# encoding: utf-8 from __future__ import unicode_literals from .common import InfoExtractor from ..compat import compat_urlparse from ..utils import ( int_or_none, qualities, unified_strdate, ) class FirstTVIE(InfoExtractor): IE_NAME = '1tv' IE_DESC = 'Первый канал' _VALID_URL = r'https?://(?:www\.)?1tv\.ru/(?:[^/]+/)+(?P[^/?#]+)' _TESTS = [{ 'url': 'http://www.1tv.ru/shows/naedine-so-vsemi/vypuski/gost-lyudmila-senchina-naedine-so-vsemi-vypusk-ot-12-02-2015', 'md5': 'a1b6b60d530ebcf8daacf4565762bbaf', 'info_dict': { 'id': '40049', 'ext': 'mp4', 'title': 'Гость Людмила Сенчина. Наедине со всеми. Выпуск от 12.02.2015', 'description': 'md5:36a39c1d19618fec57d12efe212a8370', 'thumbnail': 're:^https?://.*\.(?:jpg|JPG)$', 'upload_date': '20150212', 'duration': 2694, }, }, { 'url': 'http://www.1tv.ru/shows/dobroe-utro/pro-zdorove/vesennyaya-allergiya-dobroe-utro-fragment-vypuska-ot-07042016', 'only_matching': 'true', }] def _real_extract(self, url): display_id = self._match_id(url) webpage = self._download_webpage(url, display_id) playlist_url = compat_urlparse.urljoin(url, self._search_regex( r'data-playlist-url="([^"]+)', webpage, 'playlist url')) item = self._download_json(playlist_url, display_id)[0] video_id = item['id'] quality = qualities(('ld', 'sd', 'hd', )) formats = [] for f in item.get('mbr', []): src = f.get('src') if not src: continue fname = f.get('name') formats.append({ 'url': src, 'format_id': fname, 'quality': quality(fname), }) self._sort_formats(formats) title = self._html_search_regex( (r'
\s*

([^<]*)', r"'title'\s*:\s*'([^']+)'"), webpage, 'title', default=None) or item['title'] description = self._html_search_regex( r'
\s*
 
\s*

([^<]*)

', webpage, 'description', default=None) or self._html_search_meta( 'description', webpage, 'description') duration = int_or_none(self._html_search_meta( 'video:duration', webpage, 'video duration', fatal=False)) upload_date = unified_strdate(self._html_search_meta( 'ya:ovs:upload_date', webpage, 'upload date', fatal=False)) return { 'id': video_id, 'thumbnail': item.get('poster') or self._og_search_thumbnail(webpage), 'title': title, 'description': description, 'upload_date': upload_date, 'duration': int_or_none(duration), 'formats': formats }