from .common import InfoExtractor from ..utils import ( ExtractorError, int_or_none, parse_iso8601, ) class TriluliluIE(InfoExtractor): _VALID_URL = r'https?://(?:(?:www|m)\.)?trilulilu\.ro/(?:[^/]+/)?(?P[^/#\?]+)' _TESTS = [{ 'url': 'http://www.trilulilu.ro/big-buck-bunny-1', 'md5': '68da087b676a6196a413549212f60cc6', 'info_dict': { 'id': 'ae2899e124140b', 'ext': 'mp4', 'title': 'Big Buck Bunny', 'description': ':) pentru copilul din noi', 'uploader_id': 'chipy', 'upload_date': '20120304', 'timestamp': 1330830647, 'uploader': 'chipy', 'view_count': int, 'like_count': int, 'comment_count': int, }, }, { 'url': 'http://www.trilulilu.ro/adena-ft-morreti-inocenta', 'md5': '929dfb8729dc71750463af88bbbbf4a4', 'info_dict': { 'id': 'f299710e3c91c5', 'ext': 'mp4', 'title': 'Adena ft. Morreti - Inocenta', 'description': 'pop music', 'uploader_id': 'VEVOmixt', 'upload_date': '20151204', 'uploader': 'VEVOmixt', 'timestamp': 1449187937, 'view_count': int, 'like_count': int, 'comment_count': int, }, }] def _real_extract(self, url): display_id = self._match_id(url) media_info = self._download_json('http://m.trilulilu.ro/%s?format=json' % display_id, display_id) age_limit = 0 errors = media_info.get('errors', {}) if errors.get('friends'): raise ExtractorError('This video is private.', expected=True) elif errors.get('geoblock'): raise ExtractorError('This video is not available in your country.', expected=True) elif errors.get('xxx_unlogged'): age_limit = 18 media_class = media_info.get('class') if media_class not in ('video', 'audio'): raise ExtractorError('not a video or an audio') user = media_info.get('user', {}) thumbnail = media_info.get('cover_url') if thumbnail: thumbnail.format(width='1600', height='1200') # TODO: get correct ext for audio files stream_type = media_info.get('stream_type') formats = [{ 'url': media_info['href'], 'ext': stream_type, }] if media_info.get('is_hd'): formats.append({ 'format_id': 'hd', 'url': media_info['hrefhd'], 'ext': stream_type, }) if media_class == 'audio': formats[0]['vcodec'] = 'none' else: formats[0]['format_id'] = 'sd' return { 'id': media_info['identifier'].split('|')[1], 'display_id': display_id, 'formats': formats, 'title': media_info['title'], 'description': media_info.get('description'), 'thumbnail': thumbnail, 'uploader_id': user.get('username'), 'uploader': user.get('fullname'), 'timestamp': parse_iso8601(media_info.get('published'), ' '), 'duration': int_or_none(media_info.get('duration')), 'view_count': int_or_none(media_info.get('count_views')), 'like_count': int_or_none(media_info.get('count_likes')), 'comment_count': int_or_none(media_info.get('count_comments')), 'age_limit': age_limit, }