feat(search): compose Google and Photon through an explicit provider service

src/keylessGeocoder.js

@@ -0,0 +1,405 @@
+/** Photon/OpenStreetMap place-search adapter. */
+const PHOTON_ENDPOINT = 'https://photon.komoot.io/api/';
+
+/** Photon is a courtesy service; fail fast rather than hold the search open. */
+const PHOTON_TIMEOUT_MS = 6000;
+
+/** Bounded memo — a search box re-issues the same query on every keystroke. */
+const PHOTON_CACHE_MAX = 64;
+
+/**
+ * Candidates requested per pass. One is not enough: proximity bias reorders the
+ * list, so the name actually asked for can sit below a nearer near-miss.
+ */
+const PHOTON_RESULT_LIMIT = 5;
+const photonCache = new Map();
+
+/**
+ * Photon's own coarse class → the Google Place Type `geocodeNavigationMode`
+ * reads. Only types that function consults are worth emitting; anything else
+ * falls through to `precise-place`, which is the correct default for a house
+ * or a POI.
+ */
+const PHOTON_TYPE_TO_GOOGLE = Object.freeze({
+  country: 'country',
+  state: 'administrative_area_level_1',
+  county: 'administrative_area_level_2',
+  city: 'locality',
+  district: 'sublocality',
+  locality: 'locality',
+  street: 'route',
+});
+
+/**
+ * OSM `key=value` → Google Place Type. `null` means "this tag is a precise
+ * place" and, like any other entry, stops the coarse class from being consulted.
+ *
+ * Two failure directions matter, and both were found by querying the live
+ * service rather than by reasoning about OSM tagging:
+ *
+ * - Too coarse: a park or lake that arrives without an area type frames at
+ *   building range, which is the exact bug `geocodeNavigationMode`'s area types
+ *   exist to prevent — searching "Ho Guom" put the camera 26 m over the water.
+ * - Too broad: Photon classifies a town square as `type: 'locality'`, so
+ *   deferring to the coarse class frames Times Square as if it were a city.
+ *   The explicit `null` below is what stops that.
+ */
+const OSM_TAG_TO_GOOGLE = Object.freeze({
+  'leisure=park': 'park',
+  'leisure=nature_reserve': 'park',
+  'leisure=garden': 'park',
+  'leisure=stadium': 'stadium',
+  'boundary=national_park': 'park',
+  'boundary=protected_area': 'park',
+  'tourism=zoo': 'zoo',
+  'tourism=theme_park': 'amusement_park',
+  'amenity=university': 'university',
+  'amenity=college': 'university',
+  'amenity=grave_yard': 'cemetery',
+  'landuse=cemetery': 'cemetery',
+  'landuse=forest': 'natural_feature',
+  'aeroway=aerodrome': 'airport',
+  'shop=mall': 'shopping_mall',
+  'place=region': 'natural_feature',
+  'place=suburb': 'sublocality',
+  'place=neighbourhood': 'sublocality',
+  'place=quarter': 'sublocality',
+  'place=borough': 'sublocality',
+  'place=square': null,
+});
+
+/**
+ * OSM keys whose features are areas whatever their value — lakes, rivers,
+ * peaks, woods, protected land. Photon reports every one of these with
+ * `type: 'other'`, so the key is the only signal that they are not buildings.
+ */
+const AREA_OSM_KEYS = Object.freeze(
+  new Set(['natural', 'water', 'waterway', 'landuse']),
+);
+
+/**
+ * Google-style types for one Photon feature's properties.
+ *
+ * A specific `key=value` mapping is authoritative and ends the lookup; the
+ * coarse `type` is consulted only when no tag rule matched. Deferring to the
+ * coarse class after a tag rule is what mis-framed town squares as cities.
+ * @param {object} properties - Photon feature `properties`.
+ * @returns {string[]} Google Place Types, possibly empty (a precise place).
+ */
+export function photonResultTypes(properties) {
+  const key = String(properties?.osm_key || '');
+  const value = String(properties?.osm_value || '');
+  const tag = `${key}=${value}`;
+
+  if (Object.prototype.hasOwnProperty.call(OSM_TAG_TO_GOOGLE, tag)) {
+    const mapped = OSM_TAG_TO_GOOGLE[tag];
+    return mapped ? [mapped] : [];
+  }
+  if (AREA_OSM_KEYS.has(key)) return ['natural_feature'];
+  if (key === 'highway') return ['route'];
+
+  const coarse = PHOTON_TYPE_TO_GOOGLE[String(properties?.type || '')];
+  return coarse ? [coarse] : [];
+}
+
+/**
+ * Photon `extent` → the `{southwest,northeast}` bounds the flight code frames.
