import { normalizePhotonFeature, photonSearchUrl } from '../keylessGeocoder.js'; import { parseCoordinateQuery } from './coordinateParser.js'; import { requestWithDeadline, waitForSignal, } from '../sources/requestDeadline.js'; const MAX_QUERY_CHARS = 200; const CJK = /[\p{Script=Han}\p{Script=Hangul}\p{Script=Hiragana}\p{Script=Katakana}]/u; /** Han, Hangul or Kana text, whose place names are often two characters. */ export const isCjk = (text) => CJK.test(String(text || '')); /** The fewest characters a place name is looked up from. */ export const minPlaceChars = (text) => (isCjk(text) ? 2 : 3); /** * Text that may hold something private, which is never looked up, offered * as a place or sent anywhere: * - an "@" with something right on both sides (an email or a handle); * - seven digits or more (in any script, full-width too) with nothing but * spaces, punctuation, symbols or format characters between them * (phone, card, SSN and account numbers, and their starts); * - a phone number written with dots ("415.555.0100"); * - two letters, two digits and ten letters or digits more (an IBAN's * shape), spaces allowed; * - "password", "passcode", "passwd", "ssn", "pw", "pwd" or "bearer" with * anything after it, "token", "secret", "api key" or "key" with ":" or * "=" after it, "api key", "access token" or "secret key" with a word of * six characters or more after it, the word for "password" in the * interface's languages with anything after it, or "pin" (also "pin * code", "pin number", "pin is") with a digit after it; * - a word starting "sk_", "sk-", "rk_", "pk_live", "ghp_" or a Slack token's "xoxb-" and the like (any * case), or "AIza", "eyJ" or "AKIA"; * - a word that may be a password or key (see mayBeSecret). * The text is read after NFKC, so full-width letters and "@" count too. * Text that parses as coordinates or a grid reference is a spot, not a * secret. */ export function mayBePrivate(text) { return privacyReason(text) !== null; } /** * Whether the only thing private-looking in a line is a long run of digits, * as an address's house number and postcode can be ("Москва, ул. Тверская, * д. 13, 125009"): such a line may be searched as a place when the user * picks that on purpose. Never for an email, handle, password, key, code, * IBAN or anything else. * @param {string} text */ export function onlyNumbersLookPrivate(text) { // Read as privacyReason reads it: "⁴¹¹¹", "①②", "card" and words split by // zero-width spaces count as the digits and words they fold to. const line = foldLine(text); const groups = line.match(/\d+/g) || []; return ( privacyReason(text) === 'number' && /\p{L}{2,}/u.test(line) && // Address-sized: no group over six digits, nine digits at most in all // (a house number and a postcode), and no word that says what a number // is ("card 4111 …", "phone +1 415 …", "token 12345678"). groups.every((group) => group.length <= 6) && groups.join('').length <= 9 && !/(? (/\d/.test(text) ? 10 : 8); /** * Common words no wallet's recovery phrase holds (they are not on the BIP39 * list): a line with one is a sentence, however many short words it has. */ const NOT_SEED_WORDS = new Set([ 'the', 'and', 'are', 'for', 'from', 'with', 'how', 'why', 'which', 'would', 'could', 'should', 'please', 'does', 'was', 'were', 'flights', 'show', ]); /** * "Boise ID 83702": Idaho's code before a ZIP code, after a town's name * (not after "My" or "Employee", whose "ID" is an identity number). */ const IDAHO_BEFORE_ZIP = /(?<=(?:^|\s)(?!(?:My|Your|Our|His|Her|Their|Employee|Student|Staff|Member|User|Customer|Patient|National|Voter|Photo|Tax|Card)\b)\p{Lu}\p{Ll}+,?\s)ID(?=\.?,?\s+\d{5}(?:-\d{4})?(?!\S))/gu; /** * Words that say a number is an account, a card, a phone or a secret, in the * interface's languages where they differ ("Karte", "Telefon", "tarjeta"). */ const NUMBER_CONTEXT = /(?]