Babel Plugin Handbook

Dieses Dokument beschreibt das Erstellen von Babel Plugins.

Das Handbuch ist in verschiedenen Sprachen verfügbar. Eine vollständige Liste findest du im README.

Einführung
Basics
- ASTs
- Stages of Babel
- Parse
  - Lexical Analysis
  - Syntactic Analysis
- Transform
- Generate
- Traversal
- Visitors
- Paths
  - Paths in Visitors
- State
- Scopes
  - Bindings
API
Writing your first Babel Plugin
Transformation Operations
Plugin Options
Building Nodes
Best Practices

Einführung

Babel ist ein generischer Compiler für JavaScript. Des weiteren ist es eine Sammlung von Modulen, welche für vielfältige Arten von statischer Analyse benutzt werden können.

Statische Analyse ist die Analyse von Code ohne ihn auszuführen. (Analyse von Code mit Ausführung desselben wird als dynamische Analyse bezeichnet). Die statische Analyse kann vielfältige Anwendungen haben. Sie kann für Linting, Kompilieren, Syntax-Highlighting, Codetransformation, Optimierung, Minifikation und vielen anderen Anwendungen verwendet werden.

Babel kann zur erstellung von verschiedenen Arten von Tools verwedet werden um produktiver zu werden und bessere Programme zu schreiben.

For future updates, follow @thejameskyle on Twitter.

Basics

Babel ist ein JavaScript Compiler, genauer ein Source zu Source Compiler, auch "Transpiler" genannt. Das heißt, dass Babel als Eingabe JavsScript Code bekommt, den Code modifiziert und neuen Code als Ausgabe produziert.

ASTs

In jeden dieser Schritte ist das Erstellen oder das Arbeiten mit einem Abstrakten Syntaxbaum oder AST enthalten.

Babel verwendet einen auf ESTree basierten AST mit folgender Spezifikation.

function square(n) {
  return n * n;
}

Um mit AST Knoten zu experimentieren kann das Tool AST Explorer verwendet werden. Dieser Link enthält den AST für den Beispielcode von oben.

Das Programm kann auch als Liste wie folgt abgebildet werden:

- FunctionDeclaration:
  - id:
    - Identifier:
      - name: square
  - params [1]
    - Identifier
      - name: n
  - body:
    - BlockStatement
      - body [1]
        - ReturnStatement
          - argument
            - BinaryExpression
              - operator: *
              - left
                - Identifier
                  - name: n
              - right
                - Identifier
                  - name: n

Oder auch als JavaScript Objekt:

{
  type: "FunctionDeclaration",
  id: {
    type: "Identifier",
    name: "square"
  },
  params: [{
    type: "Identifier",
    name: "n"
  }],
  body: {
    type: "BlockStatement",
    body: [{
      type: "ReturnStatement",
      argument: {
        type: "BinaryExpression",
        operator: "*",
        left: {
          type: "Identifier",
          name: "n"
        },
        right: {
          type: "Identifier",
          name: "n"
        }
      }
    }]
  }
}

Jede Ebene des AST hat eine ähnliche Struktur:

{
  type: "FunctionDeclaration",
  id: {...},
  params: [...],
  body: {...}
}

{
  type: "Identifier",
  name: ...
}

{
  type: "BinaryExpression",
  operator: ...,
  left: {...},
  right: {...}
}

Hinweis: Der Einfachheit halber wurden einige Eigenschaften entfernt.

Jede dieser Ebenen wird auch als Knoten bezeichnet. Ein AST kann aus einem einzelnen Knoten oder hunderten wenn nicht tausenden von Knoten bestehen. Zusammen können sie die Syntax eines Programmes beschreiben und können für die statische Analyse verwendet werden.

Jeder Knoten hat das folgende Interface:

interface Node {
  type: string;
}

Das type Feld ist ein String welcher die Art des Knotens repräsentiert (z.B. "FunctionDeclaration", "Identifier", oder "BinaryExpression"). Jeder Knotentyp definiert zusätzliche Eigenschaften welche den speziellen Knotentypen genauer beschreiben.

Babel generiert zusätzliche Eigenschaften für jeden Knoten, welche die Position des Knotens im originalen Source Code beschreiben.

{
  type: ...,
  start: 0,
  end: 38,
  loc: {
    start: {
      line: 1,
      column: 0
    },
    end: {
      line: 3,
      column: 1
    }
  },
  ...
}

Die Eigenschaften start, end und loc werden zu jedem Knoten hinzugefügt.