+ * Photon orders it [west, north, east, south]; a naive [w,s,e,n] read produces
+ * an inverted box that still looks like a valid viewport.
+ * @param {number[]} extent - Photon `properties.extent`.
+ * @returns {?{southwest:{lat:number,lng:number}, northeast:{lat:number,lng:number}}}
+ */
+export function photonExtentToBounds(extent) {
+  if (!Array.isArray(extent) || extent.length !== 4) return null;
+  const [west, north, east, south] = extent;
+  if (![west, north, east, south].every(Number.isFinite)) return null;
+  if (west > east) return null; // Wrapped bounds are omitted until framing supports them.
+  if (Math.abs(north) > 90 || Math.abs(south) > 90) return null;
+  if (Math.abs(west) > 180 || Math.abs(east) > 180) return null;
+  return {
+    southwest: { lat: Math.min(north, south), lng: Math.min(west, east) },
+    northeast: { lat: Math.max(north, south), lng: Math.max(west, east) },
+  };
+}
+
+/**
+ * Human label for a Photon feature — its name plus the coarsest containing
+ * places that are not already in the name, mirroring the single-line address
+ * Google's `formatted_address` provides.
+ * @param {object} properties - Photon feature `properties`.
+ * @returns {string}
+ */
+export function photonResultLabel(properties) {
+  const name = String(properties?.name || '').trim();
+  const parts = [name];
+  for (const field of ['district', 'city', 'state', 'country']) {
+    const part = String(properties?.[field] || '').trim();
+    if (part && !parts.includes(part)) parts.push(part);
+  }
+  return parts.filter(Boolean).join(', ');
+}
+
+/**
+ * Photon GeoJSON feature → the normalized geocode result its callers consume.
+ * Returns null for a feature without usable coordinates.
+ *
+ * `name` is carried alongside the composed `label` because two call sites want
+ * different halves of the same answer: the search box shows the full label,
+ * while the annotation resolver matches OSM features on the canonical name
+ * alone. Deriving it here keeps one shape rather than two parsers of the same
+ * string.
+ *
+ * `properties.country` is deliberately NOT re-exported. Photon answers in the
+ * feature's own language ("Việt Nam", "日本", "Россия"), which no consumer in
+ * this app can match; the composed label is the only honest use for it.
+ * @param {object} feature - One entry of Photon's `features` array.
+ * @returns {?{lat:number, lng:number, name:string, label:string,
+ *   types:string[], viewport:?object}}
+ */
+export function normalizePhotonFeature(feature) {
+  if (
+    !Array.isArray(feature?.geometry?.coordinates) ||
+    feature.geometry.type !== 'Point'
+  )
+    return null;
+  const [lng, lat] = feature.geometry.coordinates;
+  if (!Number.isFinite(lat) || !Number.isFinite(lng)) return null;
+  if (Math.abs(lat) > 90 || Math.abs(lng) > 180) return null;
+
+  const properties = feature.properties || {};
+  const name = String(properties.name || '');
+  return {
+    lat,
+    lng,
+    name,
+    label: photonResultLabel(properties) || name,
+    types: photonResultTypes(properties),
+    viewport: photonExtentToBounds(properties.extent),
+  };
+}
+
+/**
+ * Photon request URL for one query.
+ *
+ * `bias` is the SAME `"swLat,swLng|neLat,neLng"` string `viewportBias()` builds
+ * for the Google call, so the two paths cannot drift apart — but it is sent as
+ * Photon's `lat`/`lon` PROXIMITY bias, not as `bbox`.
+ *
+ * That distinction is the whole correctness of this function. Google's `bounds`
+ * prefers results inside the box and still returns ones outside it; Photon's
+ * `bbox` is a hard filter. Translating the rectangle literally makes every
+ * search for somewhere off-screen return nothing — searching "Ho Guom, Ha Noi"
+ * while looking at Austin answers `features: []`. Sent as `lat`/`lon` it finds
+ * Hoan Kiem Lake, while "Sixth Street" still resolves to Austin's over Austin
+ * and to England's over London. Soft bias is the semantics Google gives, so
+ * soft bias is what the adapter must produce.
