Create title index for zims missing them

.gitignore

@@ -0,0 +1,10 @@
+.idea
+.vscode
+
+logs
+*.log
+
+bin/*.dll
+bin/kiwix*
+node_modles
+zims

bin/install-kwix.js

@@ -0,0 +1,87 @@
+import fs from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
+import https from 'node:https';
+import {execFileSync} from 'node:child_process';
+import {fileURLToPath} from 'node:url';
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+
+const VERSION = '3.8.1';
+const BASE_URL = 'https://download.kiwix.org/release/kiwix-tools';
+const PLATFORM_MAP = {linux: 'linux', darwin: 'macos', win32: 'win'};
+
+const BIN_NAME = process.platform === 'win32' ? 'kiwix-search.exe' : 'kiwix-search';
+const BIN_DIR = path.join(__dirname, '..', 'bin'); // project root/bin
+const BIN_PATH = path.join(BIN_DIR, BIN_NAME);
+
+/** Download a file, following redirects. */
+function download(url, dest) {
+	return new Promise((resolve, reject) => {
+		const file = fs.createWriteStream(dest);
+		https.get(url, res => {
+			if (res.statusCode >= 300 && res.statusCode < 400 && res.headers.location) {
+				file.close();
+				return resolve(download(res.headers.location, dest));
+			}
+			if (!res.statusCode || res.statusCode >= 400) {
+				return reject(new Error(`Download failed: ${res.statusCode} ${res.statusMessage}`));
+			}
+			res.pipe(file);
+			file.on('finish', () => file.close(resolve));
+		}).on('error', err => fs.unlink(dest, () => reject(err)));
+	});
+}
+
+/** Recursively search a directory for a file by name. */
+function findFile(dir, name) {
+	for (const entry of fs.readdirSync(dir, {withFileTypes: true})) {
+		const full = path.join(dir, entry.name);
+		if (entry.isDirectory()) {
+			const found = findFile(full, name);
+			if (found) return found;
+		} else if (entry.name === name) {
+			return full;
+		}
+	}
+	return null;
+}
+
+async function downloadAndExtract() {
+	const platformName = PLATFORM_MAP[process.platform];
+	if (!platformName) throw new Error(`Unsupported platform: ${process.platform}`);
+
+	const ext = platformName === 'win' ? 'zip' : 'tar.gz';
+	const archiveName = `kiwix-tools_${platformName}-x86_64-${VERSION}.${ext}`;
+	const url = `${BASE_URL}/${archiveName}`;
+	const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'kiwix-tools-'));
+	const archivePath = path.join(tmpDir, archiveName);
+
+	console.log('Downloading kiwix-tools from:', url);
+	await download(url, archivePath);
+
+	console.log('Extracting...');
+	execFileSync('tar', ['-xf', archivePath, '-C', tmpDir]); // bsdtar handles zip too
+
+	const extractedBin = findFile(tmpDir, BIN_NAME);
+	if (!extractedBin) throw new Error(`Could not find ${BIN_NAME} in extracted archive`);
+	const extractedDir = path.dirname(extractedBin);
+
+	await fs.promises.mkdir(BIN_DIR, {recursive: true});
+
+	// Copy the binary plus any DLLs sitting alongside it (Windows deps)
+	for (const entry of await fs.promises.readdir(extractedDir)) {
+		if (entry === BIN_NAME || entry.toLowerCase().endsWith('.dll')) {
+			await fs.promises.copyFile(path.join(extractedDir, entry), path.join(BIN_DIR, entry));
+		}
+	}
+	if (process.platform !== 'win32') await fs.promises.chmod(BIN_PATH, 0o755);
+
+	await fs.promises.rm(tmpDir, {recursive: true, force: true});
+	console.log('Installed to:', BIN_PATH);
+}
+
+downloadAndExtract().catch(err => {
+	console.error(err);
+	process.exit(1)
+}).finally(() => process.exit());