\s*[\p{L}\p{N}-]/iu.test(line) || /\b\S+\s+at\s+\S+\s+(?:dot|punto|ponto|period|point|punkt|[([{<]\s*dot\s*[)\]}>])\s+\p{L}{2,24}\b/iu.test( line, ) || /\b\S+\s+at\s+[\p{L}\p{N}-]+\s*[([{<]\s*dot\s*[)\]}>]\s*\p{L}{2,24}\b/iu.test( line, ) || /\b[\p{L}\p{N}._%+-]+\s+at\s+[\p{L}\p{N}-]+(?:\.[\p{L}\p{N}-]+)*\s*\.\s*(?:com|net|org|edu|gov|mil|io|co|me|info|biz|app|dev|ai|[a-z]{2})\b/iu.test( line, ) ) return 'other'; // An IP address ("10.0.0.1"). if (/(? { const bare = token.replace(/^[[(]+|[\]),.;]+$/g, '').replace(/%\w+$/, ''); const colons = (bare.match(/:/g) || []).length; return ( /^[0-9a-f:.]+$/i.test(bare) && colons >= 2 && (bare.includes('::') || colons >= 5) ); }) ) return 'other'; // A wallet's recovery phrase, whole or nearly ("abandon ability able // …"): twelve to twenty-four words of three to eight letters, none a // common word the phrase list lacks ("can you tell where the most …" is // a question), and not starting as a request does. Too long to be sent // for suggestions anyway, it is kept from the assistant too. const words = line.trim().toLowerCase().split(/\s+/); if ( words.length >= 12 && words.length <= 24 && words.every((word) => /^[a-z]{3,8}$/.test(word)) && !words.some((word) => NOT_SEED_WORDS.has(word)) && !REQUEST_START.test(words[0]) ) return 'other'; // Coordinates and grid references are a spot, not a number to keep: only // text the place search itself reads as coordinates, as typed ("١٢٣٤, // ٥٦٧٨" folded would read as a spot, but is not one). if (parseCoordinateQuery(input)) return null; // Seven digits or more with only spaces, dashes or other marks between // them (any dash, ".", "_", ",", ":", "|", zero-width spaces): a phone number // or its start ("+1 415 555"), a card or its start ("4111 1111"), an SSN // ("123 45 6789", "123-45-6789", "123456789"), an account number. Dots // ("4111.1111.1111.1111", "123–45–6789"); coordinates were let through // above. // A street address's house number and postcode ("Hauptstrasse 12 10115 // Berlin", "Calle Mayor 10, 28013 Madrid", "Av Paulista 1578 01310-200") // are read apart first. // So are postcodes written the way they are: a US ZIP+4 after a word // ("DC 20500-0003"), a Brazilian CEP ("01310-200") and a Japanese one // after "〒" or before the address ("〒100-0001", "100-0001 東京都"). // "#" (a street number's sign, "Calle 100 # 19-61") splits digits too. // A word that says what the number is ("card 4111 111111", "call 0207 // 946095") keeps the digits together: no address is read out of them. const named = NUMBER_CONTEXT.test(line); const digits = ( named ? line : line .replace( // After a street word, or before the city ("12, 10115 Berlin"); // "030 12345" alone is a phone number. /(?<=\p{L}\.?\s)[1-9]\d{0,3}[a-z]?,?\s+\d{5}(?:-\d{3})?(?=\s+(?!(?:thanks|thank|thx|please|pls|ok|okay|now|asap|call|text)\b)\p{L}|$)|(?<=^|\s)[1-9]\d{0,3}[a-z]?,?\s+\d{5}(?:-\d{3})?(?=\s+(?!(?:thanks|thank|thx|please|pls|ok|okay|now|asap|call|text)\b)\p{L})/giu, ' ', ) // Other postcodes after a street's house number and before the city (a // four-digit one after a comma, so "is 555 0100 thanks" stays a number): // "Nowy Świat 10, 00-950 Warszawa", "Drottninggatan 53, 111 21 // Stockholm", "Mariahilfer Straße 120, 1070 Wien", "Rua Augusta 274, // 1100-053 Lisboa", "Tverskaya 13 125009 Moscow". .replace( /(?<=\p{L}\.?\s)\d{1,4}[a-z]?(?:,\s*\d{4}(?:-\d{3})?|,?\s+(?:\d{2}-\d{3}|\d{3} \d{2}|\d{6}))(?=\s+\p{L})/giu, ' x ', ) // ZIP+4 after a state ("DC 20500-0003", "MA. 02139-4307", // "California 90210-1234"), not after any word ("Alice // 12345-6789"). .replace(US_ZIP4, ' x ') .replace(/\b\d{5}-\d{3}\b/g, ' x ') .replace( /〒\s?\d{3}-\d{4}|\b\d{3}-\d{4}(?=\s*[\p{Script=Han}\p{Script=Hiragana}\p{Script=Katakana}])|(?<=[\p{Script=Han}\p{Script=Hiragana}\p{Script=Katakana}]\s*)\d{3}-\d{4}\b/gu, ' x ', ) .replace(/#/g, ' x ') ).replace(/[^\p{L}\p{N}]/gu, ''); const longNumber = /\d{7,}/.test(digits) || // A phone number written with dots ("415.555.0100"). /\b\d{3}\.