abstract-syntax-tree

A library for working with abstract syntax trees.

Table of Contents

• Background
• Install
• Usage
• API
• REPL
• Nodes
• Optimizations
• Maintainers
• Contributing
• License

Background

An abstract syntax tree is a way to represent the source code. In case of this library it is represented in the estree format.

For example, the following source code:

const answer = 42

Has the following representation:

{
  "type": "Program",
  "body": [
    {
      "type": "VariableDeclaration",
      "declarations": [
        {
          "type": "VariableDeclarator",
          "id": {
            "type": "Identifier",
            "name": "answer"
          },
          "init": {
            "type": "Literal",
            "value": 42
          }
        }
      ],
      "kind": "const"
    }
  ]
}

The goal of this library is to consolidate common abstract syntax tree operations in one place. It uses a variety of libraries under the hood based on their performance and flexibility, e.g. meriyah for parsing and astring for source code generation.

The library exposes a set of utility methods that can be useful for analysis or transformation of abstract syntax trees. It supports functional and object-oriented programming style.

Install

npm install abstract-syntax-tree

Usage

const { parse, find } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(find(tree, "Literal")) // [ { type: 'Literal', value: 42 } ]

const AbstractSyntaxTree = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = new AbstractSyntaxTree(source)
console.log(tree.find("Literal")) // [ { type: 'Literal', value: 42 } ]

API

Static Methods

parse

The library uses meriyah to create an estree compatible abstract syntax tree. All meriyah parsing options can be passed to the parse method.

const { parse } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(tree) // { type: 'Program', body: [ ... ] }

const { parse } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source, {
  loc: true,
  ranges: true,
})
console.log(tree) // { type: 'Program', body: [ ... ], loc: {...} }

generate

The library uses astring to generate the source code. All astring generate options can be passed to the generate method.

const { parse, generate } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(generate(tree)) // 'const answer = 42;'

walk

Walk method is a thin layer over estraverse.

const { parse, walk } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
walk(tree, (node, parent) => {
  console.log(node)
  console.log(parent)
})

find

Find supports two traversal methods. You can pass a string selector or pass an object that will be compared to every node in the tree. The method returns an array of nodes.

The following selectors are supported:

• node type (Identifier)
• node attribute ([name="foo"])
• node attribute existence ([name])
• wildcard (*)

const { parse, find } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(find(tree, "VariableDeclaration")) // [ { type: 'VariableDeclaration', ... } ]
console.log(find(tree, { type: "VariableDeclaration" })) // [ { type: 'VariableDeclaration', ... } ]

serialize

Serialize can transform nodes into values. Works for: Array, Boolean, Error, Infinity, Map, NaN, Number, Object, RegExp, Set, String, Symbol, WeakMap, WeakSet, null and undefined.

const { serialize } = require("abstract-syntax-tree")
const node = {
  type: "ArrayExpression",
  elements: [
    { type: "Literal", value: 1 },
    { type: "Literal", value: 2 },
    { type: "Literal", value: 3 },
    { type: "Literal", value: 4 },
    { type: "Literal", value: 5 },
  ],
}
const array = serialize(node) // [1, 2, 3, 4, 5]

traverse

Traverse method accepts a configuration object with enter and leave callbacks. It allows multiple transformations in one traversal.

const { parse, traverse } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
traverse(tree, {
  enter(node) {},
  leave(node) {},
})

replace

Replace extends estraverse by handling replacement of give node with multiple nodes. It will also remove given node if null is returned.

const { parse, replace } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
replace(tree, (node) => {
  if (node.type === "VariableDeclaration") {
    node.kind = "let"
  }
  return node
})

remove

Remove uses estraverse and ensures that no useless nodes are left in the tree. It accepts a string, object or callback as the matching strategy.

const { parse, remove, generate } = require("abstract-syntax-tree")
const source = '"use strict"; const b = 4;'
const tree = parse(source)
remove(tree, 'Literal[value="use strict"]')

// or
// remove(tree, { type: 'Literal', value: 'use strict' })

// or
// remove(tree, (node) => {
//   if (node.type === 'Literal' && node.value === 'use strict') return null
//   return node
// })

console.log(generate(tree)) // 'const b = 4;'

each

const { parse, each } = require("abstract-syntax-tree")
const source = "const foo = 1; const bar = 2;"
const tree = parse(source)
each(tree, "VariableDeclaration", (node) => {
  console.log(node)
})

first

const { parse, first } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(first(tree, "VariableDeclaration")) // { type: 'VariableDeclaration', ... }

last

const { parse, last } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(last(tree, "VariableDeclaration")) // { type: 'VariableDeclaration', ... }

reduce

const { parse, reduce } = require("abstract-syntax-tree")
const source = "const a = 1, b = 2"
const tree = parse(source)
const value = reduce(
  tree,
  (sum, node) => {
    if (node.type === "Literal") {
      sum += node.value
    }
    return sum
  },
  0
)
console.log(value) // 3

has

const { parse, has } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(has(tree, "VariableDeclaration")) // true
console.log(has(tree, { type: "VariableDeclaration" })) // true

count

const { parse, count } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
console.log(count(tree, "VariableDeclaration")) // 1
console.log(count(tree, { type: "VariableDeclaration" })) // 1

append

Append pushes nodes to the body of the abstract syntax tree. It accepts estree nodes as input.

