repo.ts

1import { inflateSync } from "node:zlib";2import { blake3 } from "@noble/hashes/blake3.js";3import { bytesToHex } from "@noble/hashes/utils.js";4import type { BucketReader } from "./reader.ts";56export type SaveFile = { path: string; hash: string; executable: boolean };78/** The save record Koh 3 writes to `<repo>/saves/<id>` (§9.2). */9export type SaveRecord = {10	id: string;11	parent: string;12	message: string;13	created_at: number;14	machine: string;15	is_checkpoint: boolean;16	size_bytes: number;17	agent: string;18	session: string;19	restores: string;20	replayed_from: string;21	files: SaveFile[];22};2324/** A flag, clear or strip event, as Koh 3 writes it to `<repo>/events/<id>` (§9.2, §10). */25export type EventRecord = {26	id: string;27	kind: "flag" | "clear" | "strip";28	save_id: string;29	path: string;30	reason: string;31	created_at: number;32	machine: string;33};3435/** One built file of a release (§11.2). */36export type ReleaseFile = {37	platform: string;38	name: string;39	size: number;40	blake3: string;41	sha256: string;42};4344/** The release record Koh 3 writes to `<repo>/releases/<version>/release.json` (§11.2). */45export type ReleaseRecord = {46	format: 1;47	repo: string;48	version: string;49	save: string;50	published_at: number;51	machine: string;52	notes: string;53	files: ReleaseFile[];54};5556/** §11.1's platform names, Kepr 2's exact list: Peach Android and existing build files depend on them. */57export const PLATFORMS = [58	"linux-x86_64",59	"linux-aarch64",60	"macos-aarch64",61	"macos-x86_64",62	"windows-x86_64",63	"android-arm64",64	"android-x86",65] as const;6667/** A save record as it arrived, less the files §5.5 quarantines, named here escaped. */68export type ArrivingSave = SaveRecord & { quarantined: string[] };6970/** An object's stored zlib bytes and the inflated bytes they were verified from. */71export type VerifiedObject = { compressed: Uint8Array; bytes: Uint8Array };7273export class BucketError extends Error {}7475/** An object that inflates past Koh's own file limit; the paths pointing at it are quarantined. */76export class OversizedObject extends BucketError {}7778const MAX_OBJECT = 512 * 1024 * 1024;7980const SAVE_ID = /^[0-9a-f]{12}$/;81const REPO_NAME = /^[A-Za-z0-9._-]+$/;82const CONTROL_CHAR = /\p{Cc}/u;83const HEX64 = /^[0-9a-f]{64}$/;84const VERSION = /^[A-Za-z0-9.-]{1,32}$/;85// A file name that is safe as one key segment and inside a quoted Content-Disposition filename.86const FILE_NAME = /^[^/\\"\p{Cc}]{1,255}$/u;87const decoder = new TextDecoder();88// A run of whole UTF-8 sequences (RFC 3629), or one byte outside them, over a record read one89// character per byte.90const UTF8_RUN =91	/((?:[^\x80-\xff]|[\xc2-\xdf][\x80-\xbf]|\xe0[\xa0-\xbf][\x80-\xbf]|[\xe1-\xec\xee\xef][\x80-\xbf]{2}|\xed[\x80-\x9f][\x80-\xbf]|\xf0[\x90-\xbf][\x80-\xbf]{2}|[\xf1-\xf3][\x80-\xbf]{3}|\xf4[\x80-\x8f][\x80-\xbf]{2})+)|[\s\S]/g;92// A surrogate pair escape, a lone surrogate escape, or any other escape (`\\` included), whole.93const JSON_ESCAPE =94	/\\u[dD][89abAB][0-9a-fA-F]{2}\\u[dD][c-fC-F][0-9a-fA-F]{2}|\\u[dD][89a-fA-F][0-9a-fA-F]{2}|\\[\s\S]/g;9596// ponytail: one full listing per call; switch to a delimiter listing if the bucket grows large.97export async function listRepos(bucket: BucketReader): Promise<string[]> {98	const keys = await bucket.list("");99	return keys100		.filter((key) => key.endsWith("/head") && key.split("/").length === 2)101		.map((key) => key.slice(0, -"/head".length))102		.filter(isRepoName);103}104105function isRepoName(name: string): boolean {106	return REPO_NAME.test(name) && name !== "." && name !