Stages of Babel

Die drei grundsätzlichen Phasen von Babel sind Parsen(parse), Transformieren(transform) und Generieren(generate).

Parse

Die parse Phase erhält Source Code als Eingabe und produziert einen AST als Ausgabe. Die Phase wird in zwei Schritte aufgeteilt: Lexikalische Analyse und Syntaktische Analyse.

Lexical Analysis

Lexikalische Analyse nimmt einen Source Code String und wandelt ihn in einen Strom von Tokens um.

Man kann sich die Tokens als eine Liste von syntaktischen Sprachelementen vorstellen.

n * n;

[
  { type: { ... }, value: "n", start: 0, end: 1, loc: { ... } },
  { type: { ... }, value: "*", start: 2, end: 3, loc: { ... } },
  { type: { ... }, value: "n", start: 4, end: 5, loc: { ... } },
  ...
]

Jeder type hat eine Menge von Eigenschaften welche das Token beschreiben:

{
  type: {
    label: 'name',
    keyword: undefined,
    beforeExpr: false,
    startsExpr: true,
    rightAssociative: false,
    isLoop: false,
    isAssign: false,
    prefix: false,
    postfix: false,
    binop: null,
    updateContext: null
  },
  ...
}

Analog zu AST Knoten haben Tokens auch eine start, end und loc Eigenschaft.

Syntactic Analysis

Syntaktische Analyse wandelt einen Strom von Tokens in einen AST um. Diese Phase verwendet die Informationen welche in den Tokens vorhanden ist und wandelt diese in einen AST um, welcher die Struktur des Codes so darstellt welche einfacher bearbeitet werden kann.

Transform

Die Transformations-Phase traversiert einen AST und fügt neue Knoten hinzu, ändert vorhandene Knoten ab oder entfernt Knoten. Dies ist der komplexeste Schritt von Babel oder anderen Compilern. Das ist auch die Phase in welcher Plugins eingesetzt weden und ist somit ein großer Bestandteil dieses Handbuches. Deshalb werden wir diese Phase nicht weiter vertiefen.

Generate

Die Code Generierung Phase wandelt den finalen AST wieder in einen String um und erzeugt auch Source Maps.

Die Code Generierung ist ziemlich einfach: Der AST wird mit einer Tiefensuche traversiert. Dabei wird eine Zeichenkette generiert, welche den transformierten Code darstellt.

Traversal

Um einen AST zu transformieren muss der Baum rekursiv durchlaufen werden.

Angenommen, wir haben den Typ FunctionDeclaration. Er hat folgende Eigenschaften: Id, Params und Körper. Jede diese Eigenschaften hat verschachtelte Knoten.

{
  type: "FunctionDeclaration",
  id: {
    type: "Identifier",
    name: "square"
  },
  params: [{
    type: "Identifier",
    name: "n"
  }],
  body: {
    type: "BlockStatement",
    body: [{
      type: "ReturnStatement",
      argument: {
        type: "BinaryExpression",
        operator: "*",
        left: {
          type: "Identifier",
          name: "n"
        },
        right: {
          type: "Identifier",
          name: "n"
        }
      }
    }]
  }
}

So we start at the FunctionDeclaration and we know its internal properties so we visit each of them and their children in order.

Next we go to id which is an Identifier. Identifiers don't have any child node properties so we move on.

After that is params which is an array of nodes so we visit each of them. In this case it's a single node which is also an Identifier so we move on.

Then we hit body which is a BlockStatement with a property body that is an array of Nodes so we go to each of them.

The only item here is a ReturnStatement node which has an argument, we go to the argument and find a BinaryExpression.

The BinaryExpression has an operator, a left, and a right. The operator isn't a node, just a value, so we don't go to it, and instead just visit left and right.

This traversal process happens throughout the Babel transform stage.

Visitors

When we talk about "going" to a node, we actually mean we are visiting them. The reason we use that term is because there is this concept of a visitor.

Visitors are a pattern used in AST traversal across languages. Simply put they are an object with methods defined for accepting particular node types in a tree. That's a bit abstract so let's look at an example.

const MyVisitor = {
  Identifier() {
    console.log("Called!");
  }
};

Note: Identifier() { ... } is shorthand for Identifier: { enter() { ... } }.

