protobuf.js

Protocol Buffers are a language-neutral, platform-neutral, extensible way of serializing structured data for use in communications protocols, data storage, and more, originally designed at Google (see).

protobuf.js is a pure JavaScript implementation with TypeScript support for Node.js and the browser. It's easy to use, does not sacrifice on performance, has good conformance and works out of the box with .proto files!

Contents

• Installation
  How to include protobuf.js in your project.

• Usage
  A brief introduction to using the toolset.

  • Valid Message
  • Toolset
    

• Examples
  A few examples to get you started.

  • Using .proto files
  • Using JSON descriptors
  • Using reflection only
  • Using custom classes
  • Using services
  • Usage with TypeScript
    

• Additional documentation
  A list of available documentation resources.

• Performance
  A few internals and a benchmark on performance.

• Compatibility
  Notes on compatibility regarding browsers and optional libraries.

• Building
  How to build the library and its components yourself.

Installation

Node.js

1npm install protobufjs --save

1// Static code + Reflection + .proto parser
2var protobuf = require("protobufjs");
3
4// Static code + Reflection
5var protobuf = require("protobufjs/light");
6
7// Static code only
8var protobuf = require("protobufjs/minimal");

The optional command line utility to generate static code and reflection bundles lives in the protobufjs-cli package and can be installed separately:

1npm install protobufjs-cli --save-dev

Browsers

Pick the variant matching your needs and replace the version tag with the exact release your project depends upon. For example, to use the minified full variant:

1<script src="//cdn.jsdelivr.net/npm/protobufjs@7.X.X/dist/protobuf.min.js"></script>

Distribution	Location
Full	https://cdn.jsdelivr.net/npm/protobufjs/dist/
Light	https://cdn.jsdelivr.net/npm/protobufjs/dist/light/
Minimal	https://cdn.jsdelivr.net/npm/protobufjs/dist/minimal/

All variants support CommonJS and AMD loaders and export globally as window.protobuf.

Usage

Because JavaScript is a dynamically typed language, protobuf.js utilizes the concept of a valid message in order to provide the best possible performance (and, as a side product, proper typings):

Valid message

A valid message is an object (1) not missing any required fields and (2) exclusively composed of JS types understood by the wire format writer.

There are two possible types of valid messages and the encoder is able to work with both of these for convenience:

• Message instances (explicit instances of message classes with default values on their prototype) naturally satisfy the requirements of a valid message and
• Plain JavaScript objects that just so happen to be composed in a way satisfying the requirements of a valid message as well.

In a nutshell, the wire format writer understands the following types:

Field type	Expected JS type (create, encode)	Conversion (fromObject)
s-/u-/int32 s-/fixed32	number (32 bit integer)	value | 0 if signed value >>> 0 if unsigned
s-/u-/int64 s-/fixed64	Long-like (optimal) number (53 bit integer)	Long.fromValue(value) with long.js parseInt(value, 10) otherwise
float double	number	Number(value)
bool	boolean	Boolean(value)
string	string	String(value)
bytes	Uint8Array (optimal) Buffer (optimal under node) Array.<number> (8 bit integers)	base64.decode(value) if a string Object with non-zero .length is assumed to be buffer-like
enum	number (32 bit integer)	Looks up the numeric id if a string
message	Valid message	Message.fromObject(value)
repeated T	Array<T>	Copy
map<K, V>	Object<K,V>	Copy

• Explicit undefined and null are considered as not set if the field is optional.
• Maps are objects where the key is the string representation of the respective value or an 8 characters long hash string for Long-likes.

Toolset

With that in mind and again for performance reasons, each message class provides a distinct set of methods with each method doing just one thing. This avoids unnecessary assertions / redundant operations where performance is a concern but also forces a user to perform verification (of plain JavaScript objects that might just so happen to be a valid message) explicitly where necessary - for example when dealing with user input.

Note that Message below refers to any message class.

