from __future__ import unicode_literals from .common import InfoExtractor from .youtube import YoutubeIE class WimpIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?wimp\.com/(?P[^/]+)' _TESTS = [{ 'url': 'http://www.wimp.com/maru-is-exhausted/', 'md5': 'ee21217ffd66d058e8b16be340b74883', 'info_dict': { 'id': 'maru-is-exhausted', 'ext': 'mp4', 'title': 'Maru is exhausted.', 'description': 'md5:57e099e857c0a4ea312542b684a869b8', } }, { 'url': 'http://www.wimp.com/clowncar/', 'md5': '5c31ad862a90dc5b1f023956faec13fe', 'info_dict': { 'id': 'cG4CEr2aiSg', 'ext': 'webm', 'title': 'Basset hound clown car...incredible!', 'description': '5 of my Bassets crawled in this dog loo! www.bellinghambassets.com\n\nFor licensing/usage please contact: licensing(at)jukinmediadotcom', 'upload_date': '20140303', 'uploader': 'Gretchen Hoey', 'uploader_id': 'gretchenandjeff1', }, 'add_ie': ['Youtube'], }] def _real_extract(self, url): video_id = self._match_id(url) webpage = self._download_webpage(url, video_id) youtube_id = self._search_regex( (r"videoId\s*:\s*[\"']([0-9A-Za-z_-]{11})[\"']", r'data-id=["\']([0-9A-Za-z_-]{11})'), webpage, 'video URL', default=None) if youtube_id: return self.url_result(youtube_id, YoutubeIE.ie_key()) info_dict = self._extract_jwplayer_data( webpage, video_id, require_title=False) info_dict.update({ 'id': video_id, 'title': self._og_search_title(webpage), 'description': self._og_search_description(webpage), }) return info_dict