MessagePack for ECMA-262/JavaScript/TypeScript

This library is an implementation of MessagePack for TypeScript and JavaScript, providing a compact and efficient binary serialization format. Learn more about MessagePack at:

https://msgpack.org/

This library serves as a comprehensive reference implementation of MessagePack for JavaScript with a focus on accuracy, compatibility, interoperability, and performance.

Additionally, this is also a universal JavaScript library. It is compatible not only with browsers, but with Node.js or other JavaScript engines that implement ES2015+ standards. As it is written in TypeScript, this library bundles up-to-date type definition files (d.ts).

*Note that this is the second edition of "MessagePack for JavaScript". The first edition, which was implemented in ES5 and never released to npmjs.com, is tagged as classic.

Synopsis

1import { deepStrictEqual } from "assert";
2import { encode, decode } from "@msgpack/msgpack";
3
4const object = {
5  nil: null,
6  integer: 1,
7  float: Math.PI,
8  string: "Hello, world!",
9  binary: Uint8Array.from([1, 2, 3]),
10  array: [10, 20, 30],
11  map: { foo: "bar" },
12  timestampExt: new Date(),
13};
14
15const encoded: Uint8Array = encode(object);
16
17deepStrictEqual(decode(encoded), object);

Table of Contents

• Synopsis
• Table of Contents
• Install
• API
  • encode(data: unknown, options?: EncoderOptions): Uint8Array
    • EncoderOptions
  • decode(buffer: ArrayLike<number> | BufferSource, options?: DecoderOptions): unknown
    • DecoderOptions
  • decodeMulti(buffer: ArrayLike<number> | BufferSource, options?: DecoderOptions): Generator<unknown, void, unknown>
  • decodeAsync(stream: ReadableStreamLike<ArrayLike<number> | BufferSource>, options?: DecoderOptions): Promise<unknown>
  • decodeArrayStream(stream: ReadableStreamLike<ArrayLike<number> | BufferSource>, options?: DecoderOptions): AsyncIterable<unknown>
  • decodeMultiStream(stream: ReadableStreamLike<ArrayLike<number> | BufferSource>, options?: DecoderOptions): AsyncIterable<unknown>
  • Reusing Encoder and Decoder instances
• Extension Types
  • ExtensionCodec context
  • Handling BigInt with ExtensionCodec
  • The temporal module as timestamp extensions
• Faster way to decode a large array of floating point numbers
• Decoding a Blob
• MessagePack Specification
  • MessagePack Mapping Table
• Prerequisites
  • ECMA-262
  • NodeJS
  • TypeScript Compiler / Type Definitions
• Benchmark
• Distribution
  • NPM / npmjs.com
  • CDN / unpkg.com
• Deno Support
• Bun Support
• Maintenance
  • Testing
  • Continuous Integration
  • Release Engineering
  • Updating Dependencies
• License

Install

This library is published to npmjs.com as @msgpack/msgpack.

1npm install @msgpack/msgpack

API

encode(data: unknown, options?: EncoderOptions): Uint8Array

It encodes data into a single MessagePack-encoded object, and returns a byte array as Uint8Array. It throws errors if data is, or includes, a non-serializable object such as a function or a symbol.

for example:

1import { encode } from "@msgpack/msgpack";
2
3const encoded: Uint8Array = encode({ foo: "bar" });
4console.log(encoded);

If you'd like to convert an uint8array to a NodeJS Buffer, use Buffer.from(arrayBuffer, offset, length) in order not to copy the underlying ArrayBuffer, while Buffer.from(uint8array) copies it:

1import { encode } from "@msgpack/msgpack";
2
3const encoded: Uint8Array = encode({ foo: "bar" });
4
5// `buffer` refers the same ArrayBuffer as `encoded`.
6const buffer: Buffer = Buffer.from(encoded.buffer, encoded.byteOffset, encoded.byteLength);
7console.log(buffer);

EncoderOptions

Name	Type	Default
extensionCodec	ExtensionCodec	ExtensionCodec.defaultCodec
context	user-defined	-
useBigInt64	boolean	false
maxDepth	number	100
initialBufferSize	number	2048
sortKeys	boolean	false
forceFloat32	boolean	false
forceIntegerToFloat	boolean	false
ignoreUndefined	boolean	false

decode(buffer: ArrayLike<number> | BufferSource, options?: DecoderOptions): unknown

It decodes buffer that includes a MessagePack-encoded object, and returns the decoded object typed unknown.

buffer must be an array of bytes, which is typically Uint8Array or ArrayBuffer. BufferSource is defined as ArrayBuffer | ArrayBufferView.

