from .common import InfoExtractor from ..utils import ( determine_ext, float_or_none, join_nonempty, traverse_obj, url_or_none, ) class DailyWireBaseIE(InfoExtractor): _JSON_PATH = { 'episode': ('props', 'pageProps', 'episodeData', 'episode'), 'videos': ('props', 'pageProps', 'videoData', 'video'), 'podcasts': ('props', 'pageProps', 'episode'), } def _get_json(self, url): sites_type, slug = self._match_valid_url(url).group('sites_type', 'id') json_data = self._search_nextjs_data(self._download_webpage(url, slug), slug) return slug, traverse_obj(json_data, self._JSON_PATH[sites_type]) class DailyWireIE(DailyWireBaseIE): _VALID_URL = r'https?://(?:www\.)dailywire(?:\.com)/(?Pepisode|videos)/(?P[\w-]+)' _TESTS = [{ 'url': 'https://www.dailywire.com/episode/1-fauci', 'info_dict': { 'id': 'ckzsl50xnqpy30850in3v4bu7', 'ext': 'mp4', 'display_id': '1-fauci', 'title': '1. Fauci', 'description': 'md5:9df630347ef85081b7e97dd30bc22853', 'thumbnail': 'https://daily-wire-production.imgix.net/episodes/ckzsl50xnqpy30850in3v4bu7/ckzsl50xnqpy30850in3v4bu7-1648237399554.jpg', 'creator': 'Caroline Roberts', 'series_id': 'ckzplm0a097fn0826r2vc3j7h', 'series': 'China: The Enemy Within', } }, { 'url': 'https://www.dailywire.com/episode/ep-124-bill-maher', 'info_dict': { 'id': 'cl0ngbaalplc80894sfdo9edf', 'ext': 'mp3', 'display_id': 'ep-124-bill-maher', 'title': 'Ep. 124 - Bill Maher', 'thumbnail': 'https://daily-wire-production.imgix.net/episodes/cl0ngbaalplc80894sfdo9edf/cl0ngbaalplc80894sfdo9edf-1647065568518.jpg', 'creator': 'Caroline Roberts', 'description': 'md5:adb0de584bcfa9c41374999d9e324e98', 'series_id': 'cjzvep7270hp00786l9hwccob', 'series': 'The Sunday Special', } }, { 'url': 'https://www.dailywire.com/videos/the-hyperions', 'only_matching': True, }] def _real_extract(self, url): slug, episode_info = self._get_json(url) urls = traverse_obj( episode_info, (('segments', 'videoUrl'), ..., ('video', 'audio')), expected_type=url_or_none) formats, subtitles = [], {} for url in urls: if determine_ext(url) != 'm3u8': formats.append({'url': url}) continue format_, subs_ = self._extract_m3u8_formats_and_subtitles(url, slug) formats.extend(format_) self._merge_subtitles(subs_, target=subtitles) return { 'id': episode_info['id'], 'display_id': slug, 'title': traverse_obj(episode_info, 'title', 'name'), 'description': episode_info.get('description'), 'creator': join_nonempty(('createdBy', 'firstName'), ('createdBy', 'lastName'), from_dict=episode_info, delim=' '), 'duration': float_or_none(episode_info.get('duration')), 'is_live': episode_info.get('isLive'), 'thumbnail': traverse_obj(episode_info, 'thumbnail', 'image', expected_type=url_or_none), 'formats': formats, 'subtitles': subtitles, 'series_id': traverse_obj(episode_info, ('show', 'id')), 'series': traverse_obj(episode_info, ('show', 'name')), } class DailyWirePodcastIE(DailyWireBaseIE): _VALID_URL = r'https?://(?:www\.)dailywire(?:\.com)/(?Ppodcasts)/(?P[\w-]+/(?P[\w-]+))' _TESTS = [{ 'url': 'https://www.dailywire.com/podcasts/morning-wire/get-ready-for-recession-6-15-22', 'info_dict': { 'id': 'cl4f01d0w8pbe0a98ydd0cfn1', 'ext': 'm4a', 'display_id': 'get-ready-for-recession-6-15-22', 'title': 'Get Ready for Recession | 6.15.22', 'description': 'md5:c4afbadda4e1c38a4496f6d62be55634', 'thumbnail': 'https://daily-wire-production.imgix.net/podcasts/ckx4otgd71jm508699tzb6hf4-1639506575562.jpg', 'duration': 900.117667, } }] def _real_extract(self, url): slug, episode_info = self._get_json(url) audio_id = traverse_obj(episode_info, 'audioMuxPlaybackId', 'VUsAipTrBVSgzw73SpC2DAJD401TYYwEp') return { 'id': episode_info['id'], 'url': f'https://stream.media.dailywire.com/{audio_id}/audio.m4a', 'display_id': slug, 'title': episode_info.get('title'), 'duration': float_or_none(episode_info.get('duration')), 'thumbnail': episode_info.get('thumbnail'), 'description': episode_info.get('description'), }