diff options
author | Ryan Dahl <ry@tinyclouds.org> | 2019-10-09 17:10:09 -0400 |
---|---|---|
committer | Ryan Dahl <ry@tinyclouds.org> | 2019-10-09 17:10:09 -0400 |
commit | 151ce0266eb4de2c8fc600c81c192a5f791b6169 (patch) | |
tree | 7cb04016a1c7ee88adde83814548d7a9409dcde3 /std/encoding/hex.ts | |
parent | a355f7c807686918734416d91b79c26c21effba9 (diff) |
Move everything into std subdir
Diffstat (limited to 'std/encoding/hex.ts')
-rw-r--r-- | std/encoding/hex.ts | 143 |
1 files changed, 143 insertions, 0 deletions
diff --git a/std/encoding/hex.ts b/std/encoding/hex.ts new file mode 100644 index 000000000..de3d76869 --- /dev/null +++ b/std/encoding/hex.ts @@ -0,0 +1,143 @@ +// Ported from Go +// https://github.com/golang/go/blob/go1.12.5/src/encoding/hex/hex.go +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Copyright 2018-2019 the Deno authors. All rights reserved. MIT license. + +const hextable = new TextEncoder().encode("0123456789abcdef"); + +export function errInvalidByte(byte: number): Error { + return new Error( + "encoding/hex: invalid byte: " + + new TextDecoder().decode(new Uint8Array([byte])) + ); +} + +export function errLength(): Error { + return new Error("encoding/hex: odd length hex string"); +} + +// fromHexChar converts a hex character into its value and a success flag. +function fromHexChar(byte: number): [number, boolean] { + switch (true) { + case 48 <= byte && byte <= 57: // '0' <= byte && byte <= '9' + return [byte - 48, true]; + case 97 <= byte && byte <= 102: // 'a' <= byte && byte <= 'f' + return [byte - 97 + 10, true]; + case 65 <= byte && byte <= 70: // 'A' <= byte && byte <= 'F' + return [byte - 65 + 10, true]; + } + return [0, false]; +} + +/** + * EncodedLen returns the length of an encoding of n source bytes. Specifically, + * it returns n * 2. + * @param n + */ +export function encodedLen(n: number): number { + return n * 2; +} + +/** + * Encode encodes `src` into `encodedLen(src.length)` bytes of `dst`. + * As a convenience, it returns the number of bytes written to `dst` + * but this value is always `encodedLen(src.length)`. + * Encode implements hexadecimal encoding. + * @param dst + * @param src + */ +export function encode(dst: Uint8Array, src: Uint8Array): number { + const srcLength = encodedLen(src.length); + if (dst.length !== srcLength) { + throw new Error("Out of index."); + } + for (let i = 0; i < src.length; i++) { + const v = src[i]; + dst[i * 2] = hextable[v >> 4]; + dst[i * 2 + 1] = hextable[v & 0x0f]; + } + return srcLength; +} + +/** + * EncodeToString returns the hexadecimal encoding of `src`. + * @param src + */ +export function encodeToString(src: Uint8Array): string { + const dest = new Uint8Array(encodedLen(src.length)); + encode(dest, src); + return new TextDecoder().decode(dest); +} + +/** + * Decode decodes `src` into `decodedLen(src.length)` bytes + * returning the actual number of bytes written to `dst`. + * Decode expects that `src` contains only hexadecimal characters and that `src` + * has even length. + * If the input is malformed, Decode returns the number of bytes decoded before + * the error. + * @param dst + * @param src + */ +export function decode( + dst: Uint8Array, + src: Uint8Array +): [number, Error | void] { + let i = 0; + for (; i < Math.floor(src.length / 2); i++) { + const [a, aOK] = fromHexChar(src[i * 2]); + if (!aOK) { + return [i, errInvalidByte(src[i * 2])]; + } + const [b, bOK] = fromHexChar(src[i * 2 + 1]); + if (!bOK) { + return [i, errInvalidByte(src[i * 2 + 1])]; + } + + dst[i] = (a << 4) | b; + } + + if (src.length % 2 == 1) { + // Check for invalid char before reporting bad length, + // since the invalid char (if present) is an earlier problem. + const [, ok] = fromHexChar(src[i * 2]); + if (!ok) { + return [i, errInvalidByte(src[i * 2])]; + } + return [i, errLength()]; + } + + return [i, undefined]; +} + +/** + * DecodedLen returns the length of a decoding of `x` source bytes. + * Specifically, it returns `x / 2`. + * @param x + */ +export function decodedLen(x: number): number { + return Math.floor(x / 2); +} + +/** + * DecodeString returns the bytes represented by the hexadecimal string `s`. + * DecodeString expects that src contains only hexadecimal characters and that + * src has even length. + * If the input is malformed, DecodeString will throws an error. + * @param s the `string` need to decode to `Uint8Array` + */ +export function decodeString(s: string): Uint8Array { + const src = new TextEncoder().encode(s); + // We can use the source slice itself as the destination + // because the decode loop increments by one and then the 'seen' byte is not + // used anymore. + const [n, err] = decode(src, src); + + if (err) { + throw err; + } + + return src.slice(0, n); +} |