1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-25 11:41:52 +00:00
youtube-dl/youtube_dl/extractor/redbulltv.py

229 lines
9 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..compat import compat_HTTPError
from ..utils import (
float_or_none,
ExtractorError,
)
class RedBullTVIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?redbull(?:\.tv|\.com(?:/[^/]+)?(?:/tv)?)(?:/events/[^/]+)?/(?:videos?|live|(?:film|episode)s)/(?P<id>AP-\w+)'
_TESTS = [{
# film
'url': 'https://www.redbull.tv/video/AP-1Q6XCDTAN1W11',
'md5': 'fb0445b98aa4394e504b413d98031d1f',
'info_dict': {
'id': 'AP-1Q6XCDTAN1W11',
'ext': 'mp4',
'title': 'ABC of... WRC - ABC of... S1E6',
'description': 'md5:5c7ed8f4015c8492ecf64b6ab31e7d31',
'duration': 1582.04,
},
}, {
# episode
'url': 'https://www.redbull.tv/video/AP-1PMHKJFCW1W11',
'info_dict': {
'id': 'AP-1PMHKJFCW1W11',
'ext': 'mp4',
'