The buffer must include a single encoded object. If the buffer includes extra bytes after an object or the buffer is empty, it throws RangeError. To decode buffer that includes multiple encoded objects, use decodeMulti() or decodeMultiStream() (recommended) instead.

for example:

1import { decode } from "@msgpack/msgpack";
2
3const encoded: Uint8Array;
4const object = decode(encoded);
5console.log(object);

NodeJS Buffer is also acceptable because it is a subclass of Uint8Array.

DecoderOptions

Name	Type	Default
extensionCodec	ExtensionCodec	ExtensionCodec.defaultCodec
context	user-defined	-
useBigInt64	boolean	false
rawStrings	boolean	false
maxStrLength	number	4_294_967_295 (UINT32_MAX)
maxBinLength	number	4_294_967_295 (UINT32_MAX)
maxArrayLength	number	4_294_967_295 (UINT32_MAX)
maxMapLength	number	4_294_967_295 (UINT32_MAX)
maxExtLength	number	4_294_967_295 (UINT32_MAX)
mapKeyConverter	MapKeyConverterType	throw exception if key is not string or number

MapKeyConverterType is defined as (key: unknown) => string | number.

To skip UTF-8 decoding of strings, rawStrings can be set to true. In this case, strings are decoded into Uint8Array.

You can use max${Type}Length to limit the length of each type decoded.

decodeMulti(buffer: ArrayLike<number> | BufferSource, options?: DecoderOptions): Generator<unknown, void, unknown>

It decodes buffer that includes multiple MessagePack-encoded objects, and returns decoded objects as a generator. See also decodeMultiStream(), which is an asynchronous variant of this function.

This function is not recommended to decode a MessagePack binary via I/O stream including sockets because it's synchronous. Instead, decodeMultiStream() decodes a binary stream asynchronously, typically spending less CPU and memory.

for example:

1import { decode } from "@msgpack/msgpack";
2
3const encoded: Uint8Array;
4
5for (const object of decodeMulti(encoded)) {
6  console.log(object);
7}

decodeAsync(stream: ReadableStreamLike<ArrayLike<number> | BufferSource>, options?: DecoderOptions): Promise<unknown>

It decodes stream, where ReadableStreamLike<T> is defined as ReadableStream<T> | AsyncIterable<T>, in an async iterable of byte arrays, and returns decoded object as unknown type, wrapped in Promise.

This function works asynchronously, and might CPU resources more efficiently compared with synchronous decode(), because it doesn't wait for the completion of downloading.

This function is designed to work with whatwg fetch() like this:

1import { decodeAsync } from "@msgpack/msgpack";
2
3const MSGPACK_TYPE = "application/x-msgpack";
4
5const response = await fetch(url);
6const contentType = response.headers.get("Content-Type");
7if (contentType && contentType.startsWith(MSGPACK_TYPE) && response.body != null) {
8  const object = await decodeAsync(response.body);
9  // do something with object
10} else { /* handle errors */ }

decodeArrayStream(stream: ReadableStreamLike<ArrayLike<number> | BufferSource>, options?: DecoderOptions): AsyncIterable<unknown>

It is alike to decodeAsync(), but only accepts a stream that includes an array of items, and emits a decoded item one by one.

for example:

1import { decodeArrayStream } from "@msgpack/msgpack";
2
3const stream: AsyncIterator<Uint8Array>;
4
5// in an async function:
6for await (const item of decodeArrayStream(stream)) {
7  console.log(item);
8}

decodeMultiStream(stream: ReadableStreamLike<ArrayLike<number> | BufferSource>, options?: DecoderOptions): AsyncIterable<unknown>

It is alike to decodeAsync() and decodeArrayStream(), but the input stream must consist of multiple MessagePack-encoded items. This is an asynchronous variant for decodeMulti().