This is a basic visitor that when used during a traversal will call the Identifier() method for every Identifier in the tree.

So with this code the Identifier() method will be called four times with each Identifier (including square).

function square(n) {
  return n * n;
}

Called!
Called!
Called!
Called!

These calls are all on node enter. However there is also the possibility of calling a visitor method when on exit.

Imagine we have this tree structure:

- FunctionDeclaration
  - Identifier (id)
  - Identifier (params[0])
  - BlockStatement (body)
    - ReturnStatement (body)
      - BinaryExpression (argument)
        - Identifier (left)
        - Identifier (right)

As we traverse down each branch of the tree we eventually hit dead ends where we need to traverse back up the tree to get to the next node. Going down the tree we enter each node, then going back up we exit each node.

Let's walk through what this process looks like for the above tree.

Enter FunctionDeclaration
- Enter Identifier (id)
- Hit dead end
- Exit Identifier (id)
- Enter Identifier (params[0])
- Hit dead end
- Exit Identifier (params[0])
- Enter BlockStatement (body)
- Enter ReturnStatement (body)
  - Enter BinaryExpression (argument)
  - Enter Identifier (left)
    - Hit dead end
  - Exit Identifier (left)
  - Enter Identifier (right)
    - Hit dead end
  - Exit Identifier (right)
  - Exit BinaryExpression (argument)
- Exit ReturnStatement (body)
- Exit BlockStatement (body)
Exit FunctionDeclaration

So when creating a visitor you have two opportunities to visit a node.

const MyVisitor = {
  Identifier: {
    enter() {
      console.log("Entered!");
    },
    exit() {
      console.log("Exited!");
    }
  }
};

Paths

An AST generally has many Nodes, but how do Nodes relate to one another? We could have one giant mutable object that you manipulate and have full access to, or we can simplify this with Paths.

A Path is an object representation of the link between two nodes.

For example if we take the following node and its child:

{
  type: "FunctionDeclaration",
  id: {
    type: "Identifier",
    name: "square"
  },
  ...
}

And represent the child Identifier as a path, it looks something like this:

{
  "parent": {
    "type": "FunctionDeclaration",
    "id": {...},
    ....
  },
  "node": {
    "type": "Identifier",
    "name": "square"
  }
}

It also has additional metadata about the path:

{
  "parent": {...},
  "node": {...},
  "hub": {...},
  "contexts": [],
  "data": {},
  "shouldSkip": false,
  "shouldStop": false,
  "removed": false,
  "state": null,
  "opts": null,
  "skipKeys": null,
  "parentPath": null,
  "context": null,
  "container": null,
  "listKey": null,
  "inList": false,
  "parentKey": null,
  "key": null,
  "scope": null,
  "type": null,
  "typeAnnotation": null
}

As well as tons and tons of methods related to adding, updating, moving, and removing nodes, but we'll get into those later.

In a sense, paths are a reactive representation of a node's position in the tree and all sorts of information about the node. Whenever you call a method that modifies the tree, this information is updated. Babel manages all of this for you to make working with nodes easy and as stateless as possible.

Paths in Visitors

When you have a visitor that has a Identifier() method, you're actually visiting the path instead of the node. This way you are mostly working with the reactive representation of a node instead of the node itself.

const MyVisitor = {
  Identifier(path) {
    console.log("Visiting: " + path.node.name);
  }
};

a + b + c;

Visiting: a
Visiting: b
Visiting: c

State

State is the enemy of AST transformation. State will bite you over and over again and your assumptions about state will almost always be proven wrong by some syntax that you didn't consider.

Take the following code:

function square(n) {
  return n * n;
}

Let's write a quick hacky visitor that will rename n to x.

let paramName;

const MyVisitor = {
  FunctionDeclaration(path) {
    const param = path.node.params[0];
    paramName = param.name;
    param.name = "x";
  },

  Identifier(path) {
    if (path.node.name === paramName) {
      path.node.name = "x";
    }
  }
};

This might work for the above code, but we can easily break that by doing this:

function square(n) {
  return n * n;
}
n;

The better way to deal with this is recursion. So let's make like a Christopher Nolan film and put a visitor inside of a visitor.

const updateParamNameVisitor = {
  Identifier(path) {
    if (path.node.name === this.paramName) {
      path.node.name = "x";
    }
  }
};

const MyVisitor = {
  FunctionDeclaration(path) {
    const param = path.node.params[0];
    const paramName = param.name;
    param.name = "x";

    path.traverse(updateParamNameVisitor, { paramName });
  }
};

Of course, this is a contrived example but it demonstrates how to eliminate global state from your visitors.