\d{3}\.\d{4}\b/.test(line); // An IBAN. if (/\b[a-z]{2}\d{2}(?:\s?[a-z0-9]){10,}/i.test(line)) return 'other'; if ( /\b(?:password|passcode|passwd|pw|pwd)\b\s*[:=#]?\s*\S/i.test(line) || // "Bearer Island" and "SSN Station" are places: after "bearer" any value // but a capitalised name is a token, and a social security number has // digits. bearerToken(line) || /\bssn\b\s*[:=#]?\s*\d/i.test(line) || // The word for "password" in the interface's languages. // A whole word only: "Sachsenhausen" holds "senha" but is a place. /(? token.replace(/^[^\p{L}\p{N}]+|[^\p{L}\p{N}]+$/gu, '')) .some( (token) => /^(?:sk[-_]|rk_|pk_live|ghp_|gho_|ghu_|ghs_|github_pat_|glpat-|hf_|xox[abposr]-)/i.test( token, ) || /^SG\.[\w-]{8,}\.[\w-]{8,}$/.test(token) || /^(?:AIza|eyJ)/.test(token) || /^AKIA[0-9A-Z]{12,}$/.test(token) || mayBeSecret(token), ) ) return 'other'; return longNumber ? 'number' : null; } /** * A value after "bearer" that is not a word a place's name is made of * ("Bearer Island"); after "Authorization:" any value is a token. */ function bearerToken(line) { if (/\bauthori[sz]ation\s*:\s*bearer\b\s*\S/i.test(line)) return true; for (const match of line.matchAll(/\bbearer\b\s*[:=#]?\s*(\S+)/gi)) { const value = match[1].replace(/[,.;:!?)]+$/, ''); if (PLACE_WORDS.test(value)) continue; return true; } return false; } /** Words a place's name is made of, after a label word ("bearer island"). */ const PLACE_WORDS = /^(?:island|islands|isle|lake|creek|river|point|bay|mountain|mount|hill|hills|valley|station|road|street|lane|avenue|park|beach|falls|springs?|county|city|town|village|plaza|square|mall|center|centre|tower|building|court|bank)$/i; /** * A code after a label that can only be a code, without a colon ("PIN * A7B2", "verification code ABCDEF", "otp 7GH2K"): holding a digit, or in * capitals after a label that is not, and ending the line. "Gate Code Lane", * "GATE CODE LANE", "OTP Bank" and "Auth Code ABCD Street" stay places. */ function codeAfterLabel(line) { // Labels that only ever name a code take one in lower case too ("otp // abcdef", "cvv abc", "verification code abcdef"); a capitalised word // after them is a name ("OTP Bank", "MFA Boston"). // After a label in lower case, any value is a code ("verification code // Abc"); after one written as a name, only a value that is not one. const coded = /\b((?:otp|2fa|mfa|cvv|cvc|(?:verification|one-time|security|auth|login) code)(?:\s+is)?)\s+([a-z0-9][a-z0-9-]{2,})(?:\s+(?:thanks|thank you|thx|please|pls|ok))?\s*[.!]?\s*$/i.exec( line, ); if (coded) { const [, label, value] = coded; const labelIsName = /\p{Lu}/u.test(label); // A word a place's name is made of is no code ("security code plaza", // "otp bank"); after the short labels a code in letters is written in // capitals ("otp ABCDEF"), so a lower-case word is a name ("mfa // boston"). const shortLabel = /^(?:otp|2fa|mfa)\b/i.test(label); const named = PLACE_WORDS.test(value) || (shortLabel && /^\p{Ll}+$/u.test(value)); if (!labelIsName && !named) return true; if (labelIsName && !/^\p{Lu}(?:\p{Ll}+|\p{Lu}+)$/u.test(value)) return true; } // A value in mixed case that is no name ("PIN aBcD", "gate code aBcD"). const mixed = /\b(?:pin|(?:door|access|gate|alarm|entry) code)(?:\s+is)?\s+([a-z0-9-]{4,})\s*[.!]?\s*$/i.exec( line, ); if ( mixed && /\p{Ll}/u.test(mixed[1]) && /\p{Lu}/u.test(mixed[1]) && !/^\p{Lu}\p{Ll}+$/u.test(mixed[1]) ) return true; const match = /\b((?:pin|cvv|cvc|otp|2fa|mfa|(?:door|access|gate|alarm|entry|security|verification|one-time|auth|login) code)(?:\s+is)?)\s+([a-z0-9][a-z0-9-]{3,})\s*[.!]?\s*$/i.exec( line, ); if (!match) return false; const [, label, value] = match; if (/\d/.test(value)) return true; return value === value.toUpperCase() && label !