Hex and base64 string encoding and decoding for Uint8Array — like native .toHex()/.fromHex() and .toBase64()/.fromBase64(), which are not yet widely supported.
Intended to be as fast as reasonably possible using only plain JavaScript. The secret ingredients are:
- Conversion between strings and typed arrays via
TextEncoder,TextDecoder - Multi-byte reads, writes and lookups using
Uint16Array,Uint32Array - A little bit of loop unrolling (which can make a difference in Chrome/V8)
Comprehensive tests. No external dependencies. ESM and CJS exports, plus TypeScript types. 4KB zipped.
The following benchmarks were run on an M3 Pro MacBook Pro, using 32 MiB of random data, and taking the mean of 10 trials.
On this machine, across methods and browsers, the headlines are that we are:
- 4 – 36x faster than a representative JS implementation: the feross/buffer shim package
- 3 – 7x faster than Firefox's native methods (surprising — Firefox can surely improve on this)
- 5 – 16x slower than Safari's native methods
Chrome Firefox Safari
131.0.6778.86 133.0 Tech Preview 209
* Encode hex
This library 22.59 ms 26.20 ms 44.10 ms
cf. native toHex - 126.20 ms: x5 5.80 ms: /8
cf. feross/buffer.toString 813.48 ms: x36 209.50 ms: x8 368.70 ms: x8
* Decode hex
This library 60.22 ms 32.60 ms 92.40 ms
cf. native fromHex - 229.40 ms: x7 5.80 ms: /16
cf. feross/buffer.from 757.89 ms: x13 546.60 ms: x17 1371.60 ms: x15
* Encode base64
This library 16.86 ms 20.90 ms 42.60 ms
cf. native toBase64 - 84.00 ms: x4 3.60 ms: /12
cf. feross/buffer.toString 275.85 ms: x16 203.80 ms: x10 550.30 ms: x13
* Decode base64
This library 52.02 ms 36.60 ms 23.90 ms
cf. native fromBase64 - 123.00 ms: x3 4.60 ms: /5
cf. feross/buffer.from 200.80 ms: x4 248.10 ms: x7 281.70 ms: x12
To install:
npm install hextremetoHex(bytes: Uint8Array, { alphabet?: 'lower' | 'upper' } = {}): string
Encodes binary data to a hex string.
The alphabet option defaults to 'lower', but may alternatively be set to 'upper'.
Examples:
import { toHex } from 'hextreme';
toHex(new Uint8Array([254, 237, 250, 206]));
// 'feedface'
toHex(new Uint8Array([254, 237, 250, 206]), { alphabet: 'upper' });
// 'FEEDFACE'fromHex(hex: string, { onInvalidInput?: 'throw' | 'truncate' } = {}): Uint8Array
Decodes a hex string (upper-, lower- or mixed-case) to binary data.
The onInvalidInput option defaults to 'throw', in which case any non-hex character in the input causes an error to be thrown. This matches the behaviour of toHex() on a Uint8Array (where supported).
onInvalidInput may otherwise be set to 'truncate', in which case decoding stops at the first non-hex character pair encountered. This matches the behaviour of toString('hex') on a Node Buffer.
Examples:
import { fromHex } from 'hextreme';
fromHex('FEEDface');
// Uint8Array(4) [ 254, 237, 250, 206 ]
fromHex('FEEDfXce');
// Uncaught Error: Invalid pair in hex input at index 4
fromHex('FEEDfXce', { onInvalidInput: 'truncate' });
// Uint8Array(2) [ 254, 237 ]toBase64(bytes: Uint8Array, { alphabet?: 'base64' | 'base64url', omitPadding?: boolean } = {}): string;
Encodes binary data to a base64 string.
The alphabet option defaults to 'base64'. It may alternatively be set to 'base64url', in which case the + and / characters are replaced with - and _.
The omitPadding option defaults to false, so that the output string is padded to a multiple of 4 characters using the = character. It can be set to true to prevent padding being applied.
Examples:
import { toBase64 } from 'hextreme';
const bytes = new Uint8Array([133, 233, 101, 163, 255, 191, 194, 138, 229, 116]);
toBase64(bytes);
// 'hello/+/worldA=='
toBase64(bytes, { alphabet: 'base64url' });
// 'hello_-_worldA=='
toBase64(bytes, { alphabet: 'base64url', omitPadding: true });
// 'hello_-_worldA'fromBase64(base64: string, { alphabet?: 'base64' | 'base64url' | 'base64any', onInvalidInput?: 'throw' | 'skip' } = {}): Uint8Array;
Decodes a base64 string to binary data. Whitespace in the input string (spaces, tabs, \r and \n) is ignored.
The alphabet option defaults to 'base64', but may alternatively be set to 'base64url', in which case - and _ are expected instead of + and /, or 'base64any', in which case both alternatives are recognised.
The onInvalidInput option defaults to 'throw', in which case any non-base64, non-whitespace character in the input causes an error to be thrown. This matches the behaviour of toBase64() on a Uint8Array (where available).
onInvalidInput may otherwise be set to 'skip', in which case any non-base64 characters are skipped and decoding continues (apart from =, which ends decoding). This matches the behaviour of toString('base64') on a Node Buffer.
Note that decoding becomes roughly 2x slower if whitespace or invalid characters are encountered in the input string.
Examples:
import { fromBase64 } from 'hextreme';
fromBase64('hello/+/worldA==');
// Uint8Array(10) [ 133, 233, 101, 163, 255, 191, 194, 138, 229, 116 ]
fromBase64('hello/+/worldA');
// Uint8Array(10) [ 133, 233, 101, 163, 255, 191, 194, 138, 229, 116 ]
fromBase64('hello_-_worldA==', { alphabet: 'base64url' });
// Uint8Array(10) [ 133, 233, 101, 163, 255, 191, 194, 138, 229, 116 ]
fromBase64('hello_+_worldA==', { alphabet: 'base64any' });
// Uint8Array(10) [ 133, 233, 101, 163, 255, 191, 194, 138, 229, 116 ]
fromBase64('hello/:+/worldA==');
// Uncaught Error: Invalid character in base64 at index 6
fromBase64('hello/:+/worldA==', { onInvalidInput: 'skip' });
// Uint8Array(10) [ 133, 233, 101, 163, 255, 191, 194, 138, 229, 116 ]The source is in src. To build: npm run build.
To run tests: npm run test. To run a subset of tests on a big-endian platform (which has some different code paths), see big-endian/README.md.
To run benchmarks: npm run perfCli (for Node and Bun) and npm run perfBrowser.
Copyright (C) 2024 George MacKerron and released under the MIT License.