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@ztimson/zim-utils",
-	"version": "0.2.3",
+	"version": "0.2.5",
 	"description": "Native, dependency-light ZIM archive reader/searcher and Kiwix catalog downloader for Node.js",
 	"author": "Zak Timson",
 	"license": "MIT",
@@ -11,6 +11,9 @@
 		"url": "https://git.zakscode.com/ztimson/zim-utils"
 	},
 	"main": "src/index.js",
+	"scripts": {
+		"postinstall": "node ./bin/install-kwix.js"
+	},
 	"dependencies": {
 		"@ztimson/utils": "^0.30.7",
 		"lzma1": "^0.3.0",

src/reader.js

@@ -1,11 +1,10 @@
 'use strict';
 
 import fs from 'node:fs';
-import path from 'node:path';
+import {createHash} from 'node:crypto';
 import {decompressPool} from './decompress.js';
-import {fuzzyMatch, titleFromUrl} from './utils.js';
+import {fuzzyMatch, titleFromUrl, kiwixSearch} from './utils.js';
 
-const INDEX_VERSION = 1;
 const HEADER_SIZE = 80;
 const NS_CONTENT = 'C';
 const NS_METADATA = 'M';
@@ -14,18 +13,29 @@ const TITLE_SENTINEL = 0xffffffffffffffffn; // Indicator -> ZIM v6+ archives wit
 const DEFAULT_CLUSTER_CACHE_MAX = 32;
 const DEFAULT_CLUSTER_TTL = 60_000;
 
+// --- Fallback search index (only used for archives with no native title listing) ---
+const INDEX_SUFFIX = '.searchidx.bin';
+const INDEX_MAGIC = 'ZXI1';
+const INDEX_HEADER_SIZE = 24;     // magic(4) + staleness key(16) + recordCount(4)
+const INDEX_RECORD_SIZE = 20;     // keyOff(4) keyLen(2) urlOff(4) urlLen(2) titleOff(4) titleLen(2) mimetype(2)
+
 /** Native, dependency-light reader for .zim archives. Supports zstd & LZMA cluster compression. */
 export class ZimReader {
 	#fd = null;
 	#header = null;
 	#mimeTypes = [];
 	#hasTitleListing = false;
-	#index;
 	#clusterCache = new Map(); // clusterNumber -> {data, extended, timer}
 	#pending = new Map();      // clusterNumber -> Promise, dedupes concurrent misses
 	#clusterCacheMax;
 	#clusterTTL;
 
+	#indexPath;
+	#indexReady = null;   // Promise, awaited by search() before using the fallback index
+	#indexFd = null;      // open fd for the fallback index, once loaded/built
+	#indexRecordCount = 0;
+	#indexTableStart = 0;
+
 	get articleCount() { return this.#header?.articleCount ?? 0; }
 	get mediaCount() { return this.#header?.clusterCount ?? 0; }
 
@@ -34,18 +44,10 @@ export class ZimReader {
 		this.path = path;
 		this.#clusterCacheMax = clusterCacheMax;
 		this.#clusterTTL = clusterTTL;
+		this.#indexPath = `${path}${INDEX_SUFFIX}`;
 	}
 
-	/** Full O(n) scan over the URL pointer list, used when there's no title index. */
+	/** Binary search the URL pointer list for namespace+url. For exact-key lookups (readPage, metadata, icons). */
-	async #allDirents() {
-		const dirents = [];
-		for (let i = 0; i < this.#header.articleCount; i++) {
-			dirents.push(await this.#readDirent(await this.#ptr64(this.#header.urlPtrPos, i)));
-		}
-		return dirents;
-	}
-
-	/** Binary search the URL pointer list for namespace+url. */
 	async #findByUrl(url, namespace) {
 		const key = namespace + url;
 		let lo = 0, hi = this.#header.articleCount - 1;
@@ -60,6 +62,182 @@ export class ZimReader {
 		return null;
 	}
 
