|
|
# encoding: utf-8 from __future__ import unicode_literals
import re
from .common import InfoExtractor from ..utils import ExtractorError
class NewstubeIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?newstube\.ru/media/(?P<id>.+)' _TEST = { 'url': 'http://www.newstube.ru/media/telekanal-cnn-peremestil-gorod-slavyansk-v-krym', 'info_dict': { 'id': '728e0ef2-e187-4012-bac0-5a081fdcb1f6', 'ext': 'flv', 'title': 'Телеканал CNN переместил город Славянск в Крым', 'description': 'md5:419a8c9f03442bc0b0a794d689360335', 'duration': 31.05, }, 'params': { # rtmp download 'skip_download': True, }, }
def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) video_id = mobj.group('id')
page = self._download_webpage(url, video_id, 'Downloading page')
video_guid = self._html_search_regex( r'<meta property="og:video" content="https?://(?:www\.)?newstube\.ru/freshplayer\.swf\?guid=(?P<guid>[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})', page, 'video GUID')
player = self._download_xml( 'http://p.newstube.ru/v2/player.asmx/GetAutoPlayInfo6?state=&url=%s&sessionId=&id=%s&placement=profile&location=n2' % (url, video_guid), video_guid, 'Downloading player XML')
def ns(s): return s.replace('/', '/%(ns)s') % {'ns': '{http://app1.newstube.ru/N2SiteWS/player.asmx}'}
error_message = player.find(ns('./ErrorMessage')) if error_message is not None: raise ExtractorError('%s returned error: %s' % (self.IE_NAME, error_message.text), expected=True)
session_id = player.find(ns('./SessionId')).text media_info = player.find(ns('./Medias/MediaInfo')) title = media_info.find(ns('./Name')).text description = self._og_search_description(page) thumbnail = media_info.find(ns('./KeyFrame')).text duration = int(media_info.find(ns('./Duration')).text) / 1000.0
formats = []
for stream_info in media_info.findall(ns('./Streams/StreamInfo')): media_location = stream_info.find(ns('./MediaLocation')) if media_location is None: continue
server = media_location.find(ns('./Server')).text app = media_location.find(ns('./App')).text media_id = stream_info.find(ns('./Id')).text quality_id = stream_info.find(ns('./QualityId')).text name = stream_info.find(ns('./Name')).text width = int(stream_info.find(ns('./Width')).text) height = int(stream_info.find(ns('./Height')).text)
formats.append({ 'url': 'rtmp://%s/%s' % (server, app), 'app': app, 'play_path': '01/%s' % video_guid.upper(), 'rtmp_conn': ['S:%s' % session_id, 'S:%s' % media_id, 'S:n2'], 'page_url': url, 'ext': 'flv', 'format_id': quality_id, 'format_note': name, 'width': width, 'height': height, })
self._sort_formats(formats)
return { 'id': video_guid, 'title': title, 'description': description, 'thumbnail': thumbnail, 'duration': duration, 'formats': formats, }
|