ByteUtilities/src/dorkbox/bytes/Base58.kt

229 lines
8.5 KiB
Kotlin

/*
* Copyright 2023 dorkbox, llc
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* Copyright 2011 Google Inc.
* Copyright 2018 Andreas Schildbach
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
* Converted to kotlin by ligi
* https://github.com/komputing/KBase58/blob/master/kbase58/src/main/kotlin/org/komputing/kbase58/Base58.kt
*/
@file:Suppress("unused")
package dorkbox.bytes
import dorkbox.bytes.Base58.CHECKSUM_SIZE
import dorkbox.bytes.Base58.ENCODED_ZERO
import dorkbox.bytes.Base58.divmod
/**
* Base58 is a way to encode Bitcoin addresses (or arbitrary data) as alphanumeric strings.
* <p>
* Note that this is not the same base58 as used by Flickr, which you may find referenced around the Internet.
* <p>
* You may want to consider working with {@link PrefixedChecksummedBytes} instead, which
* adds support for testing the prefix and suffix bytes commonly found in addresses.
* <p>
* Satoshi explains: why base-58 instead of standard base-64 encoding?
* <ul>
* <li>Don't want 0OIl characters that look the same in some fonts and
* could be used to create visually identical looking account numbers.</li>
* <li>A string with non-alphanumeric characters is not as easily accepted as an account number.</li>
* <li>E-mail usually won't line-break if there's no punctuation to break at.</li>
* <li>Doubleclicking selects the whole number as one word if it's all alphanumeric.</li>
* </ul>
* <p>
* However, note that the encoding/decoding runs in O(n^2;) time, so it is not useful for large data.
* <p>
* The basic idea of the encoding is to treat the data bytes as a large number represented using
* base-256 digits, convert the number to be represented using base-58 digits, preserve the exact
* number of leading zeros (which are otherwise lost during the mathematical operations on the
* numbers), and finally represent the resulting base-58 digits as alphanumeric ASCII characters.
*/
object Base58 {
/**
* Gets the version number.
*/
const val version = BytesInfo.version
internal const val ENCODED_ZERO = '1'
internal const val CHECKSUM_SIZE = 4
internal const val alphabet = "123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnopqrstuvwxyz"
internal val alphabetIndices by lazy {
IntArray(128) { alphabet.indexOf(it.toChar()) }
}
/**
* Divides a number, represented as an array of bytes each containing a single digit
* in the specified base, by the given divisor. The given number is modified in-place
* to contain the quotient, and the return value is the remainder.
*
* @param number the number to divide
* @param firstDigit the index within the array of the first non-zero digit
* (this is used for optimization by skipping the leading zeros)
* @param base the base in which the number's digits are represented (up to 256)
* @param divisor the number to divide by (up to 256)
* @return the remainder of the division operation
*/
internal fun divmod(number: ByteArray, firstDigit: UInt, base: UInt, divisor: UInt): UInt {
// this is just long division which accounts for the base of the input digits
var remainder = 0.toUInt()
for (i in firstDigit until number.size.toUInt()) {
val digit = number[i.toInt()].toUByte()
val temp = remainder * base + digit
number[i.toInt()] = (temp / divisor).toByte()
remainder = temp % divisor
}
return remainder
}
}
/**
* Encodes the bytes as a base58 string (no checksum is appended).
*
* @return the base58-encoded string
*/
fun ByteArray.encodeToBase58String(): String {
val input = copyOf(size) // since we modify it in-place
if (input.isEmpty()) {
return ""
}
// Count leading zeros.
var zeros = 0
while (zeros < input.size && input[zeros].toInt() == 0) {
++zeros
}
// Convert base-256 digits to base-58 digits (plus conversion to ASCII characters)
val encoded = CharArray(input.size * 2) // upper bound
var outputStart = encoded.size
var inputStart = zeros
while (inputStart < input.size) {
encoded[--outputStart] = Base58.alphabet[Base58.divmod(input, inputStart.toUInt(), 256.toUInt(), 58.toUInt()).toInt()]
if (input[inputStart].toInt() == 0) {
++inputStart // optimization - skip leading zeros
}
}
// Preserve exactly as many leading encoded zeros in output as there were leading zeros in data.
while (outputStart < encoded.size && encoded[outputStart] == ENCODED_ZERO) {
++outputStart
}
while (--zeros >= 0) {
encoded[--outputStart] = Base58.ENCODED_ZERO
}
// Return encoded string (including encoded leading zeros).
return String(encoded, outputStart, encoded.size - outputStart)
}
/**
* Decodes the base58 string into a [ByteArray]
*
* @return the decoded data bytes
* @throws NumberFormatException if the string is not a valid base58 string
*/
@Throws(NumberFormatException::class)
fun String.decodeBase58(): ByteArray {
if (isEmpty()) {
return ByteArray(0)
}
// Convert the base58-encoded ASCII chars to a base58 byte sequence (base58 digits).
val input58 = ByteArray(length)
for (i in indices) {
val c = this[i]
val digit = if (c.code < 128) Base58.alphabetIndices[c.code] else -1
if (digit < 0) {
throw NumberFormatException("Illegal character $c at position $i")
}
input58[i] = digit.toByte()
}
// Count leading zeros.
var zeros = 0
while (zeros < input58.size && input58[zeros].toInt() == 0) {
++zeros
}
// Convert base-58 digits to base-256 digits.
val decoded = ByteArray(length)
var outputStart = decoded.size
var inputStart = zeros
while (inputStart < input58.size) {
decoded[--outputStart] = divmod(input58, inputStart.toUInt(), 58.toUInt(), 256.toUInt()).toByte()
if (input58[inputStart].toInt() == 0) {
++inputStart // optimization - skip leading zeros
}
}
// Ignore extra leading zeroes that were added during the calculation.
while (outputStart < decoded.size && decoded[outputStart].toInt() == 0) {
++outputStart
}
// Return decoded data (including original number of leading zeros).
return decoded.copyOfRange(outputStart - zeros, decoded.size)
}
/**
* Encodes the given bytes as a base58 string, a checksum is appended
*
* @return the base58-encoded string
*/
fun ByteArray.encodeToBase58WithChecksum() = ByteArray(size + Base58.CHECKSUM_SIZE).apply {
System.arraycopy(this@encodeToBase58WithChecksum, 0, this, 0, this@encodeToBase58WithChecksum.size)
val checksum = this@encodeToBase58WithChecksum.sha256().sha256()
System.arraycopy(checksum, 0, this, this@encodeToBase58WithChecksum.size, Base58.CHECKSUM_SIZE)
}.encodeToBase58String()
fun String.decodeBase58WithChecksum(): ByteArray {
val rawBytes = decodeBase58()
if (rawBytes.size < Base58.CHECKSUM_SIZE) {
throw Exception("Too short for checksum: $this l: ${rawBytes.size}")
}
val checksum = rawBytes.copyOfRange(rawBytes.size - Base58.CHECKSUM_SIZE, rawBytes.size)
val payload = rawBytes.copyOfRange(0, rawBytes.size - Base58.CHECKSUM_SIZE)
val hash = payload.sha256().sha256()
val computedChecksum = hash.copyOfRange(0, CHECKSUM_SIZE)
if (checksum.contentEquals(computedChecksum)) {
return payload
} else {
throw IllegalArgumentException("Checksum mismatch: $checksum is not computed checksum $computedChecksum")
}
}