gods-eye-view/server/providers/cctv/taiwan.js

import {
  CCTV_SOURCE_FETCH_TIMEOUT_MS,
  TAIWAN_HIGHWAY_IMAGE_ORIGINS,
} from './constants.js';
import {
  cameraDisplayCode,
  fallbackHeadingFromId,
  prioritizeSources,
} from './normalize.js';
import { readResponseTextCapped } from '../common/http.js';

/** Freeway Bureau (MOTC) CCTV static table in the national MOTC 2.0 XML
 * standard, published keyless and refreshed daily (~1,800 cameras). */
export const TAIWAN_FREEWAY_CCTV_URL =
  'https://tisvcloud.freeway.gov.tw/history/motc20/CCTV.xml';
export const TAIWAN_HIGHWAY_CCTV_URL =
  'https://cctv-maintain.thb.gov.tw/opendataCCTVs.xml';
/** The whole network is ~1.2 MB. */
export const TAIWAN_MAX_CATALOG_BYTES = 6 * 1024 * 1024;
const DEFAULT_TAIWAN_MAX_SOURCES = 600;
/** After a failed download, wait this long before asking again. The catalog
 * host does not answer from some networks outside Taiwan, and every attempt
 * there holds the shared catalog refresh until the connect timeout. */
export const TAIWAN_RETRY_AFTER_FAILURE_MS = 60 * 60 * 1000;
/** The table is republished daily; a good copy stands in for a failed
 * download for up to this long. */
const TAIWAN_LAST_GOOD_MAX_AGE_MS = 24 * 60 * 60 * 1000;

// The table publishes each camera's MJPEG stream. The same camera servers
// answer a single JPEG for the same camera ID on a sibling path, which is what
// the frame proxy can relay; only these exact pairs are ever registered.
const STREAM_TO_FRAME = new Map([
  ['cctvn.freeway.gov.tw', ['/abs2mjpg/bmjpg', '/abs2mjpg/jpg']],
  ['cctvn5.freeway.gov.tw', ['/abs2mjpg/bmjpg', '/abs2mjpg/jpg']],
  ['cctvc.freeway.gov.tw', ['/abs2mjpg/bmjpg', '/abs2mjpg/jpg']],
  [
    'cctvs.freeway.gov.tw',
    ['/live-view/mjpg/video.cgi', '/live-view/jpg/image.cgi'],
  ],
]);

const DIRECTIONS = { N: 'N', S: 'S', E: 'E', W: 'W' };
const TAIWAN_HIGHWAY_CAMERA_HOSTS = new Set(
  TAIWAN_HIGHWAY_IMAGE_ORIGINS.map((origin) => new URL(origin).hostname),
);

/** Prioritization anchors, north to south, so a cap keeps the whole island
 * rather than only greater Taipei. */
const ANCHORS = [
  { lat: 25.033, lon: 121.5654 }, // Taipei
  { lat: 24.9936, lon: 121.301 }, // Taoyuan
  { lat: 24.8138, lon: 120.9675 }, // Hsinchu
  { lat: 24.7021, lon: 121.7378 }, // Yilan
  { lat: 24.1477, lon: 120.6736 }, // Taichung
  { lat: 22.9999, lon: 120.227 }, // Tainan
  { lat: 22.6273, lon: 120.3014 }, // Kaohsiung
];

/** Taiwan main island with slack for the coast; the freeway network has no
 * offshore-island cameras. */
function isLikelyTaiwanCoordinate(lat, lon) {
  return (
    Number.isFinite(lat) &&
    Number.isFinite(lon) &&
    lat >= 21.8 &&
    lat <= 25.4 &&
    lon >= 119.9 &&
    lon <= 122.1
  );
}

const XML_ENTITIES = {
  amp: '&',
  lt: '<',
  gt: '>',
  quot: '"',
  apos: "'",
};

function codePoint(value, radix) {
  const code = parseInt(value, radix);
  return Number.isInteger(code) && code > 0 && code <= 0x10ffff
    ? String.fromCodePoint(code)
    : '';
}

function decodeXml(text) {
  return String(text ?? '')
    .replace(
      /&(?:(amp|lt|gt|quot|apos)|#(\d{1,7})|#x([0-9a-fA-F]{1,6}));/g,
      (_, name, dec, hex) =>
        name ? XML_ENTITIES[name] : codePoint(dec ?? hex, dec ? 10 : 16),
    )
    .trim();
}

function tag(block, name) {
  const match = new RegExp(`<${name}>([^<]*)</${name}>`).exec(block);
  return match ? decodeXml(match[1]) : '';
}

