import re from .common import InfoExtractor from ..utils import ( filter_dict, int_or_none, parse_age_limit, smuggle_url, traverse_obj, unsmuggle_url, url_or_none, ) class CineverseBaseIE(InfoExtractor): _VALID_URL_BASE = r'https?://www\.(?P{})'.format('|'.join(map(re.escape, ( 'cineverse.com', 'asiancrush.com', 'dovechannel.com', 'screambox.com', 'midnightpulp.com', 'fandor.com', 'retrocrush.tv', )))) class CineverseIE(CineverseBaseIE): _VALID_URL = rf'{CineverseBaseIE._VALID_URL_BASE}/watch/(?P[A-Z0-9]+)' _TESTS = [{ 'url': 'https://www.asiancrush.com/watch/DMR00018919/Women-Who-Flirt', 'skip': 'geo-blocked', 'info_dict': { 'title': 'Women Who Flirt', 'ext': 'mp4', 'id': 'DMR00018919', 'modified_timestamp': 1678744575289, 'cast': ['Xun Zhou', 'Xiaoming Huang', 'Yi-Lin Sie', 'Sonia Sui', 'Quniciren'], 'duration': 5811.597, 'description': 'md5:892fd62a05611d394141e8394ace0bc6', 'age_limit': 13, }, }, { 'url': 'https://www.retrocrush.tv/watch/1000000023016/Archenemy! Crystal Bowie', 'skip': 'geo-blocked', 'info_dict': { 'title': 'Archenemy! Crystal Bowie', 'ext': 'mp4', 'id': '1000000023016', 'episode_number': 3, 'season_number': 1, 'cast': ['Nachi Nozawa', 'Yoshiko Sakakibara', 'Toshiko Fujita'], 'age_limit': 0, 'episode': 'Episode 3', 'season': 'Season 1', 'duration': 1485.067, 'description': 'Cobra meets a beautiful bounty hunter by the name of Jane Royal.', 'series': 'Space Adventure COBRA (Original Japanese)', }, }] def _real_extract(self, url): url, smuggled_data = unsmuggle_url(url, default={}) self._initialize_geo_bypass({ 'countries': smuggled_data.get('geo_countries'), }) video_id = self._match_id(url) html = self._download_webpage(url, video_id) idetails = self._search_nextjs_data(html, video_id)['props']['pageProps']['idetails'] err_code = idetails.get('err_code') if err_code == 1002: self.raise_login_required() elif err_code == 1200: self.raise_geo_restricted( 'This video is not available from your location due to geo restriction. ' 'You may be able to bypass it by using the /details/ page instead of the /watch/ page', countries=smuggled_data.get('geo_countries')) return { 'subtitles': filter_dict({ 'en': traverse_obj(idetails, (('cc_url_vtt', 'subtitle_url'), {'url': {url_or_none}})) or None, }), 'formats': self._extract_m3u8_formats(idetails['url'], video_id), **traverse_obj(idetails, { 'title': 'title', 'id': ('details', 'item_id'), 'description': ('details', 'description'), 'duration': ('duration', {lambda x: x / 1000}), 'cast': ('details', 'cast', {lambda x: x.split(', ')}), 'modified_timestamp': ('details', 'updated_by', 0, 'update_time', 'time', {int_or_none}), 'season_number': ('details', 'season', {int_or_none}), 'episode_number': ('details', 'episode', {int_or_none}), 'age_limit': ('details', 'rating_code', {parse_age_limit}), 'series': ('details', 'series_details', 'title'), }), } class CineverseDetailsIE(CineverseBaseIE): _VALID_URL = rf'{CineverseBaseIE._VALID_URL_BASE}/details/(?P[A-Z0-9]+)' _TESTS = [{ 'url': 'https://www.retrocrush.tv/details/1000000023012/Space-Adventure-COBRA-(Original-Japanese)', 'playlist_mincount': 30, 'info_dict': { 'title': 'Space Adventure COBRA (Original Japanese)', 'id': '1000000023012', }, }, { 'url': 'https://www.asiancrush.com/details/NNVG4938/Hansel-and-Gretel', 'info_dict': { 'id': 'NNVG4938', 'ext': 'mp4', 'title': 'Hansel and Gretel', 'description': 'md5:e3e4c35309c2e82aee044f972c2fb05d', 'cast': ['Jeong-myeong Cheon', 'Eun Won-jae', 'Shim Eun-gyeong', 'Ji-hee Jin', 'Hee-soon Park', 'Lydia Park', 'Kyeong-ik Kim'], 'duration': 7030.732, }, }] def _real_extract(self, url): host, series_id = self._match_valid_url(url).group('host', 'id') html = self._download_webpage(url, series_id) pageprops = self._search_nextjs_data(html, series_id)['props']['pageProps'] geo_countries = traverse_obj(pageprops, ('itemDetailsData', 'geo_country', {lambda x: x.split(', ')})) geoblocked = traverse_obj(pageprops, ( 'itemDetailsData', 'playback_err_msg')) == 'This title is not available in your location.' def item_result(item): item_url = f'https://www.{host}/watch/{item["item_id"]}/{item["title"]}' if geoblocked: item_url = smuggle_url(item_url, {'geo_countries': geo_countries}) return self.url_result(item_url, CineverseIE) season = traverse_obj(pageprops, ('seasonEpisodes', ..., 'episodes', lambda _, v: v['item_id'] and v['title'])) if season: return self.playlist_result([item_result(ep) for ep in season], playlist_id=series_id, playlist_title=traverse_obj(pageprops, ('itemDetailsData', 'title'))) return item_result(pageprops['itemDetailsData'])