== "..";107}108109// Checked here rather than in route handlers, so no caller can build a key from unchecked input.110function requireRepo(repo: string): void {111	if (!isRepoName(repo)) throw new BucketError(`not a repo name: "${repo}"`);112}113114function requireSaveId(id: string): void {115	if (!SAVE_ID.test(id)) throw new BucketError(`not a save ID: "${id}"`);116}117118/** Applies §9.2's format rule: a missing key is version 1; anything but `koh-remote 1` is refused. */119export async function checkFormat(bucket: BucketReader, repo: string): Promise<void> {120	requireRepo(repo);121	const bytes = await bucket.get(`${repo}/format`);122	if (bytes === null) return;123	const format = decoder.decode(bytes).trim();124	if (format !== "koh-remote 1") {125		throw new BucketError(`${repo} uses remote format "${format}"; update Kepr to read it`);126	}127}128129export async function readHead(bucket: BucketReader, repo: string): Promise<string | null> {130	requireRepo(repo);131	const bytes = await bucket.get(`${repo}/head`);132	if (bytes === null) return null;133	const head = decoder.decode(bytes).trim();134	if (!SAVE_ID.test(head)) throw new BucketError(`${repo}/head is not a save ID: "${head}"`);135	return head;136}137138export async function readSave(139	bucket: BucketReader,140	repo: string,141	id: string,142): Promise<ArrivingSave> {143	requireRepo(repo);144	requireSaveId(id);145	const bytes = await bucket.get(`${repo}/saves/${id}`);146	if (bytes === null) throw new BucketError(`${repo}/saves/${id} is missing`);147	const record = parseSaveRecord(JSON.parse(decodeRecord(bytes)));148	if (record.id !== id) throw new BucketError(`${repo}/saves/${id} holds save ${record.id}`);149	return {150		...record,151		files: record.files.filter((file) => isSafePath(file.path)),152		quarantined: record.files153			.filter((file) => !isSafePath(file.path))154			.map((file) => escapePath(file.path)),155	};156}157158/**159 * A record's text, read as Koh reads it (\u00A75.5): an unpaired surrogate escape stands for its three160 * bytes, and each byte outside a valid UTF-8 sequence becomes the lone surrogate U+DC00 + byte, so161 * the path holding it fails `isWellFormed` while a genuine U+FFFD stays legal.162 */163function decodeRecord(bytes: Uint8Array): string {164	const text = Buffer.from(bytes)165		.toString("latin1")166		.replace(JSON_ESCAPE, (sequence) => {167			if (sequence.length !== 6 || sequence[1] !== "u") return sequence;168			const unit = Number.parseInt(sequence.slice(2), 16);169			return String.fromCharCode(170				0xe0 | (unit >> 12),171				0x80 | ((unit >> 6) & 0x3f),172				0x80 | (unit & 0x3f),173			);174		});175	return text.replace(UTF8_RUN, (run, valid: string | undefined) =>176		valid === undefined177			? String.fromCharCode(0xdc00 + run.charCodeAt(0))178			: decoder.decode(Buffer.from(valid, "latin1")),179	);180}181182/** Fetches an object, inflates it and checks its Blake3 against its key, as steal does (§9.4). */183export async function readObject(184	bucket: BucketReader,185	repo: string,186	hash: string,187): Promise<VerifiedObject> {188	requireRepo(repo);189	if (!HEX64.test(hash)) throw new BucketError(`not an object hash: "${hash}"`);190	const compressed = await bucket.get(`${repo}/objects/${hash}`);191	if (compressed === null) throw new BucketError(`${repo}/objects/${hash} is missing`);192	let bytes: Uint8Array;193	try {194		bytes = new Uint8Array(inflateSync(compressed, { maxOutputLength: MAX_OBJECT }));195	} catch (cause) {196		if (cause instanceof RangeError) {197			throw new OversizedObject(`${repo}/objects/${hash} inflates past 512 MiB`);198		}199		throw cause;200	}201	if (bytesToHex(blake3(bytes)) !