== label.toUpperCase(); } /** * A word that may be a password or key, with letters and digits mixed * within one run: eight characters or more, four letters or more with a digit ("hunter22", * "letmein1"), or letters, digits and a symbol together ("Tr0ub4dor&3"). * Callsigns (UAL2110, BAW15K, N123AB) and postcodes are shorter or have * fewer letters. */ function mayBeSecret(token) { // Letters and digits must mix within one run ("hunter22", "w0rd"): a // number beside a word ("Yanino-1", "2-chōme", "656/G.B.", "Route-66") // is a name. // Latin letters only: CJK addresses are written without spaces // ("東京都港区六本木6丁目", "北京市朝阳区建国路88号"). // A UK postcode written without its space ("EC1A1BB", "BT487NN"). if (/^[a-z]{1,2}\d[a-z\d]?\d[a-z]{2}$/i.test(token)) return false; const mixed = token .split(/[^\p{L}\p{N}]+/u) .some((run) => /\p{Script=Latin}/u.test(run) && /\d/.test(run)); if (!mixed) return false; const letters = (token.match(/\p{L}/gu) || []).length; const digits = (token.match(/\d/g) || []).length; if (!letters || !digits) return false; if (token.length >= 8) return true; if (token.length >= 6 && /[^\p{L}\p{N}'.,\-/]/u.test(token)) return true; return token.length >= 6 && letters >= 4; } const MIN_GAP_MS = 500; const TIMEOUT_MS = 6000; const PAUSE_AFTER_REFUSAL_MS = 60_000; const CACHE_LIMIT = 200; /** * The spacing, the pause and the cache, shared by every suggester in the tab: * a search box that is put up again starts from where the last one left off. */ const shared = { nextRequestAt: 0, pausedUntil: 0, cache: new Map(), /** Requests on their way, by cache key: the same text asked again waits. */ pending: new Map(), }; /** Forget the shared spacing, pause and cache (tests). */ export function resetPlaceSuggestions() { shared.nextRequestAt = 0; shared.pausedUntil = 0; shared.cache.clear(); shared.pending.clear(); } /** * Countries that write a house's number before its street ("10 Downing * Street", "12 rue de Rivoli"); elsewhere it follows ("Hauptstraße 12"). */ const NUMBER_FIRST = new Set([ 'US', 'CA', 'GB', 'IE', 'AU', 'NZ', 'FR', 'LU', 'MC', 'ZA', 'IN', 'PH', 'SG', 'MY', 'HK', 'TH', 'LK', 'NG', 'GH', 'KE', 'JM', 'TT', ]); /** A house's number and street as its country writes them, or ''. */ export function houseAddress({ houseNumber, street, countryCode }) { if (!houseNumber || !street) return ''; return NUMBER_FIRST.has(String(countryCode || 'US').toUpperCase()) ? `${houseNumber} ${street}` : `${street} ${houseNumber}`; } /** List Photon candidates while typing. The submitted search owns navigation. */ export function createPlaceSuggester({ endpoint, fetchImpl = (...args) => globalThis.fetch(...args), limit = 4, } = {}) { const enabled = endpoint === undefined || Boolean(endpoint); const { cache } = shared; return { enabled, async suggest(query, { signal } = {}) { signal?.throwIfAborted(); const text = String(query ?? '') .trim() .slice(0, MAX_QUERY_CHARS); // A line longer than a place's name is never sent: a pasted sentence // or phrase is no place, and may be private in ways not read here. if ( !enabled || text.length < minPlaceChars(text) || text.split(/\s+/).length > suggestMaxWords(text) || mayBePrivate(text) ) return []; const language = (globalThis.document?.documentElement?.lang || 'en') .slice(0, 2) .toLowerCase(); const key = `${endpoint ?? ''}|${limit}|${language}|${text.toLowerCase()}`; if (cache.has(key)) return cache.get(key); // The same text already on its way is that request's answer, unless // its caller cancels it: then this one asks for itself. const inFlight = shared.pending.get(key); if (inFlight) try { return await waitForSignal(inFlight, signal); } catch { signal?.throwIfAborted(); } const ask = async () => { // Concurrent callers recheck the gap after waking before sending. while (shared.nextRequestAt > Date.now()) { let timer; const pause = new Promise((resolve) => { timer = setTimeout(resolve, shared.nextRequestAt - Date.now()); }); try { await waitForSignal(pause, signal); } finally { clearTimeout(timer); } } signal?.throwIfAborted(); // The same text asked again while this