+	/** Binary search the ZIM's own title pointer list. O(log n), zero extra storage - the happy path. */
+	async #findByTitleBuiltin(title) {
+		let lo = 0, hi = this.#header.articleCount - 1;
+		while (lo <= hi) {
+			const mid = (lo + hi) >> 1;
+			const urlIdx = (await this.#read(this.#header.titlePtrPos + mid * 4, 4)).readUInt32LE(0);
+			const d = await this.#readDirent(await this.#ptr64(this.#header.urlPtrPos, urlIdx));
+			if (d.title === title) return d;
+			if (d.title < title) lo = mid + 1; else hi = mid - 1;
+		}
+		return null;
+	}
+
+	/**
+	 * Resolves a kiwix-search result (a title, not necessarily a url) to
+	 * {url, title, mimetype}. kiwix-search's fulltext index returns titles, and
+	 * a title isn't guaranteed to equal its url (unicode normalization,
+	 * disambiguation suffixes, punctuation stripping), so:
+	 *   1. URL binary search    - matches when title happens to equal url (free to check)
+	 *   2. Title pointer list   - when the archive ships one (Wikipedia etc. do)
+	 *   3. Persisted fallback index / linear scan - only for archives without (2)
+	 */
+	async #resolveSearchEntry(name) {
+		let dirent = await this.#findByUrl(name, NS_CONTENT);
+		if (dirent) return dirent;
+
+		if (this.#hasTitleListing) return this.#findByTitleBuiltin(name);
+
+		if (this.#indexReady) await this.#indexReady;
+		const hit = await this.#lookupFallbackIndex(name);
+		if (hit) return hit;
+
+		// Index unavailable (build failed - unwritable disk, etc.) or genuinely no match.
+		for (let i = 0; i < this.#header.articleCount; i++) {
+			const d = await this.#readDirent(await this.#ptr64(this.#header.urlPtrPos, i));
+			if (d.namespace === NS_CONTENT && (d.title === name || titleFromUrl(d.url) === name)) return d;
+		}
+		return null;
+	}
+
+	/**
+	 * A cheap, stable fingerprint for "is the cached index still valid for this file".
+	 * ZIM archives end with a 16-byte MD5 of their own contents (the same trailer
+	 * `zimcheck` validates against), so this is a single 16-byte read regardless of
+	 * archive size - no need to hash a multi-hundred-GB file. It's also
+	 * content-based rather than path/mtime-based, so moving or redownloading an
+	 * identical archive doesn't invalidate the cache. Falls back to a tiny
+	 * size+mtime hash only if the file is too short to have a real trailer.
+	 */
+	async #stalenessKey() {
+		const {size} = await fs.promises.stat(this.path);
+		if (size >= 16) return this.#read(size - 16, 16);
+		const stat = await fs.promises.stat(this.path);
+		return createHash('md5').update(`${stat.size}:${stat.mtimeMs}`).digest();
+	}
+
+	/** Walks every content dirent once, resolving redirects, keyed by both its title and its url-derived title. */
+	async #collectFallbackEntries() {
+		const map = new Map(); // key -> {url, title, mimetype}
+		for (let i = 0; i < this.#header.articleCount; i++) {
+			let dirent = await this.#readDirent(await this.#ptr64(this.#header.urlPtrPos, i));
+			if (dirent.namespace !== NS_CONTENT) continue;
+			dirent = await this.#resolveRedirect(dirent);
+			if (dirent.namespace !== NS_CONTENT) continue; // redirected outside content namespace
+
+			const entry = {url: dirent.url, title: dirent.title, mimetype: dirent.mimetype};
+			if (!map.has(dirent.title)) map.set(dirent.title, entry);
+			const urlTitle = titleFromUrl(dirent.url);
+			if (urlTitle !== dirent.title && !map.has(urlTitle)) map.set(urlTitle, entry);
+		}
+		return map;
+	}
+
+	/** Serializes the fallback index to disk: header, sorted fixed-width records, then a string table. */
+	async #buildFallbackIndex(stalenessKey) {
+		const map = await this.#collectFallbackEntries();
+		const keys = [...map.keys()].sort();
+
+		const records = Buffer.alloc(keys.length * INDEX_RECORD_SIZE);
+		const strings = [];
+		let tableOffset = 0;
+		keys.forEach((key, i) => {
+			const {url, title, mimetype} = map.get(key);
+			const keyBuf = Buffer.from(key, 'utf8');
+			const urlBuf = Buffer.from(url, 'utf8');