In other words, it could decode an unlimited stream and emits a decoded item one by one.

for example:

1import { decodeMultiStream } from "@msgpack/msgpack";
2
3const stream: AsyncIterator<Uint8Array>;
4
5// in an async function:
6for await (const item of decodeMultiStream(stream)) {
7  console.log(item);
8}

This function is available since v2.4.0; previously it was called as decodeStream().

Reusing Encoder and Decoder instances

Encoder and Decoder classes are provided to have better performance by reusing instances:

1import { deepStrictEqual } from "assert";
2import { Encoder, Decoder } from "@msgpack/msgpack";
3
4const encoder = new Encoder();
5const decoder = new Decoder();
6
7const encoded: Uint8Array = encoder.encode(object);
8deepStrictEqual(decoder.decode(encoded), object);

According to our benchmark, reusing Encoder instance is about 20% faster than encode() function, and reusing Decoder instance is about 2% faster than decode() function. Note that the result should vary in environments and data structure.

Encoder and Decoder take the same options as encode() and decode() respectively.

Extension Types

To handle MessagePack Extension Types, this library provides ExtensionCodec class.

This is an example to setup custom extension types that handles Map and Set classes in TypeScript:

1import { encode, decode, ExtensionCodec } from "@msgpack/msgpack";
2
3const extensionCodec = new ExtensionCodec();
4
5// Set<T>
6const SET_EXT_TYPE = 0 // Any in 0-127
7extensionCodec.register({
8  type: SET_EXT_TYPE,
9  encode: (object: unknown): Uint8Array | null => {
10    if (object instanceof Set) {
11      return encode([...object], { extensionCodec });
12    } else {
13      return null;
14    }
15  },
16  decode: (data: Uint8Array) => {
17    const array = decode(data, { extensionCodec }) as Array<unknown>;
18    return new Set(array);
19  },
20});
21
22// Map<K, V>
23const MAP_EXT_TYPE = 1; // Any in 0-127
24extensionCodec.register({
25  type: MAP_EXT_TYPE,
26  encode: (object: unknown): Uint8Array => {
27    if (object instanceof Map) {
28      return encode([...object], { extensionCodec });
29    } else {
30      return null;
31    }
32  },
33  decode: (data: Uint8Array) => {
34    const array = decode(data, { extensionCodec }) as Array<[unknown, unknown]>;
35    return new Map(array);
36  },
37});
38
39const encoded = encode([new Set<any>(), new Map<any, any>()], { extensionCodec });
40const decoded = decode(encoded, { extensionCodec });

Ensure you include your extensionCodec in any recursive encode and decode statements!

Note that extension types for custom objects must be [0, 127], while [-1, -128] is reserved for MessagePack itself.

ExtensionCodec context

When you use an extension codec, it might be necessary to have encoding/decoding state to keep track of which objects got encoded/re-created. To do this, pass a context to the EncoderOptions and DecoderOptions:

1import { encode, decode, ExtensionCodec } from "@msgpack/msgpack";
2
3class MyContext {
4  track(object: any) { /*...*/ }
5}
6
7class MyType { /* ... */ }
8
9const extensionCodec = new ExtensionCodec<MyContext>();
10
11// MyType
12const MYTYPE_EXT_TYPE = 0 // Any in 0-127
13extensionCodec.register({
14  type: MYTYPE_EXT_TYPE,
15  encode: (object, context) => {
16    if (object instanceof MyType) {
17      context.track(object);
18      return encode(object.toJSON(), { extensionCodec, context });
19    } else {
20      return null;
21    }
22  },
23  decode: (data, extType, context) => {
24    const decoded = decode(data, { extensionCodec, context });
25    const my = new MyType(decoded);
26    context.track(my);
27    return my;
28  },
29});
30
31// and later
32import { encode, decode } from "@msgpack/msgpack";
33
34const context = new MyContext();
35
36const encoded = encode({ myType: new MyType<any>() }, { extensionCodec, context });
37const decoded = decode(encoded, { extensionCodec, context });

