xrandom

Classes implementing all-purpose, rock-solid random number generators.

Library priorities:

generation of identical bit-accurate numbers regardless of the platform
reproducibility of the random results in the future
high-quality randomness
performance

It has the same API as the standard Random

import 'package:xrandom/xrandom.dart';

final random = Xrandom();

var a = random.nextBool(); 
var b = random.nextDouble();
var c = random.nextInt(n);

var unordered = [1, 2, 3, 4, 5]..shuffle(random);

Creating the object

The library provides classes that differ in the first letter: Xrandom, Qrandom, Drandom.

If you just want a random number:

final random = Xrandom();

quoteOfTheDay = quotes[ random.nextInt(quotes.length) ];

If you want the same numbers each time:

final random = Drandom(); // D is for Dumb Determinism 

test("no surprises ever", () {
    expect(random.nextInt(100), 42);
    expect(random.nextInt(100), 17);
    expect(random.nextInt(100), 96);
});

If you are solving a computational problem:

final random = Qrandom(); // Q is for Quantifiable Quality

feedMonteCarloSimulation(random);

If you prefer a specific algorithm:

final random = Splitmix64(); // see list of algorithms below

feedMonteCarloSimulation(random);

Speed

Generating random numbers with AOT-compiled binary.

Sorted by nextInt fastest to slowest (numbers show execution time)

Class	nextInt	nextDouble	nextBool
Xrandom	627	640	391
Random (dart:math)	895	929	662
Qrandom / Drandom	933	1219	398

Additions to Random

nextFloat

nextFloat() generates a floating-point value in the range 0.0 ≤ x <1.0.

Unlike the nextDouble, nextFloat prefers speed to precision. It's still a double, but it has four billion shades instead of eight quadrillions.

Speed comparison

Sorted by nextDouble fastest to slowest (numbers show execution time)

JS	Class	nextDouble	nextFloat
	Xorshift64	569	353
	Xorshift128p	635	389
✓	Xrandom	640	221
	Splitmix64	658	398
✓	Xorshift128	815	339
	Mulberry32	841	301
✓	Random (dart:math)	929
	Xoshiro256pp	1182	713
✓	Qrandom / Drandom	1219	539

nextRaw

These methods return the raw output of the generator uncompromisingly fast. Depending on the algorithm, the output is a number consisting of either 32 random bits or 64 random bits.

Xrandom combines small integers or splits large ones. The methods work with any of the generators.

JS	Method	Returns	Equivalent of
✓	`nextRaw32()`	32-bit unsigned	`nextInt(pow(2,32))`
✓	`nextRaw53()`	53-bit unsigned	`nextInt(pow(2,53))`
	`nextRaw64()`	64-bit signed	`nextInt(pow(2,64))`

Speed comparison

Sorted by nextInt fastest to slowest
(numbers show execution time)

JS	Class	nextInt	nextRaw32	nextRaw64
✓	Xrandom	627	280	549
✓	Xorshift128	726	341	782
	Xorshift64	748	346	491
	Mulberry32	767	307	709
	Xorshift128p	772	383	529
	Splitmix64	838	398	500
✓	Random (dart:math)	895
✓	XrandomHq	933	537	1186
	Xoshiro256pp	1138	703	1072

Since nextInt's return range is always limited to 32 bits, only comparison to nextRaw32 is "apples-to-apples".

Algorithms

Native	JS	Class	Algorithm	Introduced	Alias
✓	✓	`Xorshift32`	xorshift32	2003	`Xrandom`
✓		`Xorshift64`	xorshift64	2003
✓	✓	`Xorshift128`	xorshift128	2003
✓		`Splitmix64`	splitmix64	2015
✓		`Xorshift128p`	xorshift128+ v2	2015
✓		`Mulberry32`	mulberry32	2017
✓		`Xoshiro256ss`	xoshiro256** 1.0	2018
✓	✓	`Xoshiro128pp`	xoshiro128++ 1.0	2019	`Qrandom`, `Drandom`
✓		`Xoshiro256pp`	xoshiro256++ 1.0	2019

You can use any generator from the library in the same way as in the examples with the Xrandom class.

final random = Mulberry32();

quoteOfTheDay = quotes[ random.nextInt(quotes.length) ];

Compatibility

TL;DR Xrandom, Qrandom, Drandom work on all platforms. Others may not work on JS.

The library is written in pure Dart. Therefore, it works wherever Dart works.

But some of the classes need full support for 64-bit integers. JavaScript* actually only supports 53 bits. If your target platform is JavaScript, then the selection will have to be narrowed down to the options marked with [✓] checkmark in the JS column. Trying to create a incompatible object in JavaScript-transpiled code will lead to UnsupportedError.

If your code compiles to native (like in Flutter apps for Android and iOS*), 64-bit* generators will work best for you. For example, Xorshift64 for speed or Xoshiro256pp for quality.

More benchmarks

nextInt fastest to slowest (numbers show execution time)

JS	Class	nextInt	nextDouble	nextBool
✓	Xrandom	627	640	391
✓	Xorshift128	726	815	394
	Xorshift64	748	569	386
	Mulberry32	767	841	391
	Xorshift128p	772	635	394
	Splitmix64	838	658	392
✓	Random (dart:math)	895	929	662
✓	Qrandom / Drandom	933	1219	398
	Xoshiro256pp	1138	1182	406

All the benchmarks on this page are from AOT-compiled binaries running on AMD A9-9420e with Ubuntu 20.04. Time is measured in milliseconds.

The tables are created using the tabular library.

Consistency

The library has been thoroughly tested to match reference numbers generated by the same algorithms implemented in C++. Not only ints, but also numbers converted to double including all decimal places that the compiler takes into account.

The sources in C are taken directly from scientific publications or the reference implementations by the authors of the algorithms. The Xorshift128+ results are also matched to reference values from JavaScript xorshift library, which tested the 128+ similarly.

Therefore, the sequence generated for example by the Xoshiro128pp.nextRaw32() with the seed (1, 2, 3, 4) is the same as the C99 code will produce with the same seed.

The double values will also be the same as if the upper bits of uint64_t type were converted to double_t in C99 by unsafe pointer casting. No matter how exotic pointer casting sounds for Dart, and even more so for JavaScript. JavaScript doesn't even have any upper bits of uint64_t. But doubles are the same type everywhere, and their random values will be the same.

Testing is done in the GitHub Actions cloud on Windows, Ubuntu, and macOS in VM and Node.js modes.

rtmigo/xrandom_dart