const { parse, append } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
append(tree, {
  type: "ExpressionStatement",
  expression: {
    type: "CallExpression",
    callee: {
      type: "MemberExpression",
      object: {
        type: "Identifier",
        name: "console",
      },
      property: {
        type: "Identifier",
        name: "log",
      },
      computed: false,
    },
    arguments: [
      {
        type: "Identifier",
        name: "answer",
      },
    ],
  },
})

Strings will be converted into abstract syntax tree under the hood. Please note that this approach might make the code run a bit slower due to an extra interpretation step.

const { parse, append } = require("abstract-syntax-tree")
const source = "const answer = 42"
const tree = parse(source)
append(tree, "console.log(answer)")

prepend

Prepend unshifts nodes to the body of the abstract syntax tree. Accepts estree nodes or strings as input, same as append.

const { parse, prepend } = require("abstract-syntax-tree")
const source = "const a = 1;"
const tree = parse(source)
prepend(tree, {
  type: "ExpressionStatement",
  expression: {
    type: "Literal",
    value: "use strict",
  },
})

equal

const { equal } = require("abstract-syntax-tree")
console.log(
  equal({ type: "Literal", value: 42 }, { type: "Literal", value: 42 })
) // true
console.log(
  equal({ type: "Literal", value: 41 }, { type: "Literal", value: 42 })
) // false

match

const { match } = require("abstract-syntax-tree")
console.log(match({ type: "Literal", value: 42 }, "Literal[value=42]")) // true
console.log(match({ type: "Literal", value: 41 }, "Literal[value=42]")) // false

template

The function converts the input to an equivalent abstract syntax tree representation. The function uses a <%= %> syntax to insert nodes.

You can also use standard javascript types and they're going to be transformed automatically.

const { template } = require("abstract-syntax-tree")
const literal = template(42)
const nodes = template("const foo = <%= bar %>;", {
  bar: { type: "Literal", value: 1 },
})

const { template } = require("abstract-syntax-tree")
const nodes = template("function foo(<%= bar %>) {}", {
  bar: [
    { type: "Identifier", name: "baz" },
    { type: "Identifier", name: "qux" },
  ],
})

const { template } = require("abstract-syntax-tree")
const literal = template(42)
const nodes = template("const foo = <%= bar %>;", {
  bar: 1,
})

program

Creates an abstract syntax tree with a blank program.

const { program } = require("abstract-syntax-tree")
const tree = program() // { type: 'Program', sourceType: 'module', body: [] }

iife

Creates an abstract syntax tree for an immediately invoked function expression.

const { iife } = require("abstract-syntax-tree")
const node = iife() // { type: 'ExpressionStatement', expression: { ... } }

Instance Methods

Almost all of the static methods (excluding parse, generate, template and match) have their instance equivalents. There are few extra instance methods:

mark

const AbstractSyntaxTree = require("abstract-syntax-tree")
const tree = new AbstractSyntaxTree("const a = 1")
tree.mark()
console.log(tree.first("Program").cid) // 1
console.log(tree.first("VariableDeclaration").cid) // 2

wrap

const AbstractSyntaxTree = require("abstract-syntax-tree")
const source = "const a = 1"
const tree = new AbstractSyntaxTree(source)
tree.wrap((body) => {
  return [
    {
      type: "ExpressionStatement",
      expression: {
        type: "CallExpression",
        callee: {
          type: "FunctionExpression",
          params: [],
          body: {
            type: "BlockStatement",
            body,
          },
        },
        arguments: [],
      },
    },
  ]
})

unwrap

const AbstractSyntaxTree = require("abstract-syntax-tree")
const source = "(function () { console.log(1); }())"
const tree = new AbstractSyntaxTree(source)
tree.unwrap()
console.log(tree.source) // console.log(1);

Getters

body

Gives the body of the root node.

source

Gives access to the source code representation of the abstract syntax tree.

const AbstractSyntaxTree = require("abstract-syntax-tree")
const source = 'const foo = "bar";'
const tree = new AbstractSyntaxTree(source)
console.log(tree.source) // const foo = "bar";

map

Gives the source map of the source code.