+			const titleBuf = Buffer.from(title, 'utf8');
+			const base = i * INDEX_RECORD_SIZE;
+
+			records.writeUInt32LE(tableOffset, base); records.writeUInt16LE(keyBuf.length, base + 4);
+			tableOffset += keyBuf.length;
+			records.writeUInt32LE(tableOffset, base + 6); records.writeUInt16LE(urlBuf.length, base + 10);
+			tableOffset += urlBuf.length;
+			records.writeUInt32LE(tableOffset, base + 12); records.writeUInt16LE(titleBuf.length, base + 16);
+			tableOffset += titleBuf.length;
+			records.writeUInt16LE(mimetype, base + 18);
+
+			strings.push(keyBuf, urlBuf, titleBuf);
+		});
+
+		const header = Buffer.alloc(INDEX_HEADER_SIZE);
+		header.write(INDEX_MAGIC, 0, 'ascii');
+		stalenessKey.copy(header, 4);
+		header.writeUInt32LE(keys.length, 20);
+
+		const tmpPath = `${this.#indexPath}.tmp-${process.pid}`;
+		await fs.promises.writeFile(tmpPath, Buffer.concat([header, records, ...strings]));
+		await fs.promises.rename(tmpPath, this.#indexPath);
+	}
+
+	/** Opens the fallback index file and caches its header fields for querying. */
+	async #openFallbackIndex() {
+		const fd = await fs.promises.open(this.#indexPath, 'r');
+		const header = Buffer.alloc(INDEX_HEADER_SIZE);
+		await fd.read(header, 0, INDEX_HEADER_SIZE, 0);
+		this.#indexFd = fd;
+		this.#indexRecordCount = header.readUInt32LE(20);
+		this.#indexTableStart = INDEX_HEADER_SIZE + this.#indexRecordCount * INDEX_RECORD_SIZE;
+	}
+
+	/**
+	 * Validates the on-disk fallback index against the archive's current staleness
+	 * key and (re)builds it if missing/stale. Failures are swallowed - callers see
+	 * #indexFd stay null and fall through to the linear-scan last resort.
+	 */
+	async #loadOrBuildFallbackIndex() {
+		const stalenessKey = await this.#stalenessKey();
+
+		try {
+			const fd = await fs.promises.open(this.#indexPath, 'r');
+			const header = Buffer.alloc(INDEX_HEADER_SIZE);
+			await fd.read(header, 0, INDEX_HEADER_SIZE, 0);
+			const fresh = header.toString('ascii', 0, 4) === INDEX_MAGIC && header.subarray(4, 20).equals(stalenessKey);
+			if (fresh) {
+				this.#indexFd = fd;
+				this.#indexRecordCount = header.readUInt32LE(20);
+				this.#indexTableStart = INDEX_HEADER_SIZE + this.#indexRecordCount * INDEX_RECORD_SIZE;
+				return;
+			}
+			await fd.close();
+		} catch { /* missing, corrupt, or unreadable -> rebuild below */ }
+
+		await this.#buildFallbackIndex(stalenessKey);
+		await this.#openFallbackIndex();
+	}
+
+	async #indexRead(pos, length) {
+		const buf = Buffer.alloc(length);
+		await this.#indexFd.read(buf, 0, length, pos);
+		return buf;
+	}
+
+	async #indexString(offset, length) {
+		if (!length) return '';
+		return (await this.#indexRead(this.#indexTableStart + offset, length)).toString('utf8');
+	}
+
+	/** Binary search over the on-disk record array. Never loads the full index into memory. */
+	async #lookupFallbackIndex(name) {
+		if (!this.#indexFd) return null;
+		let lo = 0, hi = this.#indexRecordCount - 1;
+		while (lo <= hi) {
+			const mid = (lo + hi) >> 1;
+			const rec = await this.#indexRead(INDEX_HEADER_SIZE + mid * INDEX_RECORD_SIZE, INDEX_RECORD_SIZE);
+			const keyOff = rec.readUInt32LE(0), keyLen = rec.readUInt16LE(4);
+			const key = await this.#indexString(keyOff, keyLen);
+			if (key === name) {
+				const urlOff = rec.readUInt32LE(6), urlLen = rec.readUInt16LE(10);
+				const titleOff = rec.readUInt32LE(12), titleLen = rec.readUInt16LE(16);
+				const [url, title] = await Promise.all([this.#indexString(urlOff, urlLen), this.#indexString(titleOff, titleLen)]);
+				return {url, title, mimetype: rec.readUInt16LE(18)};
+			}
+			if (key < name) lo = mid + 1; else hi = mid - 1;
+		}
+		return null;
+	}
+
 	/** Resets a cluster's idle-eviction timer. No-op when TTL disabled. */
 	#touch(clusterNumber, entry) {
 		if (!this.#clusterTTL) return;
@@ -185,82 +363,21 @@ export class ZimReader {
 		return this.#readDirent(await this.#ptr64(this.#header.urlPtrPos, dirent.redirectIndex));
 	}
 
