_TESTS = [{
# movie
'url': 'https://api.nexx.cloud/v3/748/videos/byid/128907',
- 'md5': '16746bfc28c42049492385c989b26c4a',
+ 'md5': '828cea195be04e66057b846288295ba1',
'info_dict': {
'id': '128907',
'ext': 'mp4',
'timestamp': 1384264416,
'upload_date': '20131112',
},
- 'params': {
- 'format': 'bestvideo',
- },
}, {
# episode
'url': 'https://api.nexx.cloud/v3/741/videos/byid/247858',
'season_number': 2,
},
'params': {
- 'format': 'bestvideo',
'skip_download': True,
},
}, {
stream_data = video['streamdata']
language = general.get('language_raw') or ''
- # TODO: reverse more cdns and formats
+ # TODO: reverse more cdns
cdn = stream_data['cdnType']
assert cdn == 'azure'
azure_locator = stream_data['azureLocator']
- AZURE_URL = 'http://nx-p%02d.akamaized.net/'
+ AZURE_URL = 'http://nx%s%02d.akamaized.net/'
- for secure in ('s', ''):
- cdn_shield = stream_data.get('cdnShieldHTTP%s' % secure.upper())
- if cdn_shield:
- azure_base = 'http%s://%s' % (secure, cdn_shield)
- break
- else:
- azure_base = AZURE_URL % int(stream_data['azureAccount'].replace('nexxplayplus', ''))
+ def get_cdn_shield_base(shield_type='', prefix='-p'):
+ for secure in ('', 's'):
+ cdn_shield = stream_data.get('cdnShield%sHTTP%s' % (shield_type, secure.upper()))
+ if cdn_shield:
+ return 'http%s://%s' % (secure, cdn_shield)
+ else:
+ return AZURE_URL % (prefix, int(stream_data['azureAccount'].replace('nexxplayplus', '')))
+ azure_stream_base = get_cdn_shield_base()
is_ml = ',' in language
- azure_m3u8_url = '%s%s/%s_src%s.ism/Manifest(format=m3u8-aapl)' % (
- azure_base, azure_locator, video_id, ('_manifest' if is_ml else ''))
+ azure_manifest_url = '%s%s/%s_src%s.ism/Manifest' % (
+ azure_stream_base, azure_locator, video_id, ('_manifest' if is_ml else '')) + '%s'
protection_token = try_get(
video, lambda x: x['protectiondata']['token'], compat_str)
if protection_token:
- azure_m3u8_url += '?hdnts=%s' % protection_token
+ azure_manifest_url += '?hdnts=%s' % protection_token
formats = self._extract_m3u8_formats(
- azure_m3u8_url, video_id, 'mp4', entry_protocol='m3u8_native',
- m3u8_id='%s-hls' % cdn)
+ azure_manifest_url % '(format=m3u8-aapl)',
+ video_id, 'mp4', 'm3u8_native',
+ m3u8_id='%s-hls' % cdn, fatal=False)
+ formats.extend(self._extract_mpd_formats(
+ azure_manifest_url % '(format=mpd-time-csf)',
+ video_id, mpd_id='%s-dash' % cdn, fatal=False))
+ formats.extend(self._extract_ism_formats(
+ azure_manifest_url % '', video_id, ism_id='%s-mss' % cdn, fatal=False))
+
+ azure_progressive_base = get_cdn_shield_base('Prog', '-d')
+ azure_file_distribution = stream_data.get('azureFileDistribution')
+ if azure_file_distribution:
+ fds = azure_file_distribution.split(',')
+ if fds:
+ for fd in fds:
+ ss = fd.split(':')
+ if len(ss) == 2:
+ tbr = int_or_none(ss[0])
+ if tbr:
+ f = {
+ 'url': '%s%s/%s_src_%s_%d.mp4' % (
+ azure_progressive_base, azure_locator, video_id, ss[1], tbr),
+ 'format_id': '%s-http-%d' % (cdn, tbr),
+ 'tbr': tbr,
+ }
+ width_height = ss[1].split('x')
+ if len(width_height) == 2:
+ f.update({
+ 'width': int_or_none(width_height[0]),
+ 'height': int_or_none(width_height[1]),
+ })
+ formats.append(f)
+
self._sort_formats(formats)
return {