1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-12-21 04:57:46 +01:00
youtube-dl/youtube_dl/extractor/livestream.py
2019-01-01 23:56:05 +07:00

367 lines
13 KiB
Python

from __future__ import unicode_literals
import re
import itertools
from .common import InfoExtractor
from ..compat import (
compat_str,
compat_urlparse,
)
from ..utils import (
find_xpath_attr,
xpath_attr,
xpath_with_ns,
xpath_text,
orderedSet,
update_url_query,
int_or_none,
float_or_none,
parse_iso8601,
determine_ext,
)
class LivestreamIE(InfoExtractor):
IE_NAME = 'livestream'
_VALID_URL = r'https?://(?:new\.)?livestream\.com/(?:accounts/(?P<account_id>\d+)|(?P<account_name>[^/]+))/(?:events/(?P<event_id>\d+)|(?P<event_name>[^/]+))(?:/videos/(?P<id>\d+))?'
_TESTS = [{
'url': 'http://new.livestream.com/CoheedandCambria/WebsterHall/videos/4719370',
'md5': '53274c76ba7754fb0e8d072716f2292b',
'info_dict': {
'id': '4719370',
'ext': 'mp4',
'title': 'Live from Webster Hall NYC',
'timestamp': 1350008072,
'upload_date': '20121012',
'duration': 5968.0,
'like_count': int,
'view_count': int,
'thumbnail': r're:^http://.*\.jpg$'
}
}, {
'url': 'http://new.livestream.com/tedx/cityenglish',
'info_dict': {
'title': 'TEDCity2.0 (English)',
'id': '2245590',
},
'playlist_mincount': 4,
}, {
'url': 'http://new.livestream.com/chess24/tatasteelchess',
'info_dict': {
'title': 'Tata Steel Chess',
'id': '3705884',
},
'playlist_mincount': 60,
}, {
'url': 'https://new.livestream.com/accounts/362/events/3557232/videos/67864563/player?autoPlay=false&height=360&mute=false&width=640',
'only_matching': True,
}, {
'url': 'http://livestream.com/bsww/concacafbeachsoccercampeonato2015',
'only_matching': True,
}]
_API_URL_TEMPLATE = 'http://livestream.com/api/accounts/%s/events/%s'
def _parse_smil_formats(self, smil, smil_url, video_id, namespace=None, f4m_params=None, transform_rtmp_url=None):
base_ele = find_xpath_attr(
smil, self._xpath_ns('.//meta', namespace), 'name', 'httpBase')
base = base_ele.get('content') if base_ele is not None else 'http://livestreamvod-f.akamaihd.net/'
formats = []
video_nodes = smil.findall(self._xpath_ns('.//video', namespace))
for vn in video_nodes:
tbr = int_or_none(vn.attrib.get('system-bitrate'), 1000)
furl = (
update_url_query(compat_urlparse.urljoin(base, vn.attrib['src']), {
'v': '3.0.3',
'fp': 'WIN% 14,0,0,145',
}))
if 'clipBegin' in vn.attrib:
furl += '&ssek=' + vn.attrib['clipBegin']
formats.append({
'url': furl,
'format_id': 'smil_%d' % tbr,
'ext': 'flv',
'tbr': tbr,
'preference': -1000,
})
return formats
def _extract_video_info(self, video_data):
video_id = compat_str(video_data['id'])
FORMAT_KEYS = (
('sd', 'progressive_url'),
('hd', 'progressive_url_hd'),
)
formats = []
for format_id, key in FORMAT_KEYS:
video_url = video_data.get(key)
if video_url:
ext = determine_ext(video_url)
if ext == 'm3u8':
continue
bitrate = int_or_none(self._search_regex(
r'(\d+)\.%s' % ext, video_url, 'bitrate', default=None))
formats.append({
'url': video_url,
'format_id': format_id,
'tbr': bitrate,
'ext': ext,
})
smil_url = video_data.get('smil_url')
if smil_url:
formats.extend(self._extract_smil_formats(smil_url, video_id, fatal=False))
m3u8_url = video_data.get('m3u8_url')
if m3u8_url:
formats.extend(self._extract_m3u8_formats(
m3u8_url, video_id, 'mp4', 'm3u8_native',
m3u8_id='hls', fatal=False))
f4m_url = video_data.get('f4m_url')
if f4m_url:
formats.extend(self._extract_f4m_formats(
