more information than what is currently available on falafel (specially about tokens, empty lines and white spaces) and also needed to do the node traversing on the opposite order (start from leaf nodes). The amount of changes required to introduce the new features and the differences on the concept behind the tool justified a new project.

It was created mainly to be used on esformatter.

Besides all the regular tokens returned by esprima we also add a few more that are important for non-destructive transformations:

  • WhiteSpace
    • Can store multiple white spaces (tabs are considered white space, line breaks not). Important if you want to do non-destructive replacements that are white-space sensitive.
    • Multiple subsequent white spaces are treated as a single token.
  • LineBreak
  • LineComment
  • BlockComment

It's way easier to rebuild the JS string if the tokens already have line breaks and comments. It's also easier to identify if previous/next/current token is a LineBreak or Comment (sometimes needed for non-destructive transformations).

Rocambole structure might change in the future to keep the extraneous tokens outside the tokens array and also add an option to toggle the behavior. (issue #7)

Each Node have the following extra properties/methods:

  • parent : Node|undefined
  • toString() : string
  • next : Node|undefined
  • prev : Node|undefined
  • depth : Number
  • startToken : Token
  • endToken : Token

Each token also have:

  • prev : Token|undefined
  • next : Token|undefined

BlockComment also have:

  • originalIndent: String|undefined

To get a better idea of the generated AST structure try out rocambole-visualize.

You should treat the tokens as a linked list instead of reading the ast.tokens array (inserting/removing items from a linked list is very cheap and won't break the loop). You should grab a reference to the node.startToken and get until you find the desired token or reach the end of the program. To loop between all tokens inside a node you can do like this:

var token = node.startToken;
while (token !== {
    token =;

The method toString loops through all tokens between node.startToken and node.endToken grabbing the token.raw (used by comments) or token.value properties. To implement a method similar to falafel update() you can do this:

function update(nodestr){
    var newToken = {
        type : 'Custom', // can be anything (not used internally) 
        value : str
    // update linked list references 
    if ( node.startToken.prev ) { = newToken;
        newToken.prev = node.startToken.prev;
    if ( ) { = newToken; =;
    node.startToken = node.endToken = newToken;

I plan to create helpers as separate projects when possible.

Parses a string and instrument the AST with extra properties/methods.

var rocambole = require('rocambole');
var ast = rocambole.parse(string);
console.log( ast.startToken );
// to get a string representation of all tokens call toString() 
console.log( ast.toString() );

The moonwalk() starts at the leaf nodes and go down the tree until it reaches the root node (Program). Each node will be traversed only once.

rocambole.moonwalk(ast, function(node){
    if (node.type == 'ArrayExpression'){
        console.log( node.depth +''+ node.toString() );

Traverse order:

 Program [#18]
 `-FunctionDeclaration [#16]
   |-BlockStatement [#14]
   | |-IfStatement [#12]
   | | |-BynaryExpression [#9]
   | | | |-Identifier [#4]
   | | | `-Literal [#5]
   | | `-BlockStatement [#10]
   | |   `-ExpressionStatement [#6]
   | |     `-AssignmentExpression [#3]
   | |       |-Identifier [#1 walk starts here]
   | |       `-Literal [#2]
   | `-VariableDeclaration [#13]
   |   `-VariableDeclarator [#11]
   |     |-Identifier [#7]
   |     `-Literal [#8]
   `-ReturnStatement [#17]
     `-Identifier [#15]

This behavior is very different from node-falafel and node-burrito.

It loops through all nodes on the AST starting from the root node (Program), similar to node-falafel.

rocambole.walk(ast, function(node){

Besides the regular unit tests we also use istanbul to generate code coverage reports, tests should have at least 95% code coverage for statements, branches and lines and 100% code coverage for functions or travis build will fail.

We do not run the coverage test at each call since it slows down the performnace of the tests and it also makes it harder to see the test results. To execute tests and generate coverage report call npm test --coverage, for regular tests just do npm test.

Coverage reports are not committed to the repository since they will change at each npm test --coverage call.


  • make it compatible with esprima@2.1 (don't loop through the same CatchClause twice).
  • updated esprima to v2.0 because of ES6 features and to avoid esprima-fb bug related to RegExp.
  • aliased rocambole.recursive as rocambole.walk to avoid confusions.
  • switched esprima dependency to esprima-fb because of ES6 features.
  • really handle sparse arrays (eg. [,]), fixes moonwalk. (#15)
  • handle sparse arrays (eg. [,]). (#15)
  • only add BlockComment.originalIndent if WhiteSpace is on the start of a line.
  • add toString to empty programs AST (#16)
  • exports BYPASS_RECURSION (#8)
  • fix error if input is empty (#12)
  • support anything that implements toString() as input (#13)
  • fix originalIndent on BlockComment when prev token is not WhiteSpace.
  • add originalIndent to BlockComment (#11)
  • improve rocambole.parse() performance by 4500%. (#4)
  • improve rocambole.moonwalk() performance by 11000%.
  • fix consecutive comments before start of program. (#3)
  • fix loc info on WhiteSpace and LineBreak tokens. (#2)
  • Deprecated:
    • token.before()
    • token.after()
    • token.remove()
    • node.getTokens()
    • ast.nodes
  • avoid recursion over comments.
  • fix weird bug on esformatter introduced on v0.1.1 related to token._ast property.
  • Improve token manipulation methods behavior (before, after, remove)
  • Initial release