1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-16 23:16:51 +00:00
youtube-dl/youtube_dl/extractor/theplatform.py

295 lines
11 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
from __future__ import unicode_literals
2013-12-04 22:28:40 +00:00
import re
2015-02-14 22:56:12 +00:00
import time
import hmac
import binascii
import hashlib
2013-12-04 22:28:40 +00:00
from .common import InfoExtractor
from ..compat import (
compat_parse_qs,
compat_urllib_parse_urlparse,
)
from ..utils import (
2014-10-27 00:14:17 +00:00
determine_ext,
ExtractorError,
2013-12-04 22:28:40 +00:00
xpath_with_ns,
2015-02-14 22:56:12 +00:00
unsmuggle_url,
2015-04-08 16:29:10 +00:00
int_or_none,
2015-08-19 17:24:32 +00:00
url_basename,
float_or_none,
2013-12-04 22:28:40 +00:00
)
default_ns = 'http://www.w3.org/2005/SMIL21/Language'
_x = lambda p: xpath_with_ns(p, {'smil': default_ns})
2013-12-04 22:28:40 +00:00
2015-08-19 17:24:32 +00:00
class ThePlatformBaseIE(InfoExtractor):
def _extract_theplatform_smil(self, smil_url, video_id, note='Downloading SMIL data'):
2015-08-19 17:24:32 +00:00
meta = self._download_xml(smil_url, video_id, note=note)
try:
error_msg = next(
n.attrib['abstract']
for n in meta.findall(_x('.//smil:ref'))
if n.attrib.get('title') == 'Geographic Restriction' or n.attrib.get('title') == 'Expired')
except StopIteration:
pass
else:
raise ExtractorError(error_msg, expected=True)
formats = self._parse_smil_formats(
meta, smil_url, video_id, namespace=default_ns,
# the parameters are from syfy.com, other sites may use others,
# they also work for nbc.com
f4m_params={'g': 'UXWGVKRWHFSP', 'hdcore': '3.0.3'},
transform_rtmp_url=lambda streamer, src: (streamer, 'mp4:' + src))
for _format in formats:
ext = determine_ext(_format['url'])
if ext == 'once':
_format['ext'] = 'mp4'
self._sort_formats(formats)
subtitles = self._parse_smil_subtitles(meta, default_ns)
return formats, subtitles
2015-08-19 17:24:32 +00:00
def get_metadata(self, path, video_id):
info_url = 'http://link.theplatform.com/s/%s?format=preview' % path
2015-08-19 19:07:04 +00:00
info = self._download_json(info_url, video_id)
2015-08-19 17:24:32 +00:00
subtitles = {}
captions = info.get('captions')
if isinstance(captions, list):
for caption in captions:
lang, src, mime = caption.get('lang', 'en'), caption.get('src'), caption.get('type')
subtitles[lang] = [{
'ext': 'srt' if mime == 'text/srt' else 'ttml',
'url': src,
}]
return {
'title': info['title'],
'subtitles': subtitles,
'description': info['description'],
'thumbnail': info['defaultThumbnailUrl'],
'duration': int_or_none(info.get('duration'), 1000),
}
class ThePlatformIE(ThePlatformBaseIE):
_VALID_URL = r'''(?x)
2015-02-14 22:56:12 +00:00
(?:https?://(?:link|player)\.theplatform\.com/[sp]/(?P<provider_id>[^/]+)/
(?:(?P<media>(?:[^/]+/)+select/media/)|(?P<config>(?:[^/\?]+/(?:swf|config)|onsite)/select/))?