Handling BigInt with ExtensionCodec

This library does not handle BigInt by default, but you have two options to handle it:

• Set useBigInt64: true to map bigint to MessagePack's int64/uint64
• Define a custom ExtensionCodec to map bigint to a MessagePack's extension type

useBigInt64: true is the simplest way to handle bigint, but it has limitations:

• A bigint is encoded in 8 byte binaries even if it's a small integer
• A bigint must be smaller than the max value of the uint64 and larger than the min value of the int64. Otherwise the behavior is undefined.

So you might want to define a custom codec to handle bigint like this:

1import { deepStrictEqual } from "assert";
2import { encode, decode, ExtensionCodec, DecodeError } from "@msgpack/msgpack";
3
4// to define a custom codec:
5const BIGINT_EXT_TYPE = 0; // Any in 0-127
6const extensionCodec = new ExtensionCodec();
7extensionCodec.register({
8  type: BIGINT_EXT_TYPE,
9  encode(input: unknown): Uint8Array | null {
10    if (typeof input === "bigint") {
11      if (input <= Number.MAX_SAFE_INTEGER && input >= Number.MIN_SAFE_INTEGER) {
12        return encode(Number(input));
13      } else {
14        return encode(String(input));
15      }
16    } else {
17      return null;
18    }
19  },
20  decode(data: Uint8Array): bigint {
21    const val = decode(data);
22    if (!(typeof val === "string" || typeof val === "number")) {
23      throw new DecodeError(`unexpected BigInt source: ${val} (${typeof val})`);
24    }
25    return BigInt(val);
26  },
27});
28
29// to use it:
30const value = BigInt(Number.MAX_SAFE_INTEGER) + BigInt(1);
31const encoded = encode(value, { extensionCodec });
32deepStrictEqual(decode(encoded, { extensionCodec }), value);

The temporal module as timestamp extensions

There is a proposal for a new date/time representations in JavaScript:

• https://github.com/tc39/proposal-temporal

This library maps Date to the MessagePack timestamp extension by default, but you can re-map the temporal module (or Temporal Polyfill) to the timestamp extension like this:

1import { Instant } from "@std-proposal/temporal";
2import { deepStrictEqual } from "assert";
3import {
4  encode,
5  decode,
6  ExtensionCodec,
7  EXT_TIMESTAMP,
8  encodeTimeSpecToTimestamp,
9  decodeTimestampToTimeSpec,
10} from "@msgpack/msgpack";
11
12// to define a custom codec
13const extensionCodec = new ExtensionCodec();
14extensionCodec.register({
15  type: EXT_TIMESTAMP, // override the default behavior!
16  encode(input: unknown): Uint8Array | null {
17    if (input instanceof Instant) {
18      const sec = input.seconds;
19      const nsec = Number(input.nanoseconds - BigInt(sec) * BigInt(1e9));
20      return encodeTimeSpecToTimestamp({ sec, nsec });
21    } else {
22      return null;
23    }
24  },
25  decode(data: Uint8Array): Instant {
26    const timeSpec = decodeTimestampToTimeSpec(data);
27    const sec = BigInt(timeSpec.sec);
28    const nsec = BigInt(timeSpec.nsec);
29    return Instant.fromEpochNanoseconds(sec * BigInt(1e9) + nsec);
30  },
31});
32
33// to use it
34const instant = Instant.fromEpochMilliseconds(Date.now());
35const encoded = encode(instant, { extensionCodec });
36const decoded = decode(encoded, { extensionCodec });
37deepStrictEqual(decoded, instant);

This will become default in this library with major-version increment, if the temporal module is standardized.

Faster way to decode a large array of floating point numbers

If there are large arrays of floating point numbers in your payload, there is a way to decode it faster: define a custom extension type for Float#Array with alignment.