== hash) {202		throw new BucketError(`${repo}/objects/${hash} does not match its hash`);203	}204	return { compressed, bytes };205}206207export async function listEventIds(bucket: BucketReader, repo: string): Promise<string[]> {208	requireRepo(repo);209	const prefix = `${repo}/events/`;210	// A folder remote's half-written temp files can show up in a listing; only IDs are events.211	return (await bucket.list(prefix))212		.map((key) => key.slice(prefix.length))213		.filter((id) => SAVE_ID.test(id));214}215216export async function readEvent(217	bucket: BucketReader,218	repo: string,219	id: string,220): Promise<EventRecord> {221	requireRepo(repo);222	requireSaveId(id);223	const bytes = await bucket.get(`${repo}/events/${id}`);224	if (bytes === null) throw new BucketError(`${repo}/events/${id} is missing`);225	const record = parseEventRecord(JSON.parse(decodeRecord(bytes)));226	if (record.id !== id) throw new BucketError(`${repo}/events/${id} holds event ${record.id}`);227	return record;228}229230/** §11.1's version rule, minus `.` and `..`, which would name a folder rather than a version. */231export function isVersion(value: string): boolean {232	return VERSION.test(value) && value !== "." && value !== "..";233}234235export function isPlatform(value: string): boolean {236	return (PLATFORMS as readonly string[]).includes(value);237}238239/** The body of `<repo>/releases/latest`, trimmed, or null when the repo has never released. */240export async function readReleasesLatest(241	bucket: BucketReader,242	repo: string,243): Promise<string | null> {244	requireRepo(repo);245	const bytes = await bucket.get(`${repo}/releases/latest`);246	if (bytes === null) return null;247	const version = decoder.decode(bytes).trim();248	if (!isVersion(version))249		throw new BucketError(`${repo}/releases/latest is not a version: "${version}"`);250	return version;251}252253/** Every version with a `release.json`: a version without one doesn't exist (§11.2). */254export async function listReleaseVersions(bucket: BucketReader, repo: string): Promise<string[]> {255	requireRepo(repo);256	return (await bucket.list(`${repo}/releases/`))257		.map((key) => key.split("/"))258		.filter((parts) => parts.length === 4 && parts[3] === "release.json")259		.map((parts) => parts[2] as string)260		.filter(isVersion);261}262263export async function readRelease(264	bucket: BucketReader,265	repo: string,266	version: string,267): Promise<ReleaseRecord> {268	requireRelease(repo, version);269	const key = `${repo}/releases/${version}/release.json`;270	const bytes = await bucket.get(key);271	if (bytes === null) throw new BucketError(`${key} is missing`);272	return checkRelease(repo, version, JSON.parse(decoder.decode(bytes)));273}274275/** A parsed `<repo>/releases/<version>/release.json`, if the sync would accept it (§11.2). */276export function checkRelease(repo: string, version: string, value: unknown): ReleaseRecord {277	requireRelease(repo, version);278	const record = parseReleaseRecord(value);279	if (record.repo !== repo || record.version !== version) {280		throw new BucketError(281			`${repo}/releases/${version}/release.json holds ${record.repo} ${record.version}`,282		);283	}284	return record;285}286287function requireRelease(repo: string, version: string): void {288	requireRepo(repo);289	if (!isVersion(version)) throw new BucketError(`not a release version: "${version}"`);290}291292function parseReleaseRecord(value: unknown): ReleaseRecord {293	const r = value as Record<string, unknown>;294	const valid =295		typeof value === "object" &&296		value !== null &&297		r.format === 1 &&298		typeof r.repo === "string" &&299		typeof r.version === "string" &&300		typeof r.save === "string" &&301		SAVE_ID.test(r.save) &&302		isCount(r.published_at) &&303		typeof r.machine === "string" &&304		typeof r.notes === "string" &&305		Array.isArray(r.files) &&306		r.files.every(isReleaseFile) &&307		// One file per platform per version (§11.2).308		new Set(r.files.map((file) => file.platform)).size === r.files.length;309	if (!valid) throw new BucketError("release record does not match §11.2");310	return value as ReleaseRecord;311}312313function isReleaseFile(value: unknown): value is ReleaseFile {314	const f = value as Record<string, unknown>;315	return (316		typeof value === "object" &&317		value !