-	/** Narrows to dirents near the term's alphabetical position in the title index. */
-	async #titleIndexCandidates(term) {
-		const q = term.toLowerCase();
-		let lo = 0, hi = this.#header.articleCount - 1;
-		while (lo < hi) {
-			const mid = (lo + hi) >> 1;
-			const urlIdx = (await this.#read(this.#header.titlePtrPos + mid * 4, 4)).readUInt32LE(0);
-			const dirent = await this.#readDirent(await this.#ptr64(this.#header.urlPtrPos, urlIdx));
-			if (dirent.title.toLowerCase() < q) lo = mid + 1; else hi = mid;
-		}
-		// Widen around the prefix match since fuzzy scoring isn't purely alphabetical.
-		const start = Math.max(0, lo - 50), end = Math.min(this.#header.articleCount, lo + 200);
-		const dirents = [];
-		for (let i = start; i < end; i++) {
-			const urlIdx = (await this.#read(this.#header.titlePtrPos + i * 4, 4)).readUInt32LE(0);
-			dirents.push(await this.#readDirent(await this.#ptr64(this.#header.urlPtrPos, urlIdx)));
-		}
-		return dirents;
-	}
-
-	/** Path of the cached word index, kept in an `index/` subdirectory next to the archive. */
-	#indexPath() {
-		return path.join(path.dirname(this.path), 'index', `${path.basename(this.path)}.idx.json`);
-	}
-
-	/** Builds (or loads a cached) word -> entry index, avoiding a re-scan on every search. */
-	async #wordIndex() {
-		if (this.#index) return this.#index;
-		const cachePath = this.#indexPath();
-		const stat = await fs.promises.stat(this.path);
-
-		if (fs.existsSync(cachePath)) {
-			try {
-				const cached = JSON.parse(await fs.promises.readFile(cachePath, 'utf8'));
-				// Rebuild if stale: index predates the archive's current mtime (e.g. a re-download).
-				if (cached.version === INDEX_VERSION && cached.mtimeMs >= stat.mtimeMs) {
-					return (this.#index = cached);
-				}
-			} catch {}
-		}
-
-		// Expensive full scan — only happens once per archive (or after it changes).
-		const dirents = await this.#allDirents();
-		const entries = [];
-		const words = new Map(); // word -> [entryIndex, ...]
-		for (const d of dirents) {
-			if (d.namespace !== NS_CONTENT) continue;
-			const idx = entries.length;
-			entries.push({url: d.url, title: d.title, mimetype: d.mimetype});
-			const text = `${d.title} ${titleFromUrl(d.url)}`.toLowerCase();
-			for (const w of text.split(/\W+/).filter(Boolean)) {
-				if (!words.has(w)) words.set(w, []);
-				words.get(w).push(idx);
-			}
-		}
-
-		const index = {version: INDEX_VERSION, size: stat.size, mtimeMs: stat.mtimeMs, entries, words: Object.fromEntries(words)};
-		await fs.promises.mkdir(path.dirname(cachePath), {recursive: true});
-		await fs.promises.writeFile(cachePath, JSON.stringify(index));
-		return (this.#index = index);
-	}
-
 	async close() {
 		if (this.#fd) await this.#fd.close();
 		this.#fd = null;
+		if (this.#indexFd) await this.#indexFd.close();
+		this.#indexFd = null;
 		for (const entry of this.#clusterCache.values()) clearTimeout(entry.timer);
 		this.#clusterCache.clear();
 		this.#pending.clear();
 	}
 