An extension type's encode method can return a function that takes a parameter pos: number. This parameter can be used to make alignment of the buffer, resulting decoding it much more performant.

See an example implementation for Float32Array:

1const extensionCodec = new ExtensionCodec();
2
3const EXT_TYPE_FLOAT32ARRAY = 0; // Any in 0-127
4extensionCodec.register({
5  type: EXT_TYPE_FLOAT32ARRAY,
6  encode: (object: unknown) => {
7    if (object instanceof Float32Array) {
8      return (pos: number) => {
9        const bpe = Float32Array.BYTES_PER_ELEMENT;
10        const padding = 1 + ((bpe - ((pos + 1) % bpe)) % bpe);
11        const data = new Uint8Array(object.buffer);
12        const result = new Uint8Array(padding + data.length);
13        result[0] = padding;
14        result.set(data, padding);
15        return result;
16      };
17    }
18    return null;
19  },
20  decode: (data: Uint8Array) => {
21    const padding = data[0]!;
22    const bpe = Float32Array.BYTES_PER_ELEMENT;
23    const offset = data.byteOffset + padding;
24    const length = data.byteLength - padding;
25    return new Float32Array(data.buffer, offset, length / bpe);
26  },
27});

Decoding a Blob

Blob is a binary data container provided by browsers. To read its contents when it contains a MessagePack binary, you can use Blob#arrayBuffer() or Blob#stream(). Blob#stream() is recommended if your target platform support it. This is because streaming decode should be faster for large objects. In both ways, you need to use asynchronous API.

1async function decodeFromBlob(blob: Blob): unknown {
2  if (blob.stream) {
3    // Blob#stream(): ReadableStream<Uint8Array> (recommended)
4    return await decodeAsync(blob.stream());
5  } else {
6    // Blob#arrayBuffer(): Promise<ArrayBuffer> (if stream() is not available)
7    return decode(await blob.arrayBuffer());
8  }
9}

MessagePack Specification

This library is compatible with the "August 2017" revision of MessagePack specification at the point where timestamp ext was added:

• str/bin separation, added at August 2013
• extension types, added at August 2013
• timestamp ext type, added at August 2017

The living specification is here:

https://github.com/msgpack/msgpack

Note that as of June 2019 there're no official "version" on the MessagePack specification. See https://github.com/msgpack/msgpack/issues/195 for the discussions.

MessagePack Mapping Table

The following table shows how JavaScript values are mapped to MessagePack formats and vice versa.

The mapping of integers varies on the setting of useBigInt64.

The default, useBigInt64: false is:

Source Value	MessagePack Format	Value Decoded
null, undefined	nil	null (*1)
boolean (true, false)	bool family	boolean (true, false)
number (53-bit int)	int family	number
number (64-bit float)	float family	number
string	str family	string (*2)
ArrayBufferView	bin family	Uint8Array (*3)
Array	array family	Array
Object	map family	Object (*4)
Date	timestamp ext family	Date (*5)
bigint	N/A	N/A (*6)

• *1 Both null and undefined are mapped to nil (0xC0) type, and are decoded into null
• *2 If you'd like to skip UTF-8 decoding of strings, set rawStrings: true. In this case, strings are decoded into Uint8Array.
• *3 Any ArrayBufferViews including NodeJS's Buffer are mapped to bin family, and are decoded into Uint8Array
• *4 In handling Object, it is regarded as Record<string, unknown> in terms of TypeScript
• *5 MessagePack timestamps may have nanoseconds, which will lost when it is decoded into JavaScript Date. This behavior can be overridden by registering -1 for the extension codec.
• *6 bigint is not supported in useBigInt64: false mode, but you can define an extension codec for it.

If you set useBigInt64: true, the following mapping is used:

Source Value	MessagePack Format	Value Decoded
null, undefined	nil	null
boolean (true, false)	bool family	boolean (true, false)
number (32-bit int)	int family	number
number (except for the above)	float family	number
bigint	int64 / uint64	bigint (*7)
string	str family	string
ArrayBufferView	bin family	Uint8Array
Array	array family	Array
Object	map family	Object
Date	timestamp ext family	Date

• *7 If the bigint is larger than the max value of uint64 or smaller than the min value of int64, then the behavior is undefined.