f4m_url, video_id, f4m_id='hds', fatal=False))
self._sort_formats(formats)
comments = [{
'author_id': comment.get('author_id'),
'author': comment.get('author', {}).get('full_name'),
'id': comment.get('id'),
'text': comment['text'],
'timestamp': parse_iso8601(comment.get('created_at')),
} for comment in video_data.get('comments', {}).get('data', [])]
return {
'id': video_id,
'formats': formats,
'title': video_data['caption'],
'description': video_data.get('description'),
'thumbnail': video_data.get('thumbnail_url'),
'duration': float_or_none(video_data.get('duration'), 1000),
'timestamp': parse_iso8601(video_data.get('publish_at')),
'like_count': video_data.get('likes', {}).get('total'),
'comment_count': video_data.get('comments', {}).get('total'),
'view_count': video_data.get('views'),
'comments': comments,
}
def _extract_stream_info(self, stream_info):
broadcast_id = compat_str(stream_info['broadcast_id'])
is_live = stream_info.get('is_live')
formats = []
smil_url = stream_info.get('play_url')
if smil_url:
formats.extend(self._extract_smil_formats(smil_url, broadcast_id))
m3u8_url = stream_info.get('m3u8_url')
if m3u8_url:
formats.extend(self._extract_m3u8_formats(
m3u8_url, broadcast_id, 'mp4', 'm3u8_native',
m3u8_id='hls', fatal=False))
rtsp_url = stream_info.get('rtsp_url')
if rtsp_url:
formats.append({
'url': rtsp_url,
'format_id': 'rtsp',
})
self._sort_formats(formats)
return {
'id': broadcast_id,
'formats': formats,
'title': self._live_title(stream_info['stream_title']) if is_live else stream_info['stream_title'],
'thumbnail': stream_info.get('thumbnail_url'),
'is_live': is_live,
}
def _extract_event(self, event_data):
event_id = compat_str(event_data['id'])
account_id = compat_str(event_data['owner_account_id'])
feed_root_url = self._API_URL_TEMPLATE % (account_id, event_id) + '/feed.json'
stream_info = event_data.get('stream_info')
if stream_info:
return self._extract_stream_info(stream_info)
last_video = None
entries = []
for i in itertools.count(1):
if last_video is None:
info_url = feed_root_url
else:
info_url = '{root}?&id={id}&newer=-1&type=video'.format(
root=feed_root_url, id=last_video)
videos_info = self._download_json(
info_url, event_id, 'Downloading page {0}'.format(i))['data']
videos_info = [v['data'] for v in videos_info if v['type'] == 'video']
if not videos_info:
break
for v in videos_info:
v_id = compat_str(v['id'])
entries.append(self.url_result(
'http://livestream.com/accounts/%s/events/%s/videos/%s' % (account_id, event_id, v_id),
'Livestream', v_id, v.get('caption')))
last_video = videos_info[-1]['id']
return self.playlist_result(entries, event_id, event_data['full_name'])
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
event = mobj.group('event_id') or mobj.group('event_name')
account = mobj.group('account_id') or mobj.group('account_name')
api_url = self._API_URL_TEMPLATE % (account, event)
if video_id:
video_data = self._download_json(
api_url + '/videos/%s' % video_id, video_id)
return self._extract_video_info(video_data)
else:
event_data = self._download_json(api_url, video_id)
return self._extract_event(event_data)
# The original version of Livestream uses a different system
class LivestreamOriginalIE(InfoExtractor):
IE_NAME = 'livestream:original'
_VALID_URL = r'''(?x)https?://original\.livestream\.com/
(?P<user>[^/\?#]+)(?:/(?P<type>video|folder)
(?:(?:\?.*?Id=|/)(?P<id>.*?)(&|$))?)?
