mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-21 17:51:51 +00:00
[ustream] Support HLS streams (closes #10698)
This commit is contained in:
parent
0a439c5c4c
commit
e3d6bdc8fc
2 changed files with 126 additions and 1 deletions
|
@ -1,3 +1,8 @@
|
||||||
|
version <unreleased>
|
||||||
|
|
||||||
|
Extractors
|
||||||
|
+ [ustream] Support the new HLS streams (#10698)
|
||||||
|
|
||||||
version 2016.09.19
|
version 2016.09.19
|
||||||
|
|
||||||
Extractors
|
Extractors
|
||||||
|
|
|
@ -1,15 +1,20 @@
|
||||||
from __future__ import unicode_literals
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
import random
|
||||||
import re
|
import re
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..compat import (
|
from ..compat import (
|
||||||
|
compat_str,
|
||||||
compat_urlparse,
|
compat_urlparse,
|
||||||
)
|
)
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
|
encode_data_uri,
|
||||||
ExtractorError,
|
ExtractorError,
|
||||||
int_or_none,
|
int_or_none,
|
||||||
float_or_none,
|
float_or_none,
|
||||||
|
mimetype2ext,
|
||||||
|
str_or_none,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|
||||||
|
@ -47,8 +52,108 @@ class UstreamIE(InfoExtractor):
|
||||||
'id': '10299409',
|
'id': '10299409',
|
||||||
},
|
},
|
||||||
'playlist_count': 3,
|
'playlist_count': 3,
|
||||||
|
}, {
|
||||||
|
'url': 'http://www.ustream.tv/recorded/91343263',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '91343263',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'GitHub Universe - General Session - Day 1',
|
||||||
|
'upload_date': '20160914',
|
||||||
|
'description': 'GitHub Universe - General Session - Day 1',
|
||||||
|
'timestamp': 1473872730,
|
||||||
|
'uploader': 'wa0dnskeqkr',
|
||||||
|
'uploader_id': '38977840',
|
||||||
|
},
|
||||||
|
'params': {
|
||||||
|
'skip_download': True, # m3u8 download
|
||||||
|
},
|
||||||
}]
|
}]
|
||||||
|
|
||||||
|
def _get_stream_info(self, url, video_id, app_id_ver, extra_note=None):
|
||||||
|
def num_to_hex(n):
|
||||||
|
return hex(n)[2:]
|
||||||
|
|
||||||
|
rnd = random.randrange
|
||||||
|
|
||||||
|
if not extra_note:
|
||||||
|
extra_note = ''
|
||||||
|
|
||||||
|
conn_info = self._download_json(
|
||||||
|
'http://r%d-1-%s-recorded-lp-live.ums.ustream.tv/1/ustream' % (rnd(1e8), video_id),
|
||||||
|
video_id, note='Downloading connection info' + extra_note,
|
||||||
|
query={
|
||||||
|
'type': 'viewer',
|
||||||
|
'appId': app_id_ver[0],
|
||||||
|
'appVersion': app_id_ver[1],
|
||||||
|
'rsid': '%s:%s' % (num_to_hex(rnd(1e8)), num_to_hex(rnd(1e8))),
|
||||||
|
'rpin': '_rpin.%d' % rnd(1e15),
|
||||||
|
'referrer': url,
|
||||||
|
'media': video_id,
|
||||||
|
'application': 'recorded',
|
||||||
|
})
|
||||||
|
host = conn_info[0]['args'][0]['host']
|
||||||
|
connection_id = conn_info[0]['args'][0]['connectionId']
|
||||||
|
|
||||||
|
return self._download_json(
|
||||||
|
'http://%s/1/ustream?connectionId=%s' % (host, connection_id),
|
||||||
|
video_id, note='Downloading stream info' + extra_note)
|
||||||
|
|
||||||
|
def _get_streams(self, url, video_id, app_id_ver):
|
||||||
|
# Sometimes the return dict does not have 'stream'
|
||||||
|
for trial_count in range(3):
|
||||||
|
stream_info = self._get_stream_info(
|
||||||
|
url, video_id, app_id_ver,
|
||||||
|
extra_note=' (try %d)' % (trial_count + 1) if trial_count > 0 else '')
|