== null &&318		typeof f.platform === "string" &&319		isPlatform(f.platform) &&320		typeof f.name === "string" &&321		FILE_NAME.test(f.name) &&322		f.name !== "." &&323		f.name !== ".." &&324		isCount(f.size) &&325		typeof f.blake3 === "string" &&326		HEX64.test(f.blake3) &&327		typeof f.sha256 === "string" &&328		HEX64.test(f.sha256)329	);330}331332function parseSaveRecord(value: unknown): SaveRecord {333	const r = value as Record<string, unknown>;334	const valid =335		typeof value === "object" &&336		value !== null &&337		["message", "machine", "agent", "session"].every((key) => typeof r[key] === "string") &&338		typeof r.id === "string" &&339		SAVE_ID.test(r.id) &&340		["parent", "restores", "replayed_from"].every((key) => isOptionalSaveId(r[key])) &&341		isCount(r.created_at) &&342		isCount(r.size_bytes) &&343		typeof r.is_checkpoint === "boolean" &&344		Array.isArray(r.files) &&345		r.files.every(isSaveFile) &&346		new Set(r.files.map((file) => file.path)).size === r.files.length;347	if (!valid) throw new BucketError("save record does not match the Koh 3 schema");348	return value as SaveRecord;349}350351function parseEventRecord(value: unknown): EventRecord {352	const r = value as Record<string, unknown>;353	const valid =354		typeof value === "object" &&355		value !== null &&356		typeof r.id === "string" &&357		SAVE_ID.test(r.id) &&358		isOptionalSaveId(r.save_id) &&359		typeof r.path === "string" &&360		// A flag or clear is about one save and no path; a strip names a path, and one save or361		// (with no ID) all. Sync ignores a strip whose path is illegal, as Koh does (§5.5).362		(r.kind === "strip" ||363			((r.kind === "flag" || r.kind === "clear") && r.save_id !== "" && r.path === "")) &&364		typeof r.reason === "string" &&365		typeof r.machine === "string" &&366		isCount(r.created_at);367	if (!valid) throw new BucketError("event record does not match the Koh 3 schema");368	return value as EventRecord;369}370371function isOptionalSaveId(value: unknown): boolean {372	return value === "" || (typeof value === "string" && SAVE_ID.test(value));373}374375function isCount(value: unknown): boolean {376	return Number.isSafeInteger(value) && (value as number) >= 0;377}378379/** §5.5's legal path, clause for clause with Koh's `legal_path`: no empty, `.` or `..` segments, no backslash or control character, well-formed UTF-8. */380export function isSafePath(path: string): boolean {381	return (382		path.isWellFormed() &&383		!path.includes("\\") &&384		!CONTROL_CHAR.test(path) &&385		path.split("/").every((segment) => segment !== "" && segment !== "." && segment !== "..")386	);387}388389/** A name printable without its controls reaching the page, as Koh's `escape_path` prints it. */390export function escapePath(path: string): string {391	return path392		.replace(/[\\\p{Cc}\udc80-\udcff]/gu, (char) => {393			const code = char.charCodeAt(0);394			if (char === "\\") return "\\\\";395			// A byte `decodeRecord` found outside UTF-8.396			if (code >= 0xdc80) return `\\x${(code - 0xdc00).toString(16)}`;397			const hex = code.toString(16).padStart(2, "0");398			return code >= 0x80 ? `\\u00${hex}` : `\\x${hex}`;399		})400		.toWellFormed();401}402403function isSaveFile(value: unknown): value is SaveFile {404	const f = value as Record<string, unknown>;405	return (406		typeof value === "object" &&407		value !== null &&408		typeof f.path === "string" &&409		typeof f.hash === "string" &&410		HEX64.test(f.hash) &&411		typeof f.executable === "boolean"412	);413}