'''
_TESTS = [{
'url': 'http://original.livestream.com/dealbook/video?clipId=pla_8aa4a3f1-ba15-46a4-893b-902210e138fb',
'info_dict': {
'id': 'pla_8aa4a3f1-ba15-46a4-893b-902210e138fb',
'ext': 'mp4',
'title': 'Spark 1 (BitCoin) with Cameron Winklevoss & Tyler Winklevoss of Winklevoss Capital',
'duration': 771.301,
'view_count': int,
},
}, {
'url': 'https://original.livestream.com/newplay/folder?dirId=a07bf706-d0e4-4e75-a747-b021d84f2fd3',
'info_dict': {
'id': 'a07bf706-d0e4-4e75-a747-b021d84f2fd3',
},
'playlist_mincount': 4,
}, {
# live stream
'url': 'http://original.livestream.com/znsbahamas',
'only_matching': True,
}]
def _extract_video_info(self, user, video_id):
api_url = 'http://x%sx.api.channel.livestream.com/2.0/clipdetails?extendedInfo=true&id=%s' % (user, video_id)
info = self._download_xml(api_url, video_id)
item = info.find('channel').find('item')
title = xpath_text(item, 'title')
media_ns = {'media': 'http://search.yahoo.com/mrss'}
thumbnail_url = xpath_attr(
item, xpath_with_ns('media:thumbnail', media_ns), 'url')
duration = float_or_none(xpath_attr(
item, xpath_with_ns('media:content', media_ns), 'duration'))
ls_ns = {'ls': 'http://api.channel.livestream.com/2.0'}
view_count = int_or_none(xpath_text(
item, xpath_with_ns('ls:viewsCount', ls_ns)))
return {
'id': video_id,
'title': title,
'thumbnail': thumbnail_url,
'duration': duration,
'view_count': view_count,
}
def _extract_video_formats(self, video_data, video_id):
formats = []
progressive_url = video_data.get('progressiveUrl')
if progressive_url:
formats.append({
'url': progressive_url,
'format_id': 'http',
})
m3u8_url = video_data.get('httpUrl')
if m3u8_url:
formats.extend(self._extract_m3u8_formats(
m3u8_url, video_id, 'mp4', 'm3u8_native',
m3u8_id='hls', fatal=False))
rtsp_url = video_data.get('rtspUrl')
if rtsp_url:
formats.append({
'url': rtsp_url,
'format_id': 'rtsp',
})
self._sort_formats(formats)
return formats
def _extract_folder(self, url, folder_id):
webpage = self._download_webpage(url, folder_id)
paths = orderedSet(re.findall(
r'''(?x)(?:
<li\s+class="folder">\s*<a\s+href="|
<a\s+href="(?=https?://livestre\.am/)
)([^"]+)"''', webpage))
entries = [{
'_type': 'url',
'url': compat_urlparse.urljoin(url, p),
} for p in paths]
return self.playlist_result(entries, folder_id)
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
user = mobj.group('user')
url_type = mobj.group('type')
content_id = mobj.group('id')
if url_type == 'folder':
return self._extract_folder(url, content_id)
else:
# this url is used on mobile devices
stream_url = 'http://x%sx.api.channel.livestream.com/3.0/getstream.json' % user
info = {}
if content_id:
stream_url += '?id=%s' % content_id
info = self._extract_video_info(user, content_id)
else:
content_id = user
webpage = self._download_webpage(url, content_id)
info = {
'title': self._og_search_title(webpage),
'description': self._og_search_description(webpage),
'thumbnail': self._search_regex(r'channelLogo\.src\s*=\s*"([^"]+)"', webpage, 'thumbnail', None),
}
video_data = self._download_json(stream_url, content_id)
is_live = video_data.get('isLive')
info.update({
'id': content_id,
'title': self._live_title(info['title']) if is_live else info['title'],
'formats': self._extract_video_formats(video_data, content_id),
'is_live': is_live,
})
return info
# The server doesn't support HEAD request, the generic extractor can't detect
# the redirection
class LivestreamShortenerIE(InfoExtractor):
IE_NAME = 'livestream:shortener'
IE_DESC = False # Do not list
_VALID_URL = r'https?://livestre\.am/(?P<id>.+)'
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
id = mobj.group('id')
webpage = self._download_webpage(url, id)
return self.url_result(self._og_search_url(webpage))