|theplatform:)(?P<id>[^/\?&]+)'''
2013-12-04 22:28:40 +00:00
_TESTS = [{
2013-12-04 22:28:40 +00:00
# from http://www.metacafe.com/watch/cb-e9I_cZgTgIPd/blackberrys_big_bold_z30/
'url': 'http://link.theplatform.com/s/dJ5BDC/e9I_cZgTgIPd/meta.smil?format=smil&Tracking=true&mbr=true',
'info_dict': {
'id': 'e9I_cZgTgIPd',
'ext': 'flv',
'title': 'Blackberry\'s big, bold Z30',
'description': 'The Z30 is Blackberry\'s biggest, baddest mobile messaging device yet.',
'duration': 247,
2013-12-04 22:28:40 +00:00
},
'params': {
2013-12-04 22:28:40 +00:00
# rtmp download
'skip_download': True,
2013-12-04 22:28:40 +00:00
},
}, {
# from http://www.cnet.com/videos/tesla-model-s-a-second-step-towards-a-cleaner-motoring-future/
'url': 'http://link.theplatform.com/s/kYEXFC/22d_qsQ6MIRT',
'info_dict': {
'id': '22d_qsQ6MIRT',
'ext': 'flv',
'description': 'md5:ac330c9258c04f9d7512cf26b9595409',
'title': 'Tesla Model S: A second step towards a cleaner motoring future',
},
'params': {
# rtmp download
'skip_download': True,
}
}, {
'url': 'https://player.theplatform.com/p/D6x-PC/pulse_preview/embed/select/media/yMBg9E8KFxZD',
'info_dict': {
'id': 'yMBg9E8KFxZD',
'ext': 'mp4',
'description': 'md5:644ad9188d655b742f942bf2e06b002d',
'title': 'HIGHLIGHTS: USA bag first ever series Cup win',
}
}, {
'url': 'http://player.theplatform.com/p/NnzsPC/widget/select/media/4Y0TlYUr_ZT7',
'only_matching': True,
}, {
'url': 'http://player.theplatform.com/p/2E2eJC/nbcNewsOffsite?guid=tdy_or_siri_150701',
'md5': '734f3790fb5fc4903da391beeebc4836',
'info_dict': {
'id': 'tdy_or_siri_150701',
'ext': 'mp4',
'title': 'iPhone Siris sassy response to a math question has people talking',
'description': 'md5:a565d1deadd5086f3331d57298ec6333',
'duration': 83.0,
'thumbnail': 're:^https?://.*\.jpg$',
'timestamp': 1435752600,
'upload_date': '20150701',
'categories': ['Today/Shows/Orange Room', 'Today/Sections/Money', 'Today/Topics/Tech', "Today/Topics/Editor's picks"],
},
}]
2014-11-23 19:41:03 +00:00
2015-02-14 22:56:12 +00:00
@staticmethod
def _sign_url(url, sig_key, sig_secret, life=600, include_qs=False):
flags = '10' if include_qs else '00'
expiration_date = '%x' % (int(time.time()) + life)
def str_to_hex(str):
return binascii.b2a_hex(str.encode('ascii')).decode('ascii')
def hex_to_str(hex):
return binascii.a2b_hex(hex)
relative_path = url.split('http://link.theplatform.com/s/')[1].split('?')[0]
clear_text = hex_to_str(flags + expiration_date + str_to_hex(relative_path))
checksum = hmac.new(sig_key.encode('ascii'), clear_text, hashlib.sha1).hexdigest()
sig = flags + expiration_date + checksum + str_to_hex(sig_secret)
return '%s&sig=%s' % (url, sig)
2014-10-27 00:14:17 +00:00
def _real_extract(self, url):
2015-02-14 22:56:12 +00:00
url, smuggled_data = unsmuggle_url(url, {})
2014-10-27 00:14:17 +00:00
mobj = re.match(self._VALID_URL, url)
2015-02-14 22:56:12 +00:00
provider_id = mobj.group('provider_id')
2014-10-27 00:14:17 +00:00
video_id = mobj.group('id')
2015-02-14 22:56:12 +00:00
if not provider_id:
provider_id = 'dJ5BDC'
path = provider_id
if mobj.group('media'):
path += '/media'
path += '/' + video_id
qs_dict = compat_parse_qs(compat_urllib_parse_urlparse(url).query)
if 'guid' in qs_dict:
webpage = self._download_webpage(url, video_id)
scripts = re.findall(r'<script[^>]+src="([^"]+)"', webpage)