• Message.verify(message: Object): null|string
  verifies that a plain JavaScript object satisfies the requirements of a valid message and thus can be encoded without issues. Instead of throwing, it returns the error message as a string, if any.

  1var payload = "invalid (not an object)";
  2var err = AwesomeMessage.verify(payload);
  3if (err)
  4  throw Error(err);

• Message.encode(message: Message|Object [, writer: Writer]): Writer
  encodes a message instance or valid plain JavaScript object. This method does not implicitly verify the message and it's up to the user to make sure that the payload is a valid message.

  1var buffer = AwesomeMessage.encode(message).finish();

• Message.encodeDelimited(message: Message|Object [, writer: Writer]): Writer
  works like Message.encode but additionally prepends the length of the message as a varint.

• Message.decode(reader: Reader|Uint8Array): Message
  decodes a buffer to a message instance. If required fields are missing, it throws a util.ProtocolError with an instance property set to the so far decoded message. If the wire format is invalid, it throws an Error.

  1try {
  2  var decodedMessage = AwesomeMessage.decode(buffer);
  3} catch (e) {
  4    if (e instanceof protobuf.util.ProtocolError) {
  5      // e.instance holds the so far decoded message with missing required fields
  6    } else {
  7      // wire format is invalid
  8    }
  9}

• Message.decodeDelimited(reader: Reader|Uint8Array): Message
  works like Message.decode but additionally reads the length of the message prepended as a varint.

• Message.create(properties: Object): Message
  creates a new message instance from a set of properties that satisfy the requirements of a valid message. Where applicable, it is recommended to prefer Message.create over Message.fromObject because it doesn't perform possibly redundant conversion.

  1var message = AwesomeMessage.create({ awesomeField: "AwesomeString" });

• Message.fromObject(object: Object): Message
  converts any non-valid plain JavaScript object to a message instance using the conversion steps outlined within the table above.

  1var message = AwesomeMessage.fromObject({ awesomeField: 42 });
  2// converts awesomeField to a string

• Message.toObject(message: Message [, options: ConversionOptions]): Object
  converts a message instance to an arbitrary plain JavaScript object for interoperability with other libraries or storage. The resulting plain JavaScript object might still satisfy the requirements of a valid message depending on the actual conversion options specified, but most of the time it does not.

  1var object = AwesomeMessage.toObject(message, {
  2  enums: String,  // enums as string names
  3  longs: String,  // longs as strings (requires long.js)
  4  bytes: String,  // bytes as base64 encoded strings
  5  defaults: true, // includes default values
  6  arrays: true,   // populates empty arrays (repeated fields) even if defaults=false
  7  objects: true,  // populates empty objects (map fields) even if defaults=false
  8  oneofs: true    // includes virtual oneof fields set to the present field's name
  9});

For reference, the following diagram aims to display relationships between the different methods and the concept of a valid message:

In other words: verify indicates that calling create or encode directly on the plain object will [result in a valid message respectively] succeed. fromObject, on the other hand, does conversion from a broader range of plain objects to create valid messages. (ref)

Examples

Using .proto files

It is possible to load existing .proto files using the full library, which parses and compiles the definitions to ready to use (reflection-based) message classes:

1// awesome.proto
2package awesomepackage;
3syntax = "proto3";
4
5message AwesomeMessage {
6    string awesome_field = 1; // becomes awesomeField
7}

1protobuf.load("awesome.proto", function(err, root) {
2    if (err)
3        throw err;
4
5    // Obtain a message type
6    var AwesomeMessage = root.lookupType("awesomepackage.AwesomeMessage");
7
8    // Exemplary payload
9    var payload = { awesomeField: "AwesomeString" };
10
11    // Verify the payload if necessary (i.e. when possibly incomplete or invalid)
12    var errMsg = AwesomeMessage.verify(payload);
13    if (errMsg)
14        throw Error(errMsg);
15
16    // Create a new message
17    var message = AwesomeMessage.create(payload); // or use .fromObject if conversion is necessary
18
19    // Encode a message to an Uint8Array (browser) or Buffer (node)
20    var buffer = AwesomeMessage.encode(message).finish();
21    // ... do something with buffer
22
23    // Decode an Uint8Array (browser) or Buffer (node) to a message
24    var message = AwesomeMessage.decode(buffer);
25    // ... do something with message
26
27    // If the application uses length-delimited buffers, there is also encodeDelimited and decodeDelimited.
28
29    // Maybe convert the message back to a plain object
30    var object = AwesomeMessage.toObject(message, {
31        longs: String,
32        enums: String,
33        bytes: String,
34        // see ConversionOptions
35    });
36});