Scopes

Next let's introduce the concept of a scope. JavaScript has lexical scoping, which is a tree structure where blocks create new scope.

// global scope

function scopeOne() {
  // scope 1

  function scopeTwo() {
    // scope 2
  }
}

Whenever you create a reference in JavaScript, whether that be by a variable, function, class, param, import, label, etc., it belongs to the current scope.

var global = "I am in the global scope";

function scopeOne() {
  var one = "I am in the scope created by `scopeOne()`";

  function scopeTwo() {
    var two = "I am in the scope created by `scopeTwo()`";
  }
}

Code within a deeper scope may use a reference from a higher scope.

function scopeOne() {
  var one = "I am in the scope created by `scopeOne()`";

  function scopeTwo() {
    one = "I am updating the reference in `scopeOne` inside `scopeTwo`";
  }
}

A lower scope might also create a reference of the same name without modifying it.

function scopeOne() {
  var one = "I am in the scope created by `scopeOne()`";

  function scopeTwo() {
    var one = "I am creating a new `one` but leaving reference in `scopeOne()` alone.";
  }
}

When writing a transform, we want to be wary of scope. We need to make sure we don't break existing code while modifying different parts of it.

We may want to add new references and make sure they don't collide with existing ones. Or maybe we just want to find where a variable is referenced. We want to be able to track these references within a given scope.

A scope can be represented as:

{
  path: path,
  block: path.node,
  parentBlock: path.parent,
  parent: parentScope,
  bindings: [...]
}

When you create a new scope you do so by giving it a path and a parent scope. Then during the traversal process it collects all the references ("bindings") within that scope.

Once that's done, there's all sorts of methods you can use on scopes. We'll get into those later though.

Bindings

References all belong to a particular scope; this relationship is known as a binding.

function scopeOnce() {
  var ref = "This is a binding";

  ref; // This is a reference to a binding

  function scopeTwo() {
    ref; // This is a reference to a binding from a lower scope
  }
}

A single binding looks like this:

{
  identifier: node,
  scope: scope,
  path: path,
  kind: 'var',

  referenced: true,
  references: 3,
  referencePaths: [path, path, path],

  constant: false,
  constantViolations: [path]
}

With this information you can find all the references to a binding, see what type of binding it is (parameter, declaration, etc.), lookup what scope it belongs to, or get a copy of its identifier. You can even tell if it's constant and if not, see what paths are causing it to be non-constant.

Being able to tell if a binding is constant is useful for many purposes, the largest of which is minification.

function scopeOne() {
  var ref1 = "This is a constant binding";

  becauseNothingEverChangesTheValueOf(ref1);

  function scopeTwo() {
    var ref2 = "This is *not* a constant binding";
    ref2 = "Because this changes the value";
  }
}

API

Babel is actually a collection of modules. In this section we'll walk through the major ones, explaining what they do and how to use them.

Note: This is not a replacement for detailed API documentation which will be available elsewhere shortly.

`babylon`

Babylon is Babel's parser. Started as a fork of Acorn, it's fast, simple to use, has plugin-based architecture for non-standard features (as well as future standards).

First, let's install it.

$ npm install --save babylon

Let's start by simply parsing a string of code:

import * as babylon from "babylon";

const code = `function square(n) {
  return n * n;
}`;

babylon.parse(code);
// Node {
//   type: "File",
//   start: 0,
//   end: 38,
//   loc: SourceLocation {...},
//   program: Node {...},
//   comments: [],
//   tokens: [...]
// }

We can also pass options to parse() like so:

babylon.parse(code, {
  sourceType: "module", // default: "script"
  plugins: ["jsx"] // default: []
});

sourceType can either be "module" or "script" which is the mode that Babylon should parse in. "module" will parse in strict mode and allow module declarations, "script" will not.

Note: sourceType defaults to "script" and will error when it finds import or export. Pass sourceType: "module" to get rid of these errors.

Since Babylon is built with a plugin-based architecture, there is also a plugins option which will enable the internal plugins. Note that Babylon has not yet opened this API to external plugins, although may do so in the future.

