RandomInt64

A portable, zero-dependency JavaScript utility for efficiently generating non-deterministic sequences of random 64-bit integers.

Mixes the output of Math.random() (Xorshift128+) with additional entropy derived from performance.now(), which preserves the statistical properties of Xorshift128+ while interrupting its strict determinism. This makes the sequence of generated values much more difficult to predict, and extremely unlikely to ever repeat.

Works in the browser (with some limitations, described below), Node, and Deno (where it requires --allow-hrtime in order to function as intended).

Installation

In Node, install normally using your preferred package manager:

npm install random-int64

pnpm add random-int64

yarn add random-int64

In Deno, you may wish to add the module URL to your import map:

"imports": {
  "random-int64": "https://deno.land/x/random_int64@v0.7.0/mod.ts"
}

Otherwise, you will need to reference the URL directly in your scripts.

Usage

Two classes are available; they only differ in the generated data type:

• RandomInt64 generates BigInt values
• RandomBits64 generates 8-byte arrays (i.e. Uint8Array(8))

Both import (ES modules) and require (CommonJS) are supported. Typescript is supported.

For best performance, the class instance should be cached for repeated usage.

RandomInt64

To generate 64-bit BigInt values,

import { RandomInt64 } from 'random-int64';

const randomInt64 = new RandomInt64();
const int1 = randomInt64.create();
const int2 = randomInt64.create();

console.log(int1.toString());
console.log(int2.toString());

// 7761218963542659225
// 3510194340918650657

Options

The RandomInt64 constructor takes a single argument, a boolean value that indicates whether signed or unsigned values will be generated.

Calling new RandomInt64() with no arguments (or an argument that resolves to false) will return a class instance that generates unsigned 64-bit values between 0 and 2^64 - 1 (as in the above example).

signed

Passing in a boolean true to the constructor will result instead in a class instance that generates signed (randomly positive or negative) values between -(2^63 - 1) and 2^63 - 1.

const randomInt64 = new RandomInt64(true);
const n = randomInt64.create();

console.log(n.toString());
// -3984732910473829021

RandomBits64

To generate 64-bit Uint8Array(8) values,

import { RandomBits64 } from 'random-int64';

const randomBits64 = new RandomBits64();
const bits1 = randomBits64.create();
console.log(bits1.toString());

const bits2 = randomBits64.create();
console.log(bits2.toString());

// 76,152,199,167,4,159,105,169
// 132,202,147,183,84,9,24,190

Options

The RandomBits64 constructor also takes a single (optional) boolean argument, which in this case indicates whether or not the internal buffer will be copied for each generated value.

Calling new RandomBits64() with no arguments (or an argument that resolves to false) will return a class instance that simply exposes a reference to the internal buffer. Thus, in the above example, bits1 and bits2 actually refer to the same Uint8Array instance. (If the console.log() statements were to be grouped together at the end, they would print the same value.)

Reusing the buffer is around 2-5 times faster than copying it, but this is really only useful if you are post-processing the Uint8Array in some way (such as converting it to a string, as above).

copy

Passing in a boolean true to the constructor will create an instance that copies the internal buffer object with each created value.

This is useful if you need unique, persistent Uint8Array objects.

const randomBits64 = new RandomBits64(true);
const bits1 = randomBits64.create();
const bits2 = randomBits64.create();

console.log(bits1.toString());
console.log(bits2.toString());

// 120,204,48,235,228,142,71,137
// 106,95,14,187,47,229,188,112

Compatibility

RandomInt64 generates BigInt values, which limits the supported runtimes to:

• Node 10.4+ (tested on current LTS versions 16+)
• Deno 1+ (tested on 1.30+)
• Chrome 67+
• Edge 79+
• Firefox 68+
• Opera 54+
• Safari 14+

Note that most browsers limit the resolution of performance.now() to a few microseconds (or sometimes more: Firefox only provides 1 ms resolution), which limits the amount of entropy that can be collected on each call. In this case, RandomInt64 will still generate usable sequences of random values, however, outputs will exhibit higher degrees of determinism (comparable to just calling Math.random()).

For this reason, RandomInt64 is somewhat more suited for server-side usage. If you do use it in the browser, be sure to include the Cross-Origin-Opener-Policy: same-origin and Cross-Origin-Embedder-Policy: require-corp headers, so that your application is running in an isolated context (which enables somewhat higher resolution timers).

Disclaimer

This utility was designed for generating random 64-bit object identifiers (database keys) with good performance charateristics, uniform distribution, and a very low likelihood of ever generating a repeated sequence of values.

Although the sequences of generated values are effectively non-deterministic (they depend on the exact timing of successive calls to create()), they cannot be considered cryptographically secure. It may be possible to guess the seed state of the underlying PRNG, or the approximate timing between calls, and thus narrow down the range of values that are likely to be or have been generated.

In other words, RandomInt64 is not suitable for generating cryptographic key material.