Setters

body

Sets the body of the root node.

Transformations

toBinaryExpression

const { toBinaryExpression } = require("abstract-syntax-tree")
const expression = {
  type: "ArrayExpression",
  elements: [
    { type: "Literal", value: "foo" },
    { type: "Literal", value: "bar" },
    { type: "Literal", value: "baz" },
  ],
}
console.log(toBinaryExpression(expression)) // { type: 'BinaryExpression', ... }

Nodes

You can also use classes to create nodes.

const { ArrayExpression, Literal } = require("abstract-syntax-tree")
const expression = new ArrayExpression([
  new Literal("foo"),
  new Literal("bar"),
  new Literal("baz"),
])

Here's a list of all available nodes, with examples.

Type	Example
ArrayExpression	const foo = []
ArrayPattern	const [foo, bar] = bar
ArrowFunctionExpression	(() => {})
AssignmentExpression	foo = bar
AssignmentOperator
AssignmentPattern	function foo(bar = baz) {}
AwaitExpression	(async () => { await foo() })()
BigIntLiteral
BinaryExpression	foo + bar
BinaryOperator
BlockStatement	{ console.log(foo) }
BreakStatement	for (foo in bar) break
CallExpression	foo()
CatchClause	try {} catch (error) {}
ChainElement
ChainExpression	foo?.()
Class
ClassBody	class Foo {}
ClassDeclaration	class Foo {}
ClassExpression	(class {})
ConditionalExpression	foo ? bar : baz
ContinueStatement	while(true) { continue }
DebuggerStatement	debugger
Declaration
Directive
DoWhileStatement	do {} while (true) {}
EmptyStatement	;
ExportAllDeclaration	export * from "foo"
ExportDefaultDeclaration	export default foo
ExportNamedDeclaration	export { foo as bar }
ExportSpecifier	export { foo }
Expression
ExpressionStatement	foo
ForInStatement	for (foo in bar) {}
ForOfStatement	for (foo of bar) {}
ForStatement	for (let i = 0; i < 10; i ++) {}
Function
FunctionBody
FunctionDeclaration	function foo () {}
FunctionExpression	(function () {})
Identifier	foo
IfStatement	if (foo) {}
ImportDeclaration	import "foo"
ImportDefaultSpecifier	import foo from "bar"
ImportExpression	import(foo).then(bar)
ImportNamespaceSpecifier	import * as foo from "bar"
ImportSpecifier	import { foo } from "bar"
LabeledStatement	label: foo
Literal	42
LogicalExpression	true && false
LogicalOperator
MemberExpression	foo.bar
MetaProperty	function foo () { new.target }
MethodDefinition	class Foo { bar() {} }
ModuleDeclaration
ModuleSpecifier
NewExpression	new Foo()
Node
ObjectExpression	({})
ObjectPattern	function foo ({}) {}
Pattern
Position
Program	42
Property
RegExpLiteral
RestElement	function foo (...bar) {}
ReturnStatement	function foo () { return bar }
SequenceExpression	foo, bar
SourceLocation
SpreadElement
Statement
Super	class Foo extends Bar { constructor() { super() } }
SwitchCase	switch (foo) { case 'bar': }
SwitchStatement	switch(foo) {}
TaggedTemplateExpression	css.foo { color: red; }
TemplateLiteral	css.foo { color: red; }
ThisExpression	this.foo = 'bar'
ThrowStatement	throw new Error("foo")
TryStatement	try { foo() } catch (exception) { bar() }
UnaryExpression	!foo
UnaryOperator
UpdateExpression	foo++
UpdateOperator
VariableDeclaration	const answer = 42
VariableDeclarator	const foo = 'bar'
WhileStatement	while (true) {}
WithStatement	with (foo) {}
YieldExpression	function* foo() { yield bar }

Optimizations

How can you optimize an abstract syntax tree?

Abstract syntax tree is a tree-like structure that represents your program. The program is interpreted at some point, e.g. in your browser. Everything takes time, and the same applies to the interpretation. Some of the operations, e.g. adding numbers can be done at compile time, so that the interpreter has less work to do. Having less work to do means that your program will run faster.

