scure-base

Audited & minimal implementation of bech32, base64, base58, base32 & base16.

• 🔒 Audited by an independent security firm
• 🔻 Tree-shakeable: unused code is excluded from your builds
• 📦 ESM and common.js
• ✍️ Written in functional style, easily composable
• 💼 Matches specs
  • BIP173, BIP350 for bech32 / bech32m
  • RFC 4648 (aka RFC 3548) for Base16, Base32, Base32Hex, Base64, Base64Url
  • Base58, Base58check, Base32 Crockford
• 🪶 Only 8KB

Check out Projects using scure-base.

This library belongs to scure

scure — audited micro-libraries.

• Zero or minimal dependencies
• Highly readable TypeScript / JS code
• PGP-signed releases and transparent NPM builds
• Check out homepage & all libraries: base, bip32, bip39, btc-signer, starknet

Usage

npm install @scure/base

deno add @scure/base

We support all major platforms and runtimes. The library is hybrid ESM / Common.js package.

1import { base16, base32, base64, base58 } from '@scure/base';
2// Flavors
3import {
4  base58xmr,
5  base58xrp,
6  base32nopad,
7  base32hex,
8  base32hexnopad,
9  base32crockford,
10  base64nopad,
11  base64url,
12  base64urlnopad,
13} from '@scure/base';
14
15const data = Uint8Array.from([1, 2, 3]);
16base64.decode(base64.encode(data));
17
18// Convert utf8 string to Uint8Array
19const data2 = new TextEncoder().encode('hello');
20base58.encode(data2);
21
22// Everything has the same API except for bech32 and base58check
23base32.encode(data);
24base16.encode(data);
25base32hex.encode(data);

base58check is a special case: you need to pass sha256() function:

1import { createBase58check } from '@scure/base';
2createBase58check(sha256).encode(data);

Alternative API:

1import { str, bytes } from '@scure/base';
2const encoded = str('base64', data);
3const data = bytes('base64', encoded);

Bech32, Bech32m and Bitcoin

We provide low-level bech32 operations. If you need high-level methods for BTC (addresses, and others), use scure-btc-signer instead.

Bitcoin addresses use both 5-bit words and bytes representations. They can't be parsed using bech32.decodeToBytes. Instead, do something this:

1const decoded = bech32.decode(address);
2// NOTE: words in bitcoin addresses contain version as first element,
3// with actual witness program words in rest
4// BIP-141: The value of the first push is called the "version byte".
5// The following byte vector pushed is called the "witness program".
6const [version, ...dataW] = decoded.words;
7const program = bech32.fromWords(dataW); // actual witness program

Same applies to Lightning Invoice Protocol BOLT-11. We have many tests in ./test/bip173.test.js that serve as minimal examples of Bitcoin address and Lightning Invoice Protocol parsers. Keep in mind that you'll need to verify the examples before using them in your code.

Design rationale

The code may feel unnecessarily complicated; but actually it's much easier to reason about. Any encoding library consists of two functions:

encode(A) -> B
decode(B) -> A
  where X = decode(encode(X))
  # encode(decode(X)) can be !== X!
  # because decoding can normalize input

e.g.
base58checksum = {
  encode(): {
    // checksum
    // radix conversion
    // alphabet
  },
  decode(): {
    // alphabet
    // radix conversion
    // checksum
  }
}

But instead of creating two big functions for each specific case, we create them from tiny composable building blocks:

base58checksum = chain(checksum(), radix(), alphabet())

Which is the same as chain/pipe/sequence function in Functional Programming, but significantly more useful since it enforces same order of execution of encode/decode. Basically you only define encode (in declarative way) and get correct decode for free. So, instead of reasoning about two big functions you need only reason about primitives and encode chain. The design revealed obvious bug in older version of the lib, where xmr version of base58 had errors in decode's block processing.

Besides base-encodings, we can reuse the same approach with any encode/decode function (bytes2number, bytes2u32, etc). For example, you can easily encode entropy to mnemonic (BIP-39):

1export function getCoder(wordlist: string[]) {
2  if (!Array.isArray(wordlist) || wordlist.length !== 2 ** 11 || typeof wordlist[0] !== 'string') {
3    throw new Error('Wordlist: expected array of 2048 strings');
4  }
5  return mbc.chain(mbu.checksum(1, checksum), mbu.radix2(11, true), mbu.alphabet(wordlist));
6}

base58 is O(n^2) and radixes

Uint8Array is represented as big-endian number:

[1, 2, 3, 4, 5] -> 1*(256**4) + 2*(256**3) 3*(256**2) + 4*(256**1) + 5*(256**0)
where 256 = 2**8 (8 bits per byte)

which is then converted to a number in another radix/base (16/32/58/64, etc).

However, generic conversion between bases has quadratic O(n^2) time complexity.

Which means base58 has quadratic time complexity too. Use base58 only when you have small constant sized input, because variable length sized input from user can cause DoS.

On the other hand, if both bases are power of same number (like 2**8 <-> 2**64), there is linear algorithm. For now we have implementation for power-of-two bases only (radix2).

Security

The library has been independently audited:

• at version 1.0.0, in Jan 2022, by cure53
  • PDFs: online, offline
  • Changes since audit.
  • The audit has been funded by Ethereum Foundation with help of Nomic Labs

The library was initially developed for js-ethereum-cryptography. At commit ae00e6d7, it was extracted to a separate package called micro-base. After the audit we've decided to use @scure NPM namespace for security.

Resources

Projects using scure-base

• scure-btc-signer
• prefixed-api-key: A re-write of seamapi/prefixed-api-key that enhances the cryptographic security properties and safety when verifying a key. The keys and verifiers of these two libraries are not compatible. Motivating post on the issues with using JWT from fly.io
• coinspace wallet and its modules: ada, btc eos, sol, xmr

License

MIT (c) Paul Miller (https://paulmillr.com), see LICENSE file.