To see a full list of plugins, see the Babylon README.

`babel-traverse`

The Babel Traverse module maintains the overall tree state, and is responsible for replacing, removing, and adding nodes.

Install it by running:

$ npm install --save babel-traverse

We can use it alongside Babylon to traverse and update nodes:

import * as babylon from "babylon";
import traverse from "babel-traverse";

const code = `function square(n) {
  return n * n;
}`;

const ast = babylon.parse(code);

traverse(ast, {
  enter(path) {
    if (
      path.node.type === "Identifier" &&
      path.node.name === "n"
    ) {
      path.node.name = "x";
    }
  }
});

`babel-types`

Babel Types is a Lodash-esque utility library for AST nodes. It contains methods for building, validating, and converting AST nodes. It's useful for cleaning up AST logic with well thought out utility methods.

You can install it by running:

$ npm install --save babel-types

Then start using it:

import traverse from "babel-traverse";
import * as t from "babel-types";

traverse(ast, {
  enter(path) {
    if (t.isIdentifier(path.node, { name: "n" })) {
      path.node.name = "x";
    }
  }
});

Definitions

Babel Types has definitions for every single type of node, with information on what properties belong where, what values are valid, how to build that node, how the node should be traversed, and aliases of the Node.

A single node type definition looks like this:

defineType("BinaryExpression", {
  builder: ["operator", "left", "right"],
  fields: {
    operator: {
      validate: assertValueType("string")
    },
    left: {
      validate: assertNodeType("Expression")
    },
    right: {
      validate: assertNodeType("Expression")
    }
  },
  visitor: ["left", "right"],
  aliases: ["Binary", "Expression"]
});

Builders

You'll notice the above definition for BinaryExpression has a field for a builder.

builder: ["operator", "left", "right"]

This is because each node type gets a builder method, which when used looks like this:

t.binaryExpression("*", t.identifier("a"), t.identifier("b"));

Which creates an AST like this:

{
  type: "BinaryExpression",
  operator: "*",
  left: {
    type: "Identifier",
    name: "a"
  },
  right: {
    type: "Identifier",
    name: "b"
  }
}

Which when printed looks like this:

a * b

Builders will also validate the nodes they are creating and throw descriptive errors if used improperly. Which leads into the next type of method.

Validators

The definition for BinaryExpression also includes information on the fields of a node and how to validate them.

fields: {
  operator: {
    validate: assertValueType("string")
  },
  left: {
    validate: assertNodeType("Expression")
  },
  right: {
    validate: assertNodeType("Expression")
  }
}

This is used to create two types of validating methods. The first of which is isX.

t.isBinaryExpression(maybeBinaryExpressionNode);

This tests to make sure that the node is a binary expression, but you can also pass a second parameter to ensure that the node contains certain properties and values.

t.isBinaryExpression(maybeBinaryExpressionNode, { operator: "*" });

There is also the more, ehem, assertive version of these methods, which will throw errors instead of returning true or false.

t.assertBinaryExpression(maybeBinaryExpressionNode);
t.assertBinaryExpression(maybeBinaryExpressionNode, { operator: "*" });
// Error: Expected type "BinaryExpression" with option { "operator": "*" }

Converters

[WIP]

`babel-generator`

Babel Generator is the code generator for Babel. It takes an AST and turns it into code with sourcemaps.

Run the following to install it:

$ npm install --save babel-generator

Then use it

import * as babylon from "babylon";
import generate from "babel-generator";

const code = `function square(n) {
  return n * n;
}`;

const ast = babylon.parse(code);

generate(ast, null, code);
// {
//   code: "...",
//   map: "..."
// }

You can also pass options to generate().

generate(ast, {
  retainLines: false,
  compact: "auto",
  concise: false,
  quotes: "double",
  // ...
}, code);

`babel-template`

Babel Template is another tiny but incredibly useful module. It allows you to write strings of code with placeholders that you can use instead of manually building up a massive AST.

$ npm install --save babel-template

import template from "babel-template";
import generate from "babel-generator";
import * as t from "babel-types";

const buildRequire = template(`
  var IMPORT_NAME = require(SOURCE);
`);

const ast = buildRequire({
  IMPORT_NAME: t.identifier("myModule"),
  SOURCE: t.stringLiteral("my-module")
});

console.log(generate(ast).code);

var myModule = require("my-module");

Writing your first Babel Plugin

Now that you're familiar with all the basics of Babel, let's tie it together with the plugin API.

