diff options
Diffstat (limited to 'youtube_dl/extractor/yapfiles.py')
-rw-r--r-- | youtube_dl/extractor/yapfiles.py | 101 |
1 files changed, 0 insertions, 101 deletions
diff --git a/youtube_dl/extractor/yapfiles.py b/youtube_dl/extractor/yapfiles.py deleted file mode 100644 index cfb368de9..000000000 --- a/youtube_dl/extractor/yapfiles.py +++ /dev/null @@ -1,101 +0,0 @@ -# coding: utf-8 -from __future__ import unicode_literals - -import re - -from .common import InfoExtractor -from ..utils import ( - ExtractorError, - int_or_none, - qualities, - unescapeHTML, - url_or_none, -) - - -class YapFilesIE(InfoExtractor): - _YAPFILES_URL = r'//(?:(?:www|api)\.)?yapfiles\.ru/get_player/*\?.*?\bv=(?P<id>\w+)' - _VALID_URL = r'https?:%s' % _YAPFILES_URL - _TESTS = [{ - # with hd - 'url': 'http://www.yapfiles.ru/get_player/?v=vMDE1NjcyNDUt0413', - 'md5': '2db19e2bfa2450568868548a1aa1956c', - 'info_dict': { - 'id': 'vMDE1NjcyNDUt0413', - 'ext': 'mp4', - 'title': 'Самый худший пароль WIFI', - 'thumbnail': r're:^https?://.*\.jpg$', - 'duration': 72, - }, - }, { - # without hd - 'url': 'https://api.yapfiles.ru/get_player/?uid=video_player_1872528&plroll=1&adv=1&v=vMDE4NzI1Mjgt690b', - 'only_matching': True, - }] - - @staticmethod - def _extract_urls(webpage): - return [unescapeHTML(mobj.group('url')) for mobj in re.finditer( - r'<iframe\b[^>]+\bsrc=(["\'])(?P<url>(?:https?:)?%s.*?)\1' - % YapFilesIE._YAPFILES_URL, webpage)] - - def _real_extract(self, url): - video_id = self._match_id(url) - - webpage = self._download_webpage(url, video_id, fatal=False) - - player_url = None - query = {} - if webpage: - player_url = self._search_regex( - r'player\.init\s*\(\s*(["\'])(?P<url>(?:(?!\1).)+)\1', webpage, - 'player url', default=None, group='url') - - if not player_url: - player_url = 'http://api.yapfiles.ru/load/%s/' % video_id - query = { - 'md5': 'ded5f369be61b8ae5f88e2eeb2f3caff', - 'type': 'json', - 'ref': url, - } - - player = self._download_json( - player_url, video_id, query=query)['player'] - - playlist_url = player['playlist'] - title = player['title'] - thumbnail = player.get('poster') - - if title == 'Ролик удален' or 'deleted.jpg' in (thumbnail or ''): - raise ExtractorError( - 'Video %s has been removed' % video_id, expected=True) - - playlist = self._download_json( - playlist_url, video_id)['player']['main'] - - hd_height = int_or_none(player.get('hd')) - - QUALITIES = ('sd', 'hd') - quality_key = qualities(QUALITIES) - formats = [] - for format_id in QUALITIES: - is_hd = format_id == 'hd' - format_url = url_or_none(playlist.get( - 'file%s' % ('_hd' if is_hd else ''))) - if not format_url: - continue - formats.append({ - 'url': format_url, - 'format_id': format_id, - 'quality': quality_key(format_id), - 'height': hd_height if is_hd else None, - }) - self._sort_formats(formats) - - return { - 'id': video_id, - 'title': title, - 'thumbnail': thumbnail, - 'duration': int_or_none(player.get('length')), - 'formats': formats, - } |