Usage

const { binaryExpressionReduction } = require("abstract-syntax-tree")

What optimization techniques are available?

binaryExpressionReduction

const number = 2 + 2

In the example above we have added two numbers. We could optimize the code by:

const number = 4

The tree would be translated from:

{
  "type": "BinaryExpression",
  "operator": "+",
  "left": { "type": "Literal", "value": 2 },
  "right": { "type": "Literal", "value": 2 }
}

to

{ "type": "Literal", "value": 4 }

ifStatementRemoval

if (true) {
  console.log("foo")
} else {
  console.log("bar")
}

It seems that we'll only enter the true path. We can simplify the code to:

console.log("foo")

The tree would be translated from:

{
  "type": "IfStatement",
  "test": {
    "type": "Literal",
    "value": true
  },
  "consequent": {
    "type": "BlockStatement",
    "body": [
      {
        "type": "ExpressionStatement",
        "expression": {
          "type": "CallExpression",
          "callee": {
            "type": "MemberExpression",
            "object": {
              "type": "Identifier",
              "name": "console"
            },
            "property": {
              "type": "Identifier",
              "name": "log"
            },
            "computed": false
          },
          "arguments": [
            {
              "type": "Literal",
              "value": "foo"
            }
          ]
        }
      }
    ]
  },
  "alternate": {
    "type": "BlockStatement",
    "body": [
      {
        "type": "ExpressionStatement",
        "expression": {
          "type": "CallExpression",
          "callee": {
            "type": "MemberExpression",
            "object": {
              "type": "Identifier",
              "name": "console"
            },
            "property": {
              "type": "Identifier",
              "name": "log"
            },
            "computed": false
          },
          "arguments": [
            {
              "type": "Literal",
              "value": "bar"
            }
          ]
        }
      }
    ]
  }
}

to:

{
        "type": "CallExpression",
        "callee": {
          "type": "MemberExpression",
          "object": {
            "type": "Identifier",
            "name": "console"
          },
          "property": {
            "type": "Identifier",
            "name": "log"
          },
          "computed": false
        },
        "arguments": [
          {
            "type": "Literal",
            "value": "foo"
          }
        ]
      }

negationOperatorRemoval

if (!(foo === bar)) {
  console.log("foo")
}

It seems that our negation operator could be a part of the condition inside the brackets.

if (foo !== bar) {
  console.log("foo")
}

The tree would be translated from:

{
  "type": "UnaryExpression",
  "operator": "!",
  "prefix": true,
  "argument": {
    "type": "BinaryExpression",
    "left": {
      "type": "Identifier",
      "name": "foo"
    },
    "operator": "===",
    "right": {
      "type": "Identifier",
      "name": "bar"
    }
  }
}

to

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

logicalExpressionReduction

const foo = "bar" || "baz"

The first value is truthy so it's safe to simplify the code.

const foo = "bar"

The tree would be translated from:

{
  "type": "LogicalExpression",
  "left": {
    "type": "Literal",
    "value": "bar"
  },
  "operator": "||",
  "right": {
    "type": "Literal",
    "value": "baz"
  }
}

To:

{
  "type": "Literal",
  "value": "bar"
}

ternaryOperatorReduction

const foo = true ? "bar" : "baz"

Given a known value of the conditional expression it's possible to get the right value immediately.

const foo = "bar"

The tree would be translated from:

{
  "type": "ConditionalExpression",
  "test": {
    "type": "Literal",
    "value": true
  },
  "consequent": {
    "type": "Literal",
    "value": "bar"
  },
  "alternate": {
    "type": "Literal",
    "value": "baz"
  }
}

To:

{
  "type": "Literal",
  "value": "bar"
}

typeofOperatorReduction

const foo = typeof "bar"

It's possible to determine the type of some variables during analysis.

const foo = "string"

The tree would be translated from:

{
  "type": "UnaryExpression",
  "operator": "typeof",
  "prefix": true,
  "argument": {
    "type": "Literal",
    "value": "foo"
  }
}

To:

{
  "type": "Literal",
  "value": "string"
}

memberExpressionReduction

const foo = { bar: "baz" }.bar

Given an inlined object expression it's possible to retrieve the value immediately.

const foo = "baz"

The tree would be translated from:

{
  "type": "MemberExpression",
  "object": {
    "type": "ObjectExpression",
    "properties": [
      {
        "type": "Property",
        "method": false,
        "shorthand": false,
        "computed": false,
        "key": {
          "type": "Identifier",
          "name": "bar"
        },
        "value": {
          "type": "Literal",
          "value": "baz"
        },
        "kind": "init"
      }
    ]
  },
  "property": {
    "type": "Identifier",
    "name": "baz"
  },
  "computed": false
}

To:

{
  "type": "Literal",
  "value": "baz"
}

Browser

The library is not intended to work inside of a browser. This might change in the future, but it's a bigger lift, pretty time consuming. For now, consider exposing and using an API endpoint instead.

Maintainers

@emilos.

Contributing

All contributions are highly appreciated! Open an issue or a submit PR.

The lib follows the tdd approach and is expected to have a high code coverage. Please follow the Contributor Covenant Code of Conduct.

License

MIT © buxlabs