import json from .theplatform import ThePlatformIE from ..utils import ( determine_ext, parse_duration, parse_iso8601, ) class TheWeatherChannelIE(ThePlatformIE): _VALID_URL = r'https?://(?:www\.)?weather\.com(?P(?:/(?P[a-z]{2}-[A-Z]{2}))?/(?:[^/]+/)*video/(?P[^/?#]+))' _TESTS = [{ 'url': 'https://weather.com/series/great-outdoors/video/ice-climber-is-in-for-a-shock', 'md5': 'c4cbe74c9c17c5676b704b950b73dd92', 'info_dict': { 'id': 'cc82397e-cc3f-4d11-9390-a785add090e8', 'ext': 'mp4', 'title': 'Ice Climber Is In For A Shock', 'description': 'md5:55606ce1378d4c72e6545e160c9d9695', 'uploader': 'TWC - Digital (No Distro)', 'uploader_id': '6ccd5455-16bb-46f2-9c57-ff858bb9f62c', 'upload_date': '20160720', 'timestamp': 1469018835, } }, { 'url': 'https://weather.com/en-CA/international/videos/video/unidentified-object-falls-from-sky-in-india', 'only_matching': True, }] def _real_extract(self, url): asset_name, locale, display_id = self._match_valid_url(url).groups() if not locale: locale = 'en-US' video_data = list(self._download_json( 'https://weather.com/api/v1/p/redux-dal', display_id, data=json.dumps([{ 'name': 'getCMSAssetsUrlConfig', 'params': { 'language': locale.replace('-', '_'), 'query': { 'assetName': { '$in': asset_name, }, }, } }]).encode(), headers={ 'Content-Type': 'application/json', })['dal']['getCMSAssetsUrlConfig'].values())[0]['data'][0] video_id = video_data['id'] seo_meta = video_data.get('seometa', {}) title = video_data.get('title') or seo_meta['title'] urls = [] thumbnails = [] formats = [] for variant_id, variant_url in video_data.get('variants', []).items(): variant_url = variant_url.strip() if not variant_url or variant_url in urls: continue urls.append(variant_url) ext = determine_ext(variant_url) if ext == 'jpg': thumbnails.append({ 'url': variant_url, 'id': variant_id, }) elif ThePlatformIE.suitable(variant_url): tp_formats, _ = self._extract_theplatform_smil(variant_url, video_id) formats.extend(tp_formats) elif ext == 'm3u8': formats.extend(self._extract_m3u8_formats( variant_url, video_id, 'mp4', 'm3u8_native', m3u8_id=variant_id, fatal=False)) elif ext == 'f4m': formats.extend(self._extract_f4m_formats( variant_url, video_id, f4m_id=variant_id, fatal=False)) else: formats.append({ 'url': variant_url, 'format_id': variant_id, }) self._sort_formats(formats) cc_url = video_data.get('cc_url') return { 'id': video_id, 'display_id': display_id, 'title': title, 'description': video_data.get('description') or seo_meta.get('description') or seo_meta.get('og:description'), 'duration': parse_duration(video_data.get('duration')), 'uploader': video_data.get('providername'), 'uploader_id': video_data.get('providerid'), 'timestamp': parse_iso8601(video_data.get('publishdate')), 'subtitles': {locale[:2]: [{'url': cc_url}]} if cc_url else None, 'thumbnails': thumbnails, 'formats': formats, }