+ * @param {string} query - Free-text place name.
+ * @param {{bias?: ?string, limit?: number}} [options]
+ * @returns {string}
+ */
+export function photonSearchUrl(
+  query,
+  { bias = null, limit = PHOTON_RESULT_LIMIT } = {},
+) {
+  const url = new URL(PHOTON_ENDPOINT);
+  url.searchParams.set('q', String(query ?? ''));
+  url.searchParams.set('limit', String(limit));
+
+  const corners = String(bias ?? '').split('|');
+  if (corners.length === 2) {
+    const [swLat, swLng] = corners[0].split(',').map(Number);
+    const [neLat, neLng] = corners[1].split(',').map(Number);
+    if ([swLat, swLng, neLat, neLng].every(Number.isFinite)) {
+      url.searchParams.set('lat', String((swLat + neLat) / 2));
+      url.searchParams.set('lon', String((swLng + neLng) / 2));
+    }
+  }
+  return url.toString();
+}
+
+/**
+ * Place name reduced to comparable form: unaccented, lowercased, punctuation
+ * collapsed to single spaces. "Huế" and "Hue" have to compare equal, because a
+ * user typing either means the same city.
+ * @param {string} text
+ * @returns {string}
+ */
+export function normalizeToponym(text) {
+  return String(text ?? '')
+    .normalize('NFD')
+    .replace(/[\u0300-\u036f]/g, '')
+    .toLowerCase()
+    .replace(/[^a-z0-9]+/g, ' ')
+    .trim();
+}
+
+/**
+ * Pick the candidate whose name actually answers `target`.
+ *
+ * Matching leads rather than merely contains: "Hue" must select `Huế` and not
+ * `Nguyen Hue Road`, which contains the word but names a different thing.
+ * `allowContains` relaxes that to a substring match, used only as a second
+ * choice once no candidate leads with the name.
+ * @param {object[]} features - Photon features, in the order returned.
+ * @param {string} target - Normalized name being looked for.
+ * @param {{allowContains?: boolean}} [options]
+ * @returns {?object}
+ */
+export function selectPhotonFeature(
+  features,
+  target,
+  { allowContains = false } = {},
+) {
+  if (!target) return null;
+  const names = (Array.isArray(features) ? features : []).map((feature) => [
+    feature,
+    normalizeToponym(feature?.properties?.name),
+  ]);
+
+  for (const [feature, name] of names) {
+    if (name === target || name.startsWith(`${target} `)) return feature;
+  }
+  if (!allowContains) return null;
+  for (const [feature, name] of names) {
+    if (name.includes(target)) return feature;
+  }
+  return null;
+}
+
+/** Evict the oldest memo entries until the cache is within its cap. */
+function trimPhotonCache(cache) {
+  while (cache.size > PHOTON_CACHE_MAX) {
+    const oldest = cache.keys().next().value;
+    if (oldest === undefined) break;
+    cache.delete(oldest);
+  }
+}
+
+/**
+ * Geocode a place name without any API key, reporting whether the service
+ * ANSWERED as well as what it found.
+ *
+ * A caller that remembers negative results needs both halves. "Photon has no
+ * such place" is a verdict and may be cached; "Photon did not reply" is a
+ * network blip, and caching THAT keeps the app answering "not found" from
+ * memory long after the network came back — a whole session, for a key that
+ * would now resolve. Conflating the two is the defect this app already had to
+ * repair once in the Overpass disk cache, where refusals were persisted with
+ * normal data TTLs.
+ *
+ * Caller cancellation throws; provider failure, timeout and empty results yield a null
+ * `place`, which the caller reads exactly as it reads a Google miss.
+ * @param {string} query - Free-text place name.
+ * @param {{bias?: ?string, fetchImpl?: Function}} [options]
+ * @returns {Promise<{place: ?{lat:number, lng:number, name:string, label:string,
+ *   types:string[], viewport:?object}, answered:boolean}>}
+ */
+export async function geocodeKeylessWithOutcome(
+  query,
+  { bias = null, fetchImpl = fetch, signal, cache = photonCache } = {},
+) {
+  signal?.throwIfAborted();
+  const trimmed = String(query ?? '').trim();
+  if (!trimmed) return { place: null, answered: true };
+
+  const memoKey = `${bias ?? ''}\n${trimmed}`;
+  // Only answered outcomes are memoised, so a hit is always an answer.
