1
0
Fork 0
mirror of https://gitlab.com/dstftw/youtube-dl.git synced 2020-11-16 09:42:26 +00:00

[vesti] Add support for vesti.ru videos and live streams (Closes #2376)

This commit is contained in:
Sergey M. 2014-02-13 23:22:23 +07:00
parent 29b2aaf035
commit 1074982e6e
2 changed files with 171 additions and 0 deletions

View file

@ -230,6 +230,7 @@ from .ustream import UstreamIE, UstreamChannelIE
from .vbox7 import Vbox7IE from .vbox7 import Vbox7IE
from .veehd import VeeHDIE from .veehd import VeeHDIE
from .veoh import VeohIE from .veoh import VeohIE
from .vesti import VestiIE
from .vevo import VevoIE from .vevo import VevoIE
from .vice import ViceIE from .vice import ViceIE
from .viddler import ViddlerIE from .viddler import ViddlerIE

View file

@ -0,0 +1,170 @@
# encoding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
int_or_none
)
class VestiIE(InfoExtractor):
IE_NAME = 'vesti'
IE_DESC = 'Вести.Ru'
_VALID_URL = r'http://(?:.+?\.)?vesti\.ru/(?P<id>.+)'
_TESTS = [
{
'url': 'http://www.vesti.ru/videos?vid=575582&cid=1',
'info_dict': {
'id': '765035',
'ext': 'mp4',
'title': 'Вести.net: биткоины в России не являются законными',
'description': 'md5:d4bb3859dc1177b28a94c5014c35a36b',
'duration': 302,
},
'params': {
# m3u8 download
'skip_download': True,
},
},
{
'url': 'http://www.vesti.ru/only_video.html?vid=576180',
'info_dict': {
'id': '766048',
'ext': 'mp4',
'title': 'США заморозило, Британию затопило',
'description': 'md5:f0ed0695ec05aed27c56a70a58dc4cc1',
'duration': 87,
},
'params': {
# m3u8 download
'skip_download': True,
},
},
{
'url': 'http://sochi2014.vesti.ru/video/index/video_id/766403',
'info_dict': {
'id': '766403',
'ext': 'mp4',
'title': 'XXII зимние Олимпийские игры. Российские хоккеисты стартовали на Олимпиаде с победы',
'description': 'md5:55805dfd35763a890ff50fa9e35e31b3',
'duration': 271,
},
'params': {
# m3u8 download
'skip_download': True,
},
},
{
'url': 'http://sochi2014.vesti.ru/live/play/live_id/301',
'info_dict': {
'id': '51499',
'ext': 'flv',
'title': 'Сочи-2014. Биатлон. Индивидуальная гонка. Мужчины ',
'description': 'md5:9e0ed5c9d2fa1efbfdfed90c9a6d179c',
},
'params': {
# rtmp download
'skip_download': True,
},
'skip': 'Translation has finished'
}
]
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')
mobj = re.search(r'<meta property="og:video" content=".+?\.swf\?v?id=(?P<id>\d+).*?" />', page)
if mobj:
video_type = 'video'
video_id = mobj.group('id')
else:
mobj = re.search(
r'<div.+?id="current-video-holder".*?>\s*<iframe src="http://player\.rutv\.ru/iframe/(?P<type>[^/]+)/id/(?P<id>\d+)[^"]*"',
page)
if not mobj:
raise ExtractorError('No media found')
video_type = mobj.group('type')
video_id = mobj.group('id')
json_data = self._download_json(
'http://player.rutv.ru/iframe/%splay/id/%s' % ('live-' if video_type == 'live' else '', video_id),
video_id, 'Downloading JSON')
if json_data['errors']:
raise ExtractorError('vesti returned error: %s' % json_data['errors'], expected=True)
playlist = json_data['data']['playlist']
medialist = playlist['medialist']
media = medialist[0]
if media['errors']:
raise ExtractorError('vesti returned error: %s' % media['errors'], expected=True)
view_count = playlist.get('count_views')
priority_transport = playlist['priority_transport']
thumbnail = media['picture']
width = media['width']
height = media['height']
description = media['anons']
title = media['title']
duration = int_or_none(media.get('duration'))
formats = []
for transport, links in media['sources'].items():
for quality, url in links.items():
if transport == 'rtmp':
mobj = re.search(r'^(?P<url>rtmp://[^/]+/(?P<app>.+))/(?P<playpath>.+)$', url)
if not mobj:
continue
fmt = {
'url': mobj.group('url'),
'play_path': mobj.group('playpath'),
'app': mobj.group('app'),
'page_url': 'http://player.rutv.ru',
'player_url': 'http://player.rutv.ru/flash2v/osmf.swf?i=22',
'rtmp_live': True,
'ext': 'flv',
'vbr': int(quality),
}
elif transport == 'm3u8':
fmt = {
'url': url,
'ext': 'mp4',
}
else:
fmt = {
'url': url
}
fmt.update({
'width': width,
'height': height,
'format_id': '%s-%s' % (transport, quality),
'preference': -1 if priority_transport == transport else -2,
})
formats.append(fmt)
if not formats:
raise ExtractorError('No media links available for %s' % video_id)
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'view_count': view_count,
'duration': duration,
'formats': formats,
}