JS Binary

Encode/decode to a custom binary format, much more compact and faster than JSON/BSON

Install

npm install js-binary

Goal

This module is analogous to JSON.stringify and JSON.parse, but instead of a JSON string, the data is encoded to a custom binary format (using a Buffer instance to store the data). This format was designed to be very compact and give support for more types (like Date and Buffer).

To reduce overhead in the format, it carries no information about types. This implies that you must define the data schema to encode/decode properly. Huge plus: this automatically validates the data against the given schema (input sanitization for free!). This binary format is well suited for very regular data, like API input/output.

Note that, since it's a binary format, it was not meant to be easily viewed/edited by hand.

Usage

var user = {
	name: {
		first: 'Guilherme',
		last: 'Souza'
	},
	pass: new Buffer('042697a30b2dafbdf91bf66bdacdcba8', 'hex'),
	creationDate: new Date('2014-04-11T21:22:32.504Z'),
	active: true,
	achievements: [3, 14, 15, 92, 65, 35]
}

var Type = require('js-binary').Type
var schema = new Type({
	name: {
		first: 'string',
		last: 'string'
	},
	pass: 'Buffer',
	creationDate: 'date',
	active: 'boolean',
	achievements: ['uint'],
	'optionalField?': 'int'
})

var encoded = schema.encode(user)

var decoded = schema.decode(encoded)

Benchmark

A quick example:

{
	"name": "js-binary",
	"published": "2014-12-21T23:42:46.558Z",
	"downloads": 1717
}

(76 bytes, without formatting) versus 096a732d62696e617279e000014a6f3b531e86b5 (20 bytes)

On average, a 2x to 3x space reduction can be observed.

In the benchmark you can find another test with more data. In my machine with node 8.9.0, this is the result:

Encode
        JSON
                Time: 2.47ms
                Size: 190KiB
        js-binary
                Time: 1.453ms (-41%)
                Size: 51KiB (3.7x less)
Decode
        JSON
                Time: 2.791ms
        js-binary
                Time: 0.497ms (-82%)

Available types

Basic types

'uint': unsigned integer (between 0 and 2^53)
'int': signed integer (between -2^53 and 2^53)
'float': a 64-bit floating-point (the JavaScript number type)
'string': a utf-8 string
'Buffer': a Buffer instance
'boolean': a boolean
'regex': a JS RegExp instance
'date': a JS Date instance
'json': any data supported by JSON format. Read bellow for more
'oid': mongodb ObjectId (see bellow)

Compound types

A compound type is an object with (optional) fields. Those fields may be arrays, but with the restriction that every element has the same data schema.

Examples:

Nested fields: {a: {b: 'int', d: {e: 'int'}}}
Optional fields: {a: 'int', 'b?': 'int', 'c?': {d: 'int'}}
Array fields: {a: ['int']}
All together now: {'a?': [{'b?': 'int'}]}

Array type

An array type in which every element has the same data schema.

Examples:

Int array: ['int']
Object array: [{v: 'int', f: 'string'}]

JSON type

As stated before, the js-binary requires the data to have a rather strict schema. But sometimes, part of the data may not fit this reality. In this case, you can fallback to JSON :)

Of course, a JSON field will miss the point about space efficiency and data validation, but will gain in flexibility.

ObjectId type

js-binary gives first-class support for mongodb ObjectId. But since js-binary doesn't (and shouldn't) depend on any peculiar mongodb driver, the rules for this type are:

Encoding: any object o is accepted, as long new Buffer(String(o), 'hex') yields a 12-byte Buffer
Decoding: returns a 24-char hex-encoded string

This should be compatible with most ObjectId implementations on the wild

Spec

The binary format spec is documented in the format.md file

liusujin/js-binary