-	/** Deletes the zim archive and its cached index (if any). Safe to call on unopened readers. */
+	/** Deletes the zim archive and its cached fallback index (if any). Safe to call on unopened readers. */
 	async delete() {
 		await this.close();
-		this.#index = null;
 		await fs.promises.rm(this.path, {force: true});
-		await fs.promises.rm(this.#indexPath(), {force: true});
+		await fs.promises.rm(this.#indexPath, {force: true});
 	}
 
 	/** Reads an 'M' namespace metadata value (e.g. Name, Date, Title). Returns null if missing. */
@@ -302,6 +419,14 @@ export class ZimReader {
 			this.#fd = null;
 			throw e;
 		}
+
+		// The fallback index is only needed when the archive has no native title
+		// pointer list - #findByTitleBuiltin already covers that case in O(log n)
+		// with zero extra storage. Well-maintained archives (Wikipedia etc.)
+		// ship a title listing, so this path is expected to be rare in practice.
+		if (!this.#hasTitleListing) {
+			this.#indexReady = this.#loadOrBuildFallbackIndex().catch(() => {});
+		}
 		return this;
 	}
 
@@ -325,27 +450,15 @@ export class ZimReader {
 
 	/**
 	 * Fuzzy-ranked title search. Accepts comma-separated `terms` the same way the
-	 * catalog search does. Uses the sorted title index when present (binary search
+	 * catalog search does. Uses kiwix-search's embedded fulltext index as a prefilter
-	 * narrows the candidate window); falls back to a full linear scan otherwise
+	 * to narrow candidates before fuzzy scoring.
-	 * (common on ZIM v6+/zimit-generated archives with no title index).
 	 */
 	async search(terms, {limit = 20, htmlOnly = true} = {}) {
 		const termList = String(terms).split(',').map(t => t.trim()).filter(Boolean);
 		if (!termList.length) return [];
 
-		let candidates;
+		const names = await kiwixSearch(this.path, termList.join(' '));
-		if (this.#hasTitleListing && this.#header.articleCount > 5000) {
+		const candidates = (await Promise.all(names.map(n => this.#resolveSearchEntry(n)))).filter(Boolean);
-			candidates = await this.#titleIndexCandidates(termList[0]);
-		} else {
-			const {entries, words} = await this.#wordIndex();
-			// Pull candidates from postings of any word that starts with (or contains) the search term.
-			const q = termList[0].toLowerCase();
-			const idxSet = new Set();
-			for (const [word, postings] of Object.entries(words)) {
-				if (word.includes(q) || q.includes(word)) postings.forEach(i => idxSet.add(i));
-			}
-			candidates = [...idxSet].map(i => ({...entries[i], namespace: NS_CONTENT}));
-		}
 
 		const scored = [];
 		for (const dirent of candidates) {

src/utils.js

@@ -1,3 +1,32 @@
+import {execFile} from 'node:child_process';
+import path from 'node:path';
+import {fileURLToPath} from 'node:url';
+import {promisify} from 'node:util';
+
+/**
+ * Search a ZIM file's fulltext index.
+ * @param {string} zimPath - Path to the .zim file
+ * @param {string} pattern - Search terms
+ * @param {object} [opts]
+ * @param {boolean} [opts.suggestion] - Suggest titles from partial pattern (completion-style)
+ * @param {boolean} [opts.spelling] - Suggest spelling-corrected titles
+ * @returns {Promise<string[]>} Matching article/tag titles
+ */
+export async function kiwixSearch(zimPath, pattern, opts = {}) {
+	const execFileAsync = promisify(execFile);
+	const __dirname = path.dirname(fileURLToPath(import.meta.url));
+	const BIN_NAME = process.platform === 'win32' ? 'kiwix-search.exe' : 'kiwix-search';
+	const BIN_PATH = path.join(__dirname, '..', 'bin', BIN_NAME);
+
+	const args = [];
+	if (opts.suggestion) args.push('-s');
+	if (opts.spelling) args.push('--spelling');
+	args.push(zimPath, pattern);
+
+	const {stdout} = await execFileAsync(BIN_PATH, args);
+	return stdout.split('\n').map(line => line.trim()).filter(Boolean);
+}
+
 export function levenshtein(a, b) {
 	const m = a.length, n = b.length;
 	if (!m) return n;