from __future__ import unicode_literals from .common import InfoExtractor from ..utils import remove_end class CharlieRoseIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?charlierose\.com/video(?:s|/player)/(?P\d+)' _TEST = { 'url': 'https://charlierose.com/videos/27996', 'info_dict': { 'id': '27996', 'ext': 'mp4', 'title': 'Remembering Zaha Hadid', 'thumbnail': 're:^https?://.*\.jpg\?\d+', 'description': 'We revisit past conversations with Zaha Hadid, in memory of the world renowned Iraqi architect.', }, 'params': { # m3u8 download 'skip_download': True, } } _PLAYER_BASE = 'https://charlierose.com/video/player/%s' def _real_extract(self, url): video_id = self._match_id(url) webpage = self._download_webpage(self._PLAYER_BASE % video_id, video_id) title = remove_end(self._og_search_title(webpage), ' - Charlie Rose') entries = self._parse_html5_media_entries(self._PLAYER_BASE % video_id, webpage, video_id)[0] formats = entries['formats'] self._sort_formats(formats) self._remove_duplicate_formats(formats) return { 'id': video_id, 'title': title, 'formats': formats, 'thumbnail': self._og_search_thumbnail(webpage), 'description': self._og_search_description(webpage), 'subtitles': entries.get('subtitles'), }