||||||
|
if 'stream' in stream_info[0]['args'][0]:
|
||||||
|
return stream_info[0]['args'][0]['stream']
|
||||||
|
return []
|
||||||
|
|
||||||
|
def _parse_segmented_mp4(self, dash_stream_info):
|
||||||
|
def resolve_dash_template(template, idx, chunk_hash):
|
||||||
|
return template.replace('%', compat_str(idx), 1).replace('%', chunk_hash)
|
||||||
|
|
||||||
|
formats = []
|
||||||
|
for stream in dash_stream_info['streams']:
|
||||||
|
# Use only one provider to avoid too many formats
|
||||||
|
provider = dash_stream_info['providers'][0]
|
||||||
|
fragments = [{
|
||||||
|
'url': resolve_dash_template(
|
||||||
|
provider['url'] + stream['initUrl'], 0, dash_stream_info['hashes']['0'])
|
||||||
|
}]
|
||||||
|
for idx in range(dash_stream_info['videoLength'] // dash_stream_info['chunkTime']):
|
||||||
|
fragments.append({
|
||||||
|
'url': resolve_dash_template(
|
||||||
|
provider['url'] + stream['segmentUrl'], idx,
|
||||||
|
dash_stream_info['hashes'][compat_str(idx // 10 * 10)])
|
||||||
|
})
|
||||||
|
content_type = stream['contentType']
|
||||||
|
kind = content_type.split('/')[0]
|
||||||
|
f = {
|
||||||
|
'format_id': '-'.join(filter(None, [
|
||||||
|
'dash', kind, str_or_none(stream.get('bitrate'))])),
|
||||||
|
'protocol': 'http_dash_segments',
|
||||||
|
# TODO: generate a MPD doc for external players?
|
||||||
|
'url': encode_data_uri(b'<MPD/>', 'text/xml'),
|
||||||
|
'ext': mimetype2ext(content_type),
|
||||||
|
'height': stream.get('height'),
|
||||||
|
'width': stream.get('width'),
|
||||||
|
'fragments': fragments,
|
||||||
|
}
|
||||||
|
if kind == 'video':
|
||||||
|
f.update({
|
||||||
|
'vcodec': stream.get('codec'),
|
||||||
|
'acodec': 'none',
|
||||||
|
'vbr': stream.get('bitrate'),
|
||||||
|
})
|
||||||
|
else:
|
||||||
|
f.update({
|
||||||
|
'vcodec': 'none',
|
||||||
|
'acodec': stream.get('codec'),
|
||||||
|
'abr': stream.get('bitrate'),
|
||||||
|
})
|
||||||
|
formats.append(f)
|
||||||
|
return formats
|
||||||
|
|
||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
m = re.match(self._VALID_URL, url)
|
m = re.match(self._VALID_URL, url)
|
||||||
video_id = m.group('id')
|
video_id = m.group('id')
|
||||||
|
@ -86,7 +191,22 @@ class UstreamIE(InfoExtractor):
|
||||||
'url': video_url,
|
'url': video_url,
|
||||||
'ext': format_id,
|
'ext': format_id,
|
||||||
'filesize': filesize,
|
'filesize': filesize,
|
||||||
} for format_id, video_url in video['media_urls'].items()]
|
} for format_id, video_url in video['media_urls'].items() if video_url]
|
||||||
|
|
||||||
|
if not formats:
|
||||||
|
hls_streams = self._get_streams(url, video_id, app_id_ver=(11, 2))
|
||||||
|
if hls_streams:
|
||||||
|
# m3u8_native leads to intermittent ContentTooShortError
|
||||||
|
formats.extend(self._extract_m3u8_formats(
|
||||||
|
hls_streams[0]['url'], video_id, ext='mp4', m3u8_id='hls'))
|
||||||
|
|
||||||
|
'''
|
||||||
|
# DASH streams handling is incomplete as 'url' is missing
|
||||||
|
dash_streams = self._get_streams(url, video_id, app_id_ver=(3, 1))
|
||||||
|
if dash_streams:
|
||||||
|
formats.extend(self._parse_segmented_mp4(dash_streams))
|
||||||
|
'''
|
||||||
|
|
||||||
self._sort_formats(formats)
|
self._sort_formats(formats)
|
||||||
|
|
||||||
description = video.get('description')
|
description = video.get('description')
|
||||||
|
|
Loading…
Reference in a new issue