/**
 * Parse the MOTC 2.0 `CCTVList` document into flat rows. Regex-based like the
 * Catalonia and Tarktee parsers: the schema is flat, and only the fields used
 * below are read.
 *
 * @param {string} xml
 * @returns {Array<object>}
 */
export function parseTaiwanCctvXml(xml) {
  const rows = [];
  const blockRe = /<CCTV>([\s\S]*?)<\/CCTV>/g;
  let match;
  while ((match = blockRe.exec(String(xml || ''))) !== null) {
    const block = match[1];
    rows.push({
      cctvId: tag(block, 'CCTVID'),
      streamUrl: tag(block, 'VideoStreamURL'),
      lat: Number(tag(block, 'PositionLat')),
      lon: Number(tag(block, 'PositionLon')),
      roadName: tag(block, 'RoadName'),
      direction: tag(block, 'RoadDirection'),
      sectionStart: tag(block, 'Start'),
      sectionEnd: tag(block, 'End'),
      mile: tag(block, 'LocationMile'),
    });
  }
  return rows;
}

/**
 * Published MJPEG stream URL -> the same camera's still, plus the host and
 * camera number that identify it, or null.
 *
 * @param {string} streamUrl
 * @returns {?{frameUrl: string, host: string, camera: string}}
 */
export function taiwanFrame(streamUrl) {
  let url;
  try {
    url = new URL(String(streamUrl ?? ''));
  } catch {
    return null;
  }
  if (url.protocol !== 'https:' || url.username || url.password || url.port)
    return null;
  const paths = STREAM_TO_FRAME.get(url.hostname);
  if (!paths || url.pathname !== paths[0]) return null;
  const camera = url.searchParams.get('camera') || '';
  if (!/^[A-Za-z0-9-]{1,40}$/.test(camera)) return null;
  return {
    // Reconstruct so extra query parameters cannot alter the request.
    frameUrl: `https://${url.hostname}${paths[1]}?camera=${camera}`,
    host: url.hostname.split('.')[0],
    camera,
  };
}

/**
 * One parsed Freeway Bureau row -> one catalog source, or null.
 *
 * @param {object} row - Output of parseTaiwanCctvXml.
 * @returns {?object}
 */
export function taiwanCameraToSource(row) {
  if (!row || typeof row !== 'object') return null;
  const { lat, lon } = row;
  if (!isLikelyTaiwanCoordinate(lat, lon)) return null;
  const frame = taiwanFrame(row.streamUrl);
  if (!frame) return null;
  const imageUrl = frame.frameUrl;
  // CCTVID carries Chinese place names and a few IDs repeat across distinct
  // cameras, so the stable key is the camera server plus its camera number.
  const rawId = String(row.cctvId ?? '')
    .trim()
    .slice(0, 80);
  const cameraId = `tw-freeway-${frame.host}-${frame.camera.toLowerCase()}`;
  const road = String(row.roadName ?? '').trim();
  const direction = DIRECTIONS[String(row.direction ?? '').trim()] || '';
  const mile = String(row.mile ?? '').trim();
  const section = [row.sectionStart, row.sectionEnd]
    .map((part) => String(part ?? '').trim())
    .filter(Boolean)
    .join('→');
  // A bare carriageway letter is not a name; fall back to the published ID.
  const name =
    (road || mile || section
      ? [road, direction, mile, section].filter(Boolean).join(' ')
      : '') ||
    rawId ||
    `Freeway camera ${frame.camera}`;

  return {
    id: cameraId,
    name: name.slice(0, 160),
    // One island-wide category in the camera picker, like Finland and Norway.
    city: 'Taiwan / 臺灣',
    cityId: 'taiwan',
    provider: 'Freeway Bureau, MOTC',
    lat,
    lon,
    // The table gives the carriageway direction, not the camera's bearing,
    // so the pose uses the shared low-confidence prior.
    headingDeg: fallbackHeadingFromId(cameraId),
    headingConfidence: 'low',
    pitchDeg: -18,
    fovDeg: 44,
    rangeM: 145,
    mountHeightM: 8,
    feedType: 'image',
    url: imageUrl,
    snapshotUrl: imageUrl,
    sourceKind: 'tw-freeway-motc',
    license:
      'Freeway Bureau, MOTC open data, Open Government Data License, Taiwan (Version 1.0)',
    code: cameraDisplayCode(
      `${road} ${mile}`.trim().toUpperCase() || frame.camera,
    ),
  };
}