Start off with a function that gets passed the current babel object.

export default function(babel) {
  // plugin contents
}

Since you'll be using it so often, you'll likely want to grab just babel.types like so:

export default function({ types: t }) {
  // plugin contents
}

Then you return an object with a property visitor which is the primary visitor for the plugin.

export default function({ types: t }) {
  return {
    visitor: {
      // visitor contents
    }
  };
};

Let's write a quick plugin to show off how it works. Here's our source code:

foo === bar;

Or in AST form:

{
  type: "BinaryExpression",
  operator: "===",
  left: {
    type: "Identifier",
    name: "foo"
  },
  right: {
    type: "Identifier",
    name: "bar"
  }
}

We'll start off by adding a BinaryExpression visitor method.

export default function({ types: t }) {
  return {
    visitor: {
      BinaryExpression(path) {
        // ...
      }
    }
  };
}

Then let's narrow it down to just BinaryExpressions that are using the === operator.

visitor: {
  BinaryExpression(path) {
    if (path.node.operator !== "===") {
      return;
    }

    // ...
  }
}

Now let's replace the left property with a new identifier:

BinaryExpression(path) {
  if (path.node.operator !== "===") {
    return;
  }

  path.node.left = t.identifier("sebmck");
  // ...
}

Already if we run this plugin we would get:

sebmck === bar;

Now let's just replace the right property.

BinaryExpression(path) {
  if (path.node.operator !== "===") {
    return;
  }

  path.node.left = t.identifier("sebmck");
  path.node.right = t.identifier("dork");
}

And now for our final result:

sebmck === dork;

Awesome! Our very first Babel plugin.

Transformation Operations

Visiting

Check if a node is a certain type

If you want to check what the type of a node is, the preferred way to do so is:

BinaryExpression(path) {
  if (t.isIdentifier(path.node.left)) {
    // ...
  }
}

You can also do a shallow check for properties on that node:

BinaryExpression(path) {
  if (t.isIdentifier(path.node.left, { name: "n" })) {
    // ...
  }
}

This is functionally equivalent to:

BinaryExpression(path) {
  if (
    path.node.left != null &&
    path.node.left.type === "Identifier" &&
    path.node.left.name === "n"
  ) {
    // ...
  }
}

Check if an identifier is referenced

Identifier(path) {
  if (path.isReferencedIdentifier()) {
    // ...
  }
}

Alternatively:

Identifier(path) {
  if (t.isReferenced(path.node, path.parent)) {
    // ...
  }
}

Manipulation

Replacing a node

BinaryExpression(path) {
  path.replaceWith(
    t.binaryExpression("**", path.node.left, t.numberLiteral(2))
  );
}

  function square(n) {
-   return n * n;
+   return n ** 2;
  }

Replacing a node with multiple nodes

ReturnStatement(path) {
  path.replaceWithMultiple([
    t.expressionStatement(t.stringLiteral("Is this the real life?")),
    t.expressionStatement(t.stringLiteral("Is this just fantasy?")),
    t.expressionStatement(t.stringLiteral("(Enjoy singing the rest of the song in your head)")),
  ]);
}

  function square(n) {
-   return n * n;
+   "Is this the real life?";
+   "Is this just fantasy?";
+   "(Enjoy singing the rest of the song in your head)";
  }

Note: When replacing an expression with multiple nodes, they must be statements. This is because Babel uses heuristics extensively when replacing nodes which means that you can do some pretty crazy transformations that would be extremely verbose otherwise.

Replacing a node with a source string

FunctionDeclaration(path) {
  path.replaceWithSourceString(`function add(a, b) {
    return a + b;
  }`);
}

- function square(n) {
-   return n * n;
+ function add(a, b) {
+   return a + b;
  }

Note: It's not recommended to use this API unless you're dealing with dynamic source strings, otherwise it's more efficient to parse the code outside of the visitor.

Inserting a sibling node

FunctionDeclaration(path) {
  path.insertBefore(t.expressionStatement(t.stringLiteral("Because I'm easy come, easy go.")));
  path.insertAfter(t.expressionStatement(t.stringLiteral("A little high, little low.")));
}

+ "Because I'm easy come, easy go.";
  function square(n) {
    return n * n;
  }
+ "A little high, little low.";

Note: This should always be a statement or an array of statements. This uses the same heuristics mentioned in Replacing a node with multiple nodes.

