mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-08 20:25:17 +00:00
102 lines
3.9 KiB
Python
102 lines
3.9 KiB
Python
from __future__ import unicode_literals
|
|
|
|
import re
|
|
import json
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
compat_urllib_parse_urlparse,
|
|
compat_urlparse,
|
|
xpath_with_ns,
|
|
compat_str,
|
|
)
|
|
|
|
|
|
class LivestreamIE(InfoExtractor):
|
|
IE_NAME = 'livestream'
|
|
_VALID_URL = r'http://new\.livestream\.com/.*?/(?P<event_name>.*?)(/videos/(?P<id>\d+))?/?$'
|
|
_TEST = {
|
|
'url': 'http://new.livestream.com/CoheedandCambria/WebsterHall/videos/4719370',
|
|
'md5': '53274c76ba7754fb0e8d072716f2292b',
|
|
'info_dict': {
|
|
'id': '4719370',
|
|
'ext': 'mp4',
|
|
'title': 'Live from Webster Hall NYC',
|
|
'upload_date': '20121012',
|
|
}
|
|
}
|
|
|
|
def _extract_video_info(self, video_data):
|
|
video_url = video_data.get('progressive_url_hd') or video_data.get('progressive_url')
|
|
return {
|
|
'id': compat_str(video_data['id']),
|
|
'url': video_url,
|
|
'ext': 'mp4',
|
|
'title': video_data['caption'],
|
|
'thumbnail': video_data['thumbnail_url'],
|
|
'upload_date': video_data['updated_at'].replace('-', '')[:8],
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
video_id = mobj.group('id')
|
|
event_name = mobj.group('event_name')
|
|
webpage = self._download_webpage(url, video_id or event_name)
|
|
|
|
if video_id is None:
|
|
# This is an event page:
|
|
config_json = self._search_regex(
|
|
r'window.config = ({.*?});', webpage, 'window config')
|
|
info = json.loads(config_json)['event']
|
|
videos = [self._extract_video_info(video_data['data'])
|
|
for video_data in info['feed']['data'] if video_data['type'] == 'video']
|
|
return self.playlist_result(videos, info['id'], info['full_name'])
|
|
else:
|
|
og_video = self._og_search_video_url(webpage, 'player url')
|
|
query_str = compat_urllib_parse_urlparse(og_video).query
|
|
query = compat_urlparse.parse_qs(query_str)
|
|
api_url = query['play_url'][0].replace('.smil', '')
|
|
info = json.loads(self._download_webpage(
|
|
api_url, video_id, 'Downloading video info'))
|
|
return self._extract_video_info(info)
|
|
|
|
|
|
# The original version of Livestream uses a different system
|
|
class LivestreamOriginalIE(InfoExtractor):
|
|
IE_NAME = 'livestream:original'
|
|
_VALID_URL = r'https?://www\.livestream\.com/(?P<user>[^/]+)/video\?.*?clipId=(?P<id>.*?)(&|$)'
|
|
_TEST = {
|
|
'url': 'http://www.livestream.com/dealbook/video?clipId=pla_8aa4a3f1-ba15-46a4-893b-902210e138fb',
|
|
'info_dict': {
|
|
'id': 'pla_8aa4a3f1-ba15-46a4-893b-902210e138fb',
|
|
'ext': 'flv',
|
|
'title': 'Spark 1 (BitCoin) with Cameron Winklevoss & Tyler Winklevoss of Winklevoss Capital',
|
|
},
|
|
'params': {
|
|
# rtmp
|
|
'skip_download': True,
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
video_id = mobj.group('id')
|
|
user = mobj.group('user')
|
|
api_url = 'http://x{0}x.api.channel.livestream.com/2.0/clipdetails?extendedInfo=true&id={1}'.format(user, video_id)
|
|
|
|
info = self._download_xml(api_url, video_id)
|
|
item = info.find('channel').find('item')
|
|
ns = {'media': 'http://search.yahoo.com/mrss'}
|
|
thumbnail_url = item.find(xpath_with_ns('media:thumbnail', ns)).attrib['url']
|
|
# Remove the extension and number from the path (like 1.jpg)
|
|
path = self._search_regex(r'(user-files/.+)_.*?\.jpg$', thumbnail_url, 'path')
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': item.find('title').text,
|
|
'url': 'rtmp://extondemand.livestream.com/ondemand',
|
|
'play_path': 'mp4:trans/dv15/mogulus-{0}.mp4'.format(path),
|
|
'ext': 'flv',
|
|
'thumbnail': thumbnail_url,
|
|
}
|