Prerequisites

This is a universal JavaScript library that supports major browsers and NodeJS.

ECMA-262

• ES2015 language features
• ES2024 standard library, including:
  • Typed arrays (ES2015)
  • Async iterations (ES2018)
  • Features added in ES2015-ES2022
• whatwg encodings (TextEncoder and TextDecoder)

ES2022 standard library used in this library can be polyfilled with core-js.

IE11 is no longer supported. If you'd like to use this library in IE11, use v2.x versions.

NodeJS

NodeJS v18 is required.

TypeScript Compiler / Type Definitions

This module requires type definitions of AsyncIterator, ArrayBufferLike, whatwg streams, and so on. They are provided by "lib": ["ES2024", "DOM"] in tsconfig.json.

Regarding the TypeScript compiler version, only the latest TypeScript is tested in development.

Benchmark

Run-time performance is not the only reason to use MessagePack, but it's important to choose MessagePack libraries, so a benchmark suite is provided to monitor the performance of this library.

V8's built-in JSON has been improved for years, esp. JSON.parse() is significantly improved in V8/7.6, it is the fastest deserializer as of 2019, as the benchmark result bellow suggests.

However, MessagePack can handles binary data effectively, actual performance depends on situations. Esp. streaming-decoding may be significantly faster than non-streaming decoding if it's effective. You'd better take benchmark on your own use-case if performance matters.

Benchmark on NodeJS/v22.13.1 (V8/12.4)

operation	op	ms	op/s
buf = Buffer.from(JSON.stringify(obj));	1348700	5000	269740
obj = JSON.parse(buf.toString("utf-8"));	1700300	5000	340060
buf = require("msgpack-lite").encode(obj);	591300	5000	118260
obj = require("msgpack-lite").decode(buf);	539500	5000	107900
buf = require("@msgpack/msgpack").encode(obj);	1238700	5000	247740
obj = require("@msgpack/msgpack").decode(buf);	1402000	5000	280400
buf = /* @msgpack/msgpack */ encoder.encode(obj);	1379800	5000	275960
obj = /* @msgpack/msgpack */ decoder.decode(buf);	1406100	5000	281220

Note that JSON cases use Buffer to emulate I/O where a JavaScript string must be converted into a byte array encoded in UTF-8, whereas MessagePack modules deal with byte arrays.

Distribution

NPM / npmjs.com

The NPM package distributed in npmjs.com includes both ES2015+ and ES5 files:

• dist/ is compiled into ES2020 with CommomJS, provided for NodeJS v10
• dist.umd/ is compiled into ES5 with UMD
  • dist.umd/msgpack.min.js - the minified file
  • dist.umd/msgpack.js - the non-minified file
• dist.esm/ is compiled into ES2020 with ES modules, provided for webpack-like bundlers and NodeJS's ESM-mode

If you use NodeJS and/or webpack, their module resolvers use the suitable one automatically.

CDN / unpkg.com

This library is available via CDN:

1<script crossorigin src="https://unpkg.com/@msgpack/msgpack"></script>

It loads MessagePack module to the global object.

Deno Support

You can use this module on Deno.

See example/deno-*.ts for examples.

deno.land/x is not supported.

Bun Support

You can use this module on Bun.

Maintenance

Testing

For simple testing:

npm run test

Continuous Integration

This library uses GitHub Actions.

Test matrix:

• NodeJS
  • v18 / v20 / v22
• Browsers:
  • Chrome, Firefox
• Deno
• Bun

Release Engineering

1# run tests on NodeJS, Chrome, and Firefox
2make test-all
3
4# edit the changelog
5code CHANGELOG.md
6
7# bump version
8npm version patch|minor|major
9
10# run the publishing task
11make publish

Updating Dependencies

1npm run update-dependencies

License

Copyright 2019 The MessagePack community.

This software uses the ISC license:

https://opensource.org/licenses/ISC

See LICENSE for details.