/**
 * One Highway Bureau row -> a source whose first MJPEG frame is served as a
 * snapshot by the existing camera-frame proxy.
 *
 * @param {object} row - Output of parseTaiwanCctvXml.
 * @returns {?object}
 */
export function taiwanHighwayCameraToSource(row) {
  if (!row || typeof row !== 'object') return null;
  const { lat, lon } = row;
  if (!isLikelyTaiwanCoordinate(lat, lon)) return null;

  let url;
  try {
    url = new URL(String(row.streamUrl ?? ''));
  } catch {
    return null;
  }
  if (
    url.protocol !== 'https:' ||
    url.username ||
    url.password ||
    url.port ||
    url.search ||
    url.hash ||
    !TAIWAN_HIGHWAY_CAMERA_HOSTS.has(url.hostname) ||
    !/^\/[A-Za-z0-9._+~-]{1,100}$/.test(url.pathname) ||
    url.pathname.includes('..')
  )
    return null;

  const rawId = String(row.cctvId ?? '').trim();
  if (!/^[A-Za-z0-9-]{1,80}$/.test(rawId)) return null;
  const cameraId = 'tw-thb-' + rawId.toLowerCase();
  const road = String(row.roadName ?? '').trim();
  const direction = DIRECTIONS[String(row.direction ?? '').trim()] || '';
  const mile = String(row.mile ?? '').trim();
  const section = [row.sectionStart, row.sectionEnd]
    .map((part) => String(part ?? '').trim())
    .filter(Boolean)
    .join('→');
  const name =
    (road || mile || section
      ? [road, direction, mile, section].filter(Boolean).join(' ')
      : '') || rawId;

  return {
    id: cameraId,
    name: name.slice(0, 160),
    city: 'Taiwan / 臺灣',
    cityId: 'taiwan',
    provider: 'Highway Bureau, MOTC',
    lat,
    lon,
    headingDeg: fallbackHeadingFromId(cameraId),
    headingConfidence: 'low',
    pitchDeg: -18,
    fovDeg: 44,
    rangeM: 145,
    mountHeightM: 8,
    feedType: 'mjpeg',
    url: url.href,
    snapshotUrl: url.href,
    sourceKind: 'taiwan-highway-motc',
    license: 'Highway Bureau, MOTC open data',
    code: cameraDisplayCode((road + ' ' + mile).trim().toUpperCase() || rawId),
  };
}
let lastGood = null;
let lastGoodAt = 0;
let retryAfter = 0;

/** Forget the last good table and any failure backoff (tests only). */
export function resetTaiwanSourcesCache() {
  lastGood = null;
  lastGoodAt = 0;
  retryAfter = 0;
}

function fallbackSources() {
  return lastGood && Date.now() - lastGoodAt <= TAIWAN_LAST_GOOD_MAX_AGE_MS
    ? lastGood
    : [];
}

function taiwanSourceCap() {
  const rawCap = Number(
    process.env.CCTV_TAIWAN_MAX_SOURCES || DEFAULT_TAIWAN_MAX_SOURCES,
  );
  return Number.isFinite(rawCap)
    ? Math.max(1, Math.min(2000, Math.floor(rawCap)))
    : DEFAULT_TAIWAN_MAX_SOURCES;
}