feed_id = None
# feed id usually locates in the last script.
# Seems there's no pattern for the interested script filename, so
# I try one by one
for script in reversed(scripts):
feed_script = self._download_webpage(script, video_id, 'Downloading feed script')
feed_id = self._search_regex(r'defaultFeedId\s*:\s*"([^"]+)"', feed_script, 'default feed id', default=None)
if feed_id is not None:
break
if feed_id is None:
raise ExtractorError('Unable to find feed id')
return self.url_result('http://feed.theplatform.com/f/%s/%s?byGuid=%s' % (
provider_id, feed_id, qs_dict['guid'][0]))
if smuggled_data.get('force_smil_url', False):
smil_url = url
elif mobj.group('config'):
2014-11-23 19:41:03 +00:00
config_url = url + '&form=json'
2014-10-27 00:14:17 +00:00
config_url = config_url.replace('swf/', 'config/')
config_url = config_url.replace('onsite/', 'onsite/config/')
config = self._download_json(config_url, video_id, 'Downloading config')
if 'releaseUrl' in config:
release_url = config['releaseUrl']
else:
release_url = 'http://link.theplatform.com/s/%s?mbr=true' % path
smil_url = release_url + '&format=SMIL&formats=MPEG4&manifest=f4m'
2014-10-27 00:14:17 +00:00
else:
smil_url = 'http://link.theplatform.com/s/%s/meta.smil?format=smil&mbr=true' % path
2015-02-14 22:56:12 +00:00
sig = smuggled_data.get('sig')
if sig:
smil_url = self._sign_url(smil_url, sig['key'], sig['secret'])
2013-12-04 22:28:40 +00:00
formats, subtitles = self._extract_theplatform_smil(smil_url, video_id)
2015-08-19 17:24:32 +00:00
ret = self.get_metadata(path, video_id)
combined_subtitles = self._merge_subtitles(ret.get('subtitles', {}), subtitles)
2015-08-19 17:24:32 +00:00
ret.update({
'id': video_id,
'formats': formats,
'subtitles': combined_subtitles,
2015-08-19 17:24:32 +00:00
})
2013-12-04 22:28:40 +00:00
2015-08-19 17:24:32 +00:00
return ret
2015-08-19 17:24:32 +00:00
class ThePlatformFeedIE(ThePlatformBaseIE):
_URL_TEMPLATE = '%s//feed.theplatform.com/f/%s/%s?form=json&byGuid=%s'
_VALID_URL = r'https?://feed\.theplatform\.com/f/(?P<provider_id>[^/]+)/(?P<feed_id>[^?/]+)\?(?:[^&]+&)*byGuid=(?P<id>[a-zA-Z0-9_]+)'
_TEST = {
# From http://player.theplatform.com/p/7wvmTC/MSNBCEmbeddedOffSite?guid=n_hardball_5biden_140207
'url': 'http://feed.theplatform.com/f/7wvmTC/msnbc_video-p-test?form=json&pretty=true&range=-40&byGuid=n_hardball_5biden_140207',
'md5': '22d2b84f058d3586efcd99e57d59d314',
'info_dict': {
'id': 'n_hardball_5biden_140207',
'ext': 'mp4',
'title': 'The Biden factor: will Joe run in 2016?',
'description': 'Could Vice President Joe Biden be preparing a 2016 campaign? Mark Halperin and Sam Stein weigh in.',
'thumbnail': 're:^https?://.*\.jpg$',
'upload_date': '20140208',
'timestamp': 1391824260,
'duration': 467.0,
'categories': ['MSNBC/Issues/Democrats', 'MSNBC/Issues/Elections/Election 2016'],
},
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
provider_id = mobj.group('provider_id')
feed_id = mobj.group('feed_id')
real_url = self._URL_TEMPLATE % (self.http_scheme(), provider_id, feed_id, video_id)
feed = self._download_json(real_url, video_id)
entry = feed['entries'][0]
formats = []
subtitles = {}
2015-08-19 17:24:32 +00:00
first_video_id = None
duration = None
for item in entry['media$content']:
smil_url = item['plfile$url'] + '&format=SMIL&Tracking=true&Embedded=true&formats=MPEG4,F4M'
cur_video_id = url_basename(smil_url)
if first_video_id is None:
first_video_id = cur_video_id
duration = float_or_none(item.get('plfile$duration'))
cur_formats, cur_subtitles = self._extract_theplatform_smil(smil_url, video_id, 'Downloading SMIL data for %s' % cur_video_id)
formats.extend(cur_formats)
subtitles = self._merge_subtitles(subtitles, cur_subtitles)
self._sort_formats(formats)
2013-12-04 22:28:40 +00:00
2015-08-19 17:24:32 +00:00
thumbnails = [{
'url': thumbnail['plfile$url'],
'width': int_or_none(thumbnail.get('plfile$width')),
'height': int_or_none(thumbnail.get('plfile$height')),
} for thumbnail in entry.get('media$thumbnails', [])]
timestamp = int_or_none(entry.get('media$availableDate'), scale=1000)
categories = [item['media$name'] for item in entry.get('media$categories', [])]
ret = self.get_metadata('%s/%s' % (provider_id, first_video_id), video_id)
subtitles = self._merge_subtitles(subtitles, ret['subtitles'])
2015-08-19 17:24:32 +00:00
ret.update({
2013-12-04 22:28:40 +00:00
'id': video_id,
'formats': formats,
'subtitles': subtitles,
2015-08-19 17:24:32 +00:00
'thumbnails': thumbnails,
'duration': duration,
'timestamp': timestamp,
'categories': categories,
})
return ret