Additionally, promise syntax can be used by omitting the callback, if preferred:

1protobuf.load("awesome.proto")
2    .then(function(root) {
3       ...
4    });

Using JSON descriptors

The library utilizes JSON descriptors that are equivalent to a .proto definition. For example, the following is identical to the .proto definition seen above:

1// awesome.json
2{
3  "nested": {
4    "awesomepackage": {
5      "nested": {
6        "AwesomeMessage": {
7          "fields": {
8            "awesomeField": {
9              "type": "string",
10              "id": 1
11            }
12          }
13        }
14      }
15    }
16  }
17}

JSON descriptors closely resemble the internal reflection structure:

Type (T)	Extends	Type-specific properties
ReflectionObject		options
Namespace	ReflectionObject	nested
Root	Namespace	nested
Type	Namespace	fields
Enum	ReflectionObject	values
Field	ReflectionObject	rule, type, id
MapField	Field	keyType
OneOf	ReflectionObject	oneof (array of field names)
Service	Namespace	methods
Method	ReflectionObject	type, requestType, responseType, requestStream, responseStream

• Bold properties are required. Italic types are abstract.
• T.fromJSON(name, json) creates the respective reflection object from a JSON descriptor
• T#toJSON() creates a JSON descriptor from the respective reflection object (its name is used as the key within the parent)

Exclusively using JSON descriptors instead of .proto files enables the use of just the light library (the parser isn't required in this case).

A JSON descriptor can either be loaded the usual way:

1protobuf.load("awesome.json", function(err, root) {
2    if (err) throw err;
3
4    // Continue at "Obtain a message type" above
5});

Or it can be loaded inline:

1var jsonDescriptor = require("./awesome.json"); // exemplary for node
2
3var root = protobuf.Root.fromJSON(jsonDescriptor);
4
5// Continue at "Obtain a message type" above

Using reflection only

Both the full and the light library include full reflection support. One could, for example, define the .proto definitions seen in the examples above using just reflection:

1...
2var Root  = protobuf.Root,
3    Type  = protobuf.Type,
4    Field = protobuf.Field;
5
6var AwesomeMessage = new Type("AwesomeMessage").add(new Field("awesomeField", 1, "string"));
7
8var root = new Root().define("awesomepackage").add(AwesomeMessage);
9
10// Continue at "Create a new message" above
11...

Detailed information on the reflection structure is available within the API documentation.

Using custom classes

Message classes can also be extended with custom functionality and it is also possible to register a custom constructor with a reflected message type:

1...
2
3// Define a custom constructor
4function AwesomeMessage(properties) {
5    // custom initialization code
6    ...
7}
8
9// Register the custom constructor with its reflected type (*)
10root.lookupType("awesomepackage.AwesomeMessage").ctor = AwesomeMessage;
11
12// Define custom functionality
13AwesomeMessage.customStaticMethod = function() { ... };
14AwesomeMessage.prototype.customInstanceMethod = function() { ... };
15
16// Continue at "Create a new message" above

(*) Besides referencing its reflected type through AwesomeMessage.$type and AwesomeMesage#$type, the respective custom class is automatically populated with:

• AwesomeMessage.create
• AwesomeMessage.encode and AwesomeMessage.encodeDelimited
• AwesomeMessage.decode and AwesomeMessage.decodeDelimited
• AwesomeMessage.verify
• AwesomeMessage.fromObject, AwesomeMessage.toObject and AwesomeMessage#toJSON

Afterwards, decoded messages of this type are instanceof AwesomeMessage.