one waited is answered now. if (cache.has(key)) return cache.get(key); if (Date.now() < shared.pausedUntil) return []; const url = new URL(photonSearchUrl(text, { endpoint, limit })); if (['en', 'de', 'fr', 'it'].includes(language)) url.searchParams.set('lang', language); shared.nextRequestAt = Date.now() + MIN_GAP_MS; let features; try { features = await requestWithDeadline( async (requestSignal) => { const response = await fetchImpl(url, { signal: requestSignal }); requestSignal.throwIfAborted(); if (!response.ok) throw new Error(`HTTP ${response.status}`); const data = await response.json(); return data?.features; }, { signal, timeoutMs: TIMEOUT_MS }, ); } catch (error) { signal?.throwIfAborted(); // A refusal, a timeout or a failed request (a refusal sent without // CORS headers arrives as one) all hold suggestions off a while. void error; shared.pausedUntil = Date.now() + PAUSE_AFTER_REFUSAL_MS; return []; } signal?.throwIfAborted(); if (!Array.isArray(features)) return []; const seen = new Set(); const places = []; for (const feature of features) { const normalized = normalizePhotonFeature(feature); if (!normalized) continue; const { city, state, country, countrycode: countryCode, osm_value: osmType, type, street, housenumber: houseNumber, locality, district, } = feature.properties ?? {}; // A house has no name of its own: its street (or, without one, its // locality) and number are its name, written in the country's own // order, so two houses on one street stay two places. const address = houseAddress({ houseNumber, street: street || locality || district || city, countryCode, }); const name = normalized.name || address || city || state || country; if (!name) continue; const detail = [...new Set([city, state, country])] .filter((part) => part && part !== name) .join(', '); const id = `${name}|${detail}`; if (seen.has(id)) continue; seen.add(id); const { southwest, northeast } = normalized.viewport ?? {}; places.push({ name, detail, latitude: normalized.lat, longitude: normalized.lng, type: (osmType === 'administrative' ? type : osmType || type) || '', state: state || null, country: country || null, countryCode: /^[a-z]{2}$/i.test(countryCode ?? '') ? countryCode.toUpperCase() : null, extent: southwest ? [southwest.lng, northeast.lat, northeast.lng, southwest.lat] : null, }); if (places.length >= limit) break; } cache.set(key, places); if (cache.size > CACHE_LIMIT) cache.delete(cache.keys().next().value); return places; }; const asking = ask(); shared.pending.set(key, asking); try { return await asking; } finally { if (shared.pending.get(key) === asking) shared.pending.delete(key); } }, }; } /** * The cities that lie in a region: inside its extent (`[west, north, east, * south]`, as Photon gives it) and, when the region names a country by its * ISO code, in that country. Without either there is nothing to tell a * region's cities from its neighbours', so none are listed. * @param {Array<{latitude: number, longitude: number, country?: string}>} cities * Bundled cities, `country` an ISO 3166-1 code. * @param {{extent?: ?Array, countryCode?: ?string}} region */ export function citiesInRegion(cities, { extent, countryCode } = {}) { const box = Array.isArray(extent) && extent.length === 4 ? extent : null; if (!box && !countryCode) return []; return cities.filter( (city) => (!countryCode || city.country === countryCode) && (!box || inExtent(box, city)), ); } /** * Whether a place lies in an extent `[west, north, east, south]`. An extent * across the antimeridian (west east of east) runs from west round to east. * @param {Array} extent * @param {{latitude: number, longitude: number}} place */ export function inExtent([west, north, east, south], place) { const lon = place.longitude; const inLongitude = west <= east ? lon >= west && lon <= east : lon >= west || lon <= east; return place.latitude <= north && place.latitude >= south && inLongitude; }