Removing a node

FunctionDeclaration(path) {
  path.remove();
}

- function square(n) {
-   return n * n;
- }

Replacing a parent

BinaryExpression(path) {
  path.parentPath.replaceWith(
    t.expressionStatement(t.stringLiteral("Anyway the wind blows, doesn't really matter to me, to me."))
  );
}

  function square(n) {
-   return n * n;
+   "Anyway the wind blows, doesn't really matter to me, to me.";
  }

Removing a parent

BinaryExpression(path) {
  path.parentPath.remove();
}

  function square(n) {
-   return n * n;
  }

Scope

Checking if a local variable is bound

FunctionDeclaration(path) {
  if (path.scope.hasBinding("n")) {
    // ...
  }
}

This will walk up the scope tree and check for that particular binding.

You can also check if a scope has its own binding:

FunctionDeclaration(path) {
  if (path.scope.hasOwnBinding("n")) {
    // ...
  }
}

Generating a UID

This will generate an identifier that doesn't collide with any locally defined variables.

FunctionDeclaration(path) {
  path.scope.generateUidIdentifier("uid");
  // Node { type: "Identifier", name: "_uid" }
  path.scope.generateUidIdentifier("uid");
  // Node { type: "Identifier", name: "_uid2" }
}

Pushing a variable declaration to a parent scope

Sometimes you may want to push a VariableDeclaration so you can assign to it.

FunctionDeclaration(path) {
  const id = path.scope.generateUidIdentifierBasedOnNode(path.node.id);
  path.remove();
  path.scope.parent.push({ id, init: path.node });
}

- function square(n) {
+ var _square = function square(n) {
    return n * n;
- }
+ };

Rename a binding and its references

FunctionDeclaration(path) {
  path.scope.rename("n", "x");
}

- function square(n) {
-   return n * n;
+ function square(x) {
+   return x * x;
  }

Alternatively, you can rename a binding to a generated unique identifier:

FunctionDeclaration(path) {
  path.scope.rename("n");
}

- function square(n) {
-   return n * n;
+ function square(_n) {
+   return _n * _n;
  }

Plugin Options

If you would like to let your users customize the behavior of your Babel plugin you can accept plugin specific options which users can specify like this:

{
  plugins: [
    ["my-plugin", {
      "option1": true,
      "option2": false
    }]
  ]
}

These options then get passed into plugin visitors through the state object:

export default function({ types: t }) {
  return {
    visitor: {
      FunctionDeclaration(path, state) {
        console.log(state.opts);
        // { option1: true, option2: false }
      }
    }
  }
}

These options are plugin-specific and you cannot access options from other plugins.

Building Nodes

When writing transformations you'll often want to build up some nodes to insert into the AST. As mentioned previously, you can do this using the builder methods in the babel-types package.

The method name for a builder is simply the name of the node type you want to build except with the first letter lowercased. For example if you wanted to build a MemberExpression you would use t.memberExpression(...).

The arguments of these builders are decided by the node definition. There's some work that's being done to generate easy-to-read documentation on the definitions, but for now they can all be found here.

A node definition looks like the following:

defineType("MemberExpression", {
  builder: ["object", "property", "computed"],
  visitor: ["object", "property"],
  aliases: ["Expression", "LVal"],
  fields: {
    object: {
      validate: assertNodeType("Expression")
    },
    property: {
      validate(node, key, val) {
        let expectedType = node.computed ? "Expression" : "Identifier";
        assertNodeType(expectedType)(node, key, val);
      }
    },
    computed: {
      default: false
    }
  }
});

Here you can see all the information about this particular node type, including how to build it, traverse it, and validate it.

By looking at the builder property, you can see the 3 arguments that will be needed to call the builder method (t.memberExpression).

builder: ["object", "property", "computed"],

Note that sometimes there are more properties that you can customize on the node than the builder array contains. This is to keep the builder from having too many arguments. In these cases you need to set the properties manually. An example of this is ClassMethod.

You can see the validation for the builder arguments with the fields object.

fields: {
  object: {
    validate: assertNodeType("Expression")
  },
  property: {
    validate(node, key, val) {
      let expectedType = node.computed ? "Expression" : "Identifier";
      assertNodeType(expectedType)(node, key, val);
    }
  },
  computed: {
    default: false
  }
}

You can see that object needs to be an Expression, property either needs to be an Expression or an Identifier depending on if the member expression is computed or not and computed is simply a boolean that defaults to false.