/** Fetch the public provincial-road CCTV table as a no-key fallback. */
async function loadTaiwanHighwaySources() {
  const signal = AbortSignal.timeout(CCTV_SOURCE_FETCH_TIMEOUT_MS);
  const response = await fetch(TAIWAN_HIGHWAY_CCTV_URL, {
    headers: { Accept: 'application/xml,text/xml' },
    redirect: 'manual',
    signal,
  });
  if (!response.ok) {
    await response.body?.cancel().catch(() => {});
    throw new Error(
      response.status >= 300 && response.status < 400
        ? 'Highway Bureau catalog redirected; redirects are not followed'
        : 'Highway Bureau catalog HTTP ' + response.status,
    );
  }
  const xml = await readResponseTextCapped(
    response,
    TAIWAN_MAX_CATALOG_BYTES,
    signal,
  );
  const cameras = new Map();
  for (const row of parseTaiwanCctvXml(xml)) {
    const camera = taiwanHighwayCameraToSource(row);
    if (camera && !cameras.has(camera.id)) cameras.set(camera.id, camera);
  }
  if (!cameras.size)
    throw new Error('no usable Highway Bureau cameras in the table');
  const sources = prioritizeSources(
    [...cameras.values()],
    taiwanSourceCap(),
    ANCHORS,
  );
  console.log(
    '[CCTV] Loaded Taiwan Highway Bureau fallback: ' +
      cameras.size +
      ' (using ' +
      sources.length +
      ')',
  );
  return sources;
}

/**
 * Fetch Taiwan Freeway Bureau cameras from the keyless MOTC 2.0 XML table.
 * One bounded request per shared catalog refresh. A failed download serves the
 * last good table (up to a day old) and is not retried for an hour, so a host
 * that cannot reach the Freeway Bureau does not wait on it every refresh.
 *
 * @returns {Promise<Array<object>>}
 */
export async function loadTaiwanSources() {
  if (Date.now() < retryAfter) return fallbackSources();
  try {
    const signal = AbortSignal.timeout(CCTV_SOURCE_FETCH_TIMEOUT_MS);
    const response = await fetch(
      process.env.CCTV_TAIWAN_URL || TAIWAN_FREEWAY_CCTV_URL,
      {
        headers: { Accept: 'application/xml,text/xml' },
        redirect: 'manual',
        signal,
      },
    );
    if (!response.ok) {
      await response.body?.cancel().catch(() => {});
      throw new Error(
        response.status >= 300 && response.status < 400
          ? 'catalog redirected; redirects are not followed'
          : 'HTTP ' + response.status,
      );
    }
    const xml = await readResponseTextCapped(
      response,
      TAIWAN_MAX_CATALOG_BYTES,
      signal,
    );
    const cameras = new Map();
    for (const row of parseTaiwanCctvXml(xml)) {
      const camera = taiwanCameraToSource(row);
      if (camera && !cameras.has(camera.id)) cameras.set(camera.id, camera);
    }
    if (!cameras.size) throw new Error('no usable cameras in the table');
    const sources = prioritizeSources(
      [...cameras.values()],
      taiwanSourceCap(),
      ANCHORS,
    );
    console.log(
      '[CCTV] Loaded Taiwan camera sources: ' +
        cameras.size +
        ' (using ' +
        sources.length +
        ')',
    );
    lastGood = sources;
    lastGoodAt = Date.now();
    retryAfter = 0;
    return sources;
  } catch (error) {
    retryAfter = Date.now() + TAIWAN_RETRY_AFTER_FAILURE_MS;
    const fallback = fallbackSources();
    if (fallback.length) {
      console.warn(
        '[CCTV] Taiwan Freeway camera download failed (' +
          (error?.message || error) +
          '); serving the last good ' +
          fallback.length +
          ' cameras, next attempt in 1 h',
      );
      return fallback;
    }
    try {
      const sources = await loadTaiwanHighwaySources();
      lastGood = sources;
      lastGoodAt = Date.now();
      console.warn(
        '[CCTV] Taiwan Freeway camera download failed (' +
          (error?.message || error) +
          '); using ' +
          sources.length +
          ' Highway Bureau cameras as fallback, next attempt in 1 h',
      );
      return sources;
    } catch (fallbackError) {
      console.warn(
        '[CCTV] Taiwan Freeway camera download failed (' +
          (error?.message || error) +
          '); Highway Bureau fallback failed (' +
          (fallbackError?.message || fallbackError) +
          '); no Taiwan cameras, next attempt in 1 h',
      );
      return [];
    }
  }
}