"use strict"; // @TODO: Migrate this to a better named package... import * as errors from "@ethersproject/errors"; import { Bytes, concat, hexlify } from "@ethersproject/bytes"; import { toUtf8Bytes } from "@ethersproject/strings"; import { keccak256 } from "@ethersproject/keccak256"; /////////////////////////////// const Zeros = new Uint8Array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]); const Partition = new RegExp("^((.*)\\.)?([^.]+)$"); const UseSTD3ASCIIRules = new RegExp("^[a-z0-9.-]*$"); export function namehash(name: string): string { if (typeof(name) !== "string") { errors.throwError("invalid address - " + String(name), errors.INVALID_ARGUMENT, { argument: "name", value: name }); } name = name.toLowerCase(); // Supporting the full UTF-8 space requires additional (and large) // libraries, so for now we simply do not support them. // It should be fairly easy in the future to support systems with // String.normalize, but that is future work. if (!name.match(UseSTD3ASCIIRules)) { errors.throwError("contains invalid UseSTD3ASCIIRules characters", errors.INVALID_ARGUMENT, { argument: "name", value: name }); } let result: string | Uint8Array = Zeros; while (name.length) { let partition = name.match(Partition); let label = toUtf8Bytes(partition[3]); result = keccak256(concat([result, keccak256(label)])); name = partition[2] || ""; } return hexlify(result); } export function id(text: string): string { return keccak256(toUtf8Bytes(text)); } export const messagePrefix = "\x19Ethereum Signed Message:\n"; export function hashMessage(message: Bytes | string): string { if (typeof(message) === "string") { message = toUtf8Bytes(message); } return keccak256(concat([ toUtf8Bytes(messagePrefix), toUtf8Bytes(String(message.length)), message ])); }