So we can construct a MemberExpression by doing the following:

t.memberExpression(
  t.identifier('object'),
  t.identifier('property')
  // `computed` is optional
);

Which will result in:

object.property

However, we said that object needed to be an Expression so why is Identifier valid?

Well if we look at the definition of Identifier we can see that it has an aliases property which states that it is also an expression.

aliases: ["Expression", "LVal"],

So since MemberExpression is a type of Expression, we could set it as the object of another MemberExpression:

t.memberExpression(
  t.memberExpression(
    t.identifier('member'),
    t.identifier('expression')
  ),
  t.identifier('property')
)

Which will result in:

member.expression.property

It's very unlikely that you will ever memorize the builder method signatures for every node type. So you should take some time and understand how they are generated from the node definitions.

You can find all of the actual definitions here and you can see them documented here

Best Practices

I'll be working on this section over the coming weeks.

Avoid traversing the AST as much as possible

Traversing the AST is expensive, and it's easy to accidentally traverse the AST more than necessary. This could be thousands if not tens of thousands of extra operations.

Babel optimizes this as much as possible, merging visitors together if it can in order to do everything in a single traversal.

Merge visitors whenever possible

When writing visitors, it may be tempting to call path.traverse in multiple places where they are logically necessary.

path.traverse({
  Identifier(path) {
    // ...
  }
});

path.traverse({
  BinaryExpression(path) {
    // ...
  }
});

However, it is far better to write these as a single visitor that only gets run once. Otherwise you are traversing the same tree multiple times for no reason.

path.traverse({
  Identifier(path) {
    // ...
  },
  BinaryExpression(path) {
    // ...
  }
});

Do not traverse when manual lookup will do

It may also be tempting to call path.traverse when looking for a particular node type.

const visitorOne = {
  Identifier(path) {
    // ...
  }
};

const MyVisitor = {
  FunctionDeclaration(path) {
    path.get('params').traverse(visitorOne);
  }
};

However, if you are looking for something specific and shallow, there is a good chance you can manually lookup the nodes you need without performing a costly traversal.

const MyVisitor = {
  FunctionDeclaration(path) {
    path.node.params.forEach(function() {
      // ...
    });
  }
};

Optimizing nested visitors

When you are nesting visitors, it might make sense to write them nested in your code.

const MyVisitor = {
  FunctionDeclaration(path) {
    path.traverse({
      Identifier(path) {
        // ...
      }
    });
  }
};

However, this creates a new visitor object everytime FunctionDeclaration() is called above, which Babel then needs to explode and validate every single time. This can be costly, so it is better to hoist the visitor up.

const visitorOne = {
  Identifier(path) {
    // ...
  }
};

const MyVisitor = {
  FunctionDeclaration(path) {
    path.traverse(visitorOne);
  }
};

If you need some state within the nested visitor, like so:

const MyVisitor = {
  FunctionDeclaration(path) {
    var exampleState = path.node.params[0].name;

    path.traverse({
      Identifier(path) {
        if (path.node.name === exampleState) {
          // ...
        }
      }
    });
  }
};

You can pass it in as state to the traverse() method and have access to it on this in the visitor.

const visitorOne = {
  Identifier(path) {
    if (path.node.name === this.exampleState) {
      // ...
    }
  }
};

const MyVisitor = {
  FunctionDeclaration(path) {
    var exampleState = path.node.params[0].name;
    path.traverse(visitorOne, { exampleState });
  }
};

Being aware of nested structures

Sometimes when thinking about a given transform, you might forget that the given structure can be nested.

For example, imagine we want to lookup the constructor ClassMethod from the Foo ClassDeclaration.

class Foo {
  constructor() {
    // ...
  }
}

const constructorVisitor = {
  ClassMethod(path) {
    if (path.node.name === 'constructor') {
      // ...
    }
  }
}

const MyVisitor = {
  ClassDeclaration(path) {
    if (path.node.id.name === 'Foo') {
      path.traverse(constructorVisitor);
    }
  }
}

We are ignoring the fact that classes can be nested and using the traversal above we will hit a nested constructor as well:

class Foo {
  constructor() {
    class Bar {
      constructor() {
        // ...
      }
    }
  }
}