1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-23 05:53:21 +08:00
youtube-dl/youtube_dl/extractor/blinkx.py

94 lines
3.3 KiB
Python
Raw Normal View History

2013-12-16 20:56:13 +08:00
import datetime
import json
import re
from .common import InfoExtractor
from ..utils import (
remove_start,
)
class BlinkxIE(InfoExtractor):
_VALID_URL = r'^(?:https?://(?:www\.)blinkx\.com/#?ce/|blinkx:)(?P<id>[^?]+)'
2013-12-16 20:56:13 +08:00
_IE_NAME = u'blinkx'
_TEST = {
u'url': u'http://www.blinkx.com/ce/8aQUy7GVFYgFzpKhT0oqsilwOGFRVXk3R1ZGWWdGenBLaFQwb3FzaWx3OGFRVXk3R1ZGWWdGenB',
u'file': u'8aQUy7GV.mp4',
u'md5': u'2e9a07364af40163a908edbf10bb2492',
u'info_dict': {
u"title": u"Police Car Rolls Away",
u"uploader": u"stupidvideos.com",
u"upload_date": u"20131215",
u"description": u"A police car gently rolls away from a fight. Maybe it felt weird being around a confrontation and just had to get out of there!",
u"duration": 14.886,
u"thumbnails": [{
"width": 100,
"height": 76,
"url": "http://cdn.blinkx.com/stream/b/41/StupidVideos/20131215/1873969261/1873969261_tn_0.jpg",
}],
},
}
def _real_extract(self, url):
m = re.match(self._VALID_URL, url)
video_id = m.group('id')
display_id = video_id[:8]
api_url = (u'https://apib4.blinkx.com/api.php?action=play_video&' +
u'video=%s' % video_id)
data_json = self._download_webpage(api_url, display_id)
data = json.loads(data_json)['api']['results'][0]
dt = datetime.datetime.fromtimestamp(data['pubdate_epoch'])
upload_date = dt.strftime('%Y%m%d')
duration = None
thumbnails = []
formats = []
for m in data['media']:
if m['type'] == 'jpg':
thumbnails.append({
'url': m['link'],
'width': int(m['w']),
'height': int(m['h']),
})
elif m['type'] == 'original':
duration = m['d']
elif m['type'] == 'youtube':
yt_id = m['link']
self.to_screen(u'Youtube video detected: %s' % yt_id)
return self.url_result(yt_id, 'Youtube', video_id=yt_id)
2013-12-16 20:56:13 +08:00
elif m['type'] in ('flv', 'mp4'):
vcodec = remove_start(m['vcodec'], 'ff')
acodec = remove_start(m['acodec'], 'ff')
tbr = (int(m['vbr']) + int(m['abr'])) // 1000
2013-12-16 20:56:13 +08:00
format_id = (u'%s-%sk-%s' %
(vcodec,
tbr,
2013-12-16 20:56:13 +08:00
m['w']))
formats.append({
'format_id': format_id,
'url': m['link'],
'vcodec': vcodec,
'acodec': acodec,
'abr': int(m['abr']) // 1000,
'vbr': int(m['vbr']) // 1000,
'tbr': tbr,
2013-12-16 20:56:13 +08:00
'width': int(m['w']),
'height': int(m['h']),
})
self._sort_formats(formats)
2013-12-16 20:56:13 +08:00
return {
'id': display_id,
'fullid': video_id,
'title': data['title'],
'formats': formats,
'uploader': data['channel_name'],
'upload_date': upload_date,
'description': data.get('description'),
'thumbnails': thumbnails,
'duration': duration,
}