+  const cached = cache.get(memoKey);
+  if (cached && cached.expires > Date.now())
+    return { place: cached.place, answered: true };
+  cache.delete(memoKey);
+
+  /** One request: the feature array Photon returned, or null if it did not answer. */
+  const ask = async (url) => {
+    try {
+      signal?.throwIfAborted();
+      const deadline = AbortSignal.timeout(PHOTON_TIMEOUT_MS);
+      const requestSignal = signal
+        ? AbortSignal.any([signal, deadline])
+        : deadline;
+      const response = await fetchImpl(url, { signal: requestSignal });
+      if (!response.ok) return null;
+      const data = await response.json();
+      requestSignal.throwIfAborted();
+      if (!Array.isArray(data?.features)) return null;
+      if (data.features.some((feature) => !normalizePhotonFeature(feature)))
+        return null;
+      return data.features;
+    } catch {
+      signal?.throwIfAborted();
+      return null;
+    }
+  };
+
+  // Bias decides WHICH match wins, never WHAT counts as a match. Photon scores
+  // proximity against the name, so from Austin "Huế" came back as Hutto, Texas
+  // and "Hạ Long" as Long Branch — near misses beating the exact name by
+  // distance alone. Measured over twelve Vietnamese place names, three resolved
+  // to the wrong continent. So a biased answer is only accepted when it leads
+  // with the name asked for; otherwise the same query runs unbiased, which is
+  // what a user searching for somewhere off-screen meant in the first place.
+  let feature = null;
+  let everyAskAnswered = true;
+  if (bias) {
+    const biased = await ask(photonSearchUrl(trimmed, { bias }));
+    if (biased === null) everyAskAnswered = false;
+    else feature = selectPhotonFeature(biased, normalizeToponym(trimmed));
+  }
+
+  signal?.throwIfAborted();
+  if (!feature) {
+    const anywhere = await ask(photonSearchUrl(trimmed, { bias: null }));
+    if (anywhere === null) everyAskAnswered = false;
+    else {
+      // "Hoan Kiem Lake, Hanoi" names a place and then the region holding it, the
+      // convention every geocoder's free-text field follows. The head segment is
+      // the thing being searched for; the tail only says where to look.
+      const head = normalizeToponym(trimmed.split(',')[0]);
+      feature =
+        selectPhotonFeature(anywhere, head, { allowContains: true }) ||
+        anywhere[0];
+    }
+  }
+
+  signal?.throwIfAborted();
+  const place = normalizePhotonFeature(feature);
+  // A found place is self-evidently an answer; a null one only counts as an
+  // answer when every request it rests on actually came back.
+  const answered = Boolean(place) || everyAskAnswered;
+
+  // Answered misses are cached too: a typo re-issued on every keystroke should
+  // cost one request, not one per stroke. Unanswered ones are not, so the next
+  // attempt retries instead of replaying a network blip.
+  if (answered) {
+    cache.set(memoKey, {
+      place,
+      expires: Date.now() + (place ? 300_000 : 30_000),
+    });
+    trimPhotonCache(cache);
+  }
+  return { place, answered };
+}
+
+/**
+ * Geocode a place name without any API key. Cancellation throws; network failure, a
+ * timeout, or an empty result all resolve to null, which the caller reads as
+ * "not found" exactly as it reads a Google miss.
+ * @param {string} query - Free-text place name.
+ * @param {{bias?: ?string, fetchImpl?: Function}} [options]
+ * @returns {Promise<?{lat:number, lng:number, name:string, label:string, types:string[],
+ *   viewport:?object}>}
+ */
+export async function geocodeKeyless(query, options) {
+  return (await geocodeKeylessWithOutcome(query, options)).place;
+}
+
+/** Construct an independent Photon provider with a bounded query cache. */
+export function createPhotonGeocoder({
+  fetchImpl = (...args) => fetch(...args),
+} = {}) {
+  const cache = new Map();
+  return {
+    geocode: (query, options = {}) =>
+      geocodeKeylessWithOutcome(query, { ...options, fetchImpl, cache }),
+  };
+}