Alternatively, it is also possible to reuse and extend the internal constructor if custom initialization code is not required:

1...
2
3// Reuse the internal constructor
4var AwesomeMessage = root.lookupType("awesomepackage.AwesomeMessage").ctor;
5
6// Define custom functionality
7AwesomeMessage.customStaticMethod = function() { ... };
8AwesomeMessage.prototype.customInstanceMethod = function() { ... };
9
10// Continue at "Create a new message" above

Using services

The library also supports consuming services but it doesn't make any assumptions about the actual transport channel. Instead, a user must provide a suitable RPC implementation, which is an asynchronous function that takes the reflected service method, the binary request and a node-style callback as its parameters:

1function rpcImpl(method, requestData, callback) {
2    // perform the request using an HTTP request or a WebSocket for example
3    var responseData = ...;
4    // and call the callback with the binary response afterwards:
5    callback(null, responseData);
6}

Below is a working example with a typescript implementation using grpc npm package.

1const grpc = require('grpc')
2
3const Client = grpc.makeGenericClientConstructor({})
4const client = new Client(
5  grpcServerUrl,
6  grpc.credentials.createInsecure()
7)
8
9const rpcImpl = function(method, requestData, callback) {
10  client.makeUnaryRequest(
11    method.name,
12    arg => arg,
13    arg => arg,
14    requestData,
15    callback
16  )
17}

Example:

1// greeter.proto
2syntax = "proto3";
3
4service Greeter {
5    rpc SayHello (HelloRequest) returns (HelloReply) {}
6}
7
8message HelloRequest {
9    string name = 1;
10}
11
12message HelloReply {
13    string message = 1;
14}

1...
2var Greeter = root.lookup("Greeter");
3var greeter = Greeter.create(/* see above */ rpcImpl, /* request delimited? */ false, /* response delimited? */ false);
4
5greeter.sayHello({ name: 'you' }, function(err, response) {
6    console.log('Greeting:', response.message);
7});

Services also support promises:

1greeter.sayHello({ name: 'you' })
2    .then(function(response) {
3        console.log('Greeting:', response.message);
4    });

There is also an example for streaming RPC.

Note that the service API is meant for clients. Implementing a server-side endpoint pretty much always requires transport channel (i.e. http, websocket, etc.) specific code with the only common denominator being that it decodes and encodes messages.

Usage with TypeScript

The library ships with its own type definitions and modern editors like Visual Studio Code will automatically detect and use them for code completion.

The npm package depends on @types/node because of Buffer and @types/long because of Long. If you are not building for node and/or not using long.js, it should be safe to exclude them manually.

Using the JS API

The API shown above works pretty much the same with TypeScript. However, because everything is typed, accessing fields on instances of dynamically generated message classes requires either using bracket-notation (i.e. message["awesomeField"]) or explicit casts. Alternatively, it is possible to use a typings file generated for its static counterpart.

1import { load } from "protobufjs"; // respectively "./node_modules/protobufjs"
2
3load("awesome.proto", function(err, root) {
4  if (err)
5    throw err;
6
7  // example code
8  const AwesomeMessage = root.lookupType("awesomepackage.AwesomeMessage");
9
10  let message = AwesomeMessage.create({ awesomeField: "hello" });
11  console.log(`message = ${JSON.stringify(message)}`);
12
13  let buffer = AwesomeMessage.encode(message).finish();
14  console.log(`buffer = ${Array.prototype.toString.call(buffer)}`);
15
16  let decoded = AwesomeMessage.decode(buffer);
17  console.log(`decoded = ${JSON.stringify(decoded)}`);
18});

Using generated static code

If you generated static code to bundle.js using the CLI and its type definitions to bundle.d.ts, then you can just do:

1import { AwesomeMessage } from "./bundle.js";
2
3// example code
4let message = AwesomeMessage.create({ awesomeField: "hello" });
5let buffer  = AwesomeMessage.encode(message).finish();
6let decoded = AwesomeMessage.decode(buffer);

Using decorators

The library also includes an early implementation of decorators.

Note that decorators are an experimental feature in TypeScript and that declaration order is important depending on the JS target. For example, @Field.d(2, AwesomeArrayMessage) requires that AwesomeArrayMessage has been defined earlier when targeting ES5.

1import { Message, Type, Field, OneOf } from "protobufjs/light"; // respectively "./node_modules/protobufjs/light.js"
2
3export class AwesomeSubMessage extends Message<AwesomeSubMessage> {
4
5  @Field.d(1, "string")
6  public awesomeString: string;
7
8}
9
10export enum AwesomeEnum {
11  ONE = 1,
12  TWO = 2
13}
14
15@Type.d("SuperAwesomeMessage")
16export class AwesomeMessage extends Message<AwesomeMessage> {
17
18  @Field.d(1, "string", "optional", "awesome default string")
19  public awesomeField: string;
20
21  @Field.d(2, AwesomeSubMessage)
22  public awesomeSubMessage: AwesomeSubMessage;
23
24  @Field.d(3, AwesomeEnum, "optional", AwesomeEnum.ONE)
25  public awesomeEnum: AwesomeEnum;
26
27  @OneOf.d("awesomeSubMessage", "awesomeEnum")
28  public which: string;
29
30}
31
32// example code
33let message = new AwesomeMessage({ awesomeField: "hello" });
34let buffer  = AwesomeMessage.encode(message).finish();
35let decoded = AwesomeMessage.decode(buffer);

Supported decorators are:

• Type.d(typeName?: string)   (optional)
  annotates a class as a protobuf message type. If typeName is not specified, the constructor's runtime function name is used for the reflected type.

• Field.d<T>(fieldId: number, fieldType: string | Constructor<T>, fieldRule?: "optional" | "required" | "repeated", defaultValue?: T)
  annotates a property as a protobuf field with the specified id and protobuf type.

• MapField.d<T extends { [key: string]: any }>(fieldId: number, fieldKeyType: string, fieldValueType. string | Constructor<{}>)
  annotates a property as a protobuf map field with the specified id, protobuf key and value type.

• OneOf.d<T extends string>(...fieldNames: string[])
  annotates a property as a protobuf oneof covering the specified fields.

Other notes:

• Decorated types reside in protobuf.roots["decorated"] using a flat structure, so no duplicate names.
• Enums are copied to a reflected enum with a generic name on decorator evaluation because referenced enum objects have no runtime name the decorator could use.
• Default values must be specified as arguments to the decorator instead of using a property initializer for proper prototype behavior.
• Property names on decorated classes must not be renamed on compile time (i.e. by a minifier) because decorators just receive the original field name as a string.

ProTip! Not as pretty, but you can use decorators in plain JavaScript as well.

Additional documentation

Protocol Buffers

• Google's Developer Guide

protobuf.js

• API Documentation
• CHANGELOG
• Frequently asked questions on our wiki

Community

• Questions and answers on StackOverflow

Performance

The package includes a benchmark that compares protobuf.js performance to native JSON (as far as this is possible) and Google's JS implementation. On an i7-2600K running node 6.9.1 it yields:

benchmarking encoding performance ...

protobuf.js (reflect) x 541,707 ops/sec ±1.13% (87 runs sampled)
protobuf.js (static) x 548,134 ops/sec ±1.38% (89 runs sampled)
JSON (string) x 318,076 ops/sec ±0.63% (93 runs sampled)
JSON (buffer) x 179,165 ops/sec ±2.26% (91 runs sampled)
google-protobuf x 74,406 ops/sec ±0.85% (86 runs sampled)

   protobuf.js (static) was fastest
  protobuf.js (reflect) was 0.9% ops/sec slower (factor 1.0)
          JSON (string) was 41.5% ops/sec slower (factor 1.7)
          JSON (buffer) was 67.6% ops/sec slower (factor 3.1)
        google-protobuf was 86.4% ops/sec slower (factor 7.3)

benchmarking decoding performance ...

protobuf.js (reflect) x 1,383,981 ops/sec ±0.88% (93 runs sampled)
protobuf.js (static) x 1,378,925 ops/sec ±0.81% (93 runs sampled)
JSON (string) x 302,444 ops/sec ±0.81% (93 runs sampled)
JSON (buffer) x 264,882 ops/sec ±0.81% (93 runs sampled)
google-protobuf x 179,180 ops/sec ±0.64% (94 runs sampled)

  protobuf.js (reflect) was fastest
   protobuf.js (static) was 0.3% ops/sec slower (factor 1.0)
          JSON (string) was 78.1% ops/sec slower (factor 4.6)
          JSON (buffer) was 80.8% ops/sec slower (factor 5.2)
        google-protobuf was 87.0% ops/sec slower (factor 7.7)

benchmarking combined performance ...

protobuf.js (reflect) x 275,900 ops/sec ±0.78% (90 runs sampled)
protobuf.js (static) x 290,096 ops/sec ±0.96% (90 runs sampled)
JSON (string) x 129,381 ops/sec ±0.77% (90 runs sampled)
JSON (buffer) x 91,051 ops/sec ±0.94% (90 runs sampled)
google-protobuf x 42,050 ops/sec ±0.85% (91 runs sampled)

   protobuf.js (static) was fastest
  protobuf.js (reflect) was 4.7% ops/sec slower (factor 1.0)
          JSON (string) was 55.3% ops/sec slower (factor 2.2)
          JSON (buffer) was 68.6% ops/sec slower (factor 3.2)
        google-protobuf was 85.5% ops/sec slower (factor 6.9)

These results are achieved by

• generating type-specific encoders, decoders, verifiers and converters at runtime
• configuring the reader/writer interface according to the environment
• using node-specific functionality where beneficial and, of course
• avoiding unnecessary operations through splitting up the toolset.

You can also run the benchmark ...

$> npm run bench

and the profiler yourself (the latter requires a recent version of node):

$> npm run prof <encode|decode|encode-browser|decode-browser> [iterations=10000000]

Note that as of this writing, the benchmark suite performs significantly slower on node 7.2.0 compared to 6.9.1 because moths.

Compatibility

• Works in all modern and not-so-modern browsers except IE8.
• Because the internals of this package do not rely on google/protobuf/descriptor.proto, options are parsed and presented literally.
• If typed arrays are not supported by the environment, plain arrays will be used instead.
• Support for pre-ES5 environments (except IE8) can be achieved by using a polyfill.
• Support for Content Security Policy-restricted environments (like Chrome extensions without unsafe-eval) can be achieved by generating and using static code instead.
• If a proper way to work with 64 bit values (uint64, int64 etc.) is required, just install long.js alongside this library. All 64 bit numbers will then be returned as a Long instance instead of a possibly unsafe JavaScript number (see).
• For descriptor.proto interoperability, see ext/descriptor

Building

To build the library or its components yourself, clone it from GitHub and install the development dependencies:

$> git clone https://github.com/protobufjs/protobuf.js.git
$> cd protobuf.js
$> npm install

Building the respective development and production versions with their respective source maps to dist/:

$> npm run build

Building the documentation to docs/:

$> npm run docs

Building the TypeScript definition to index.d.ts:

$> npm run build:types

Browserify integration

By default, protobuf.js integrates into any browserify build-process without requiring any optional modules. Hence:

• If int64 support is required, explicitly require the long module somewhere in your project as it will be excluded otherwise. This assumes that a global require function is present that protobuf.js can call to obtain the long module.

  If there is no global require function present after bundling, it's also possible to assign the long module programmatically:

  1var Long = ...;
  2
  3protobuf.util.Long = Long;
  4protobuf.configure();

• If you have any special requirements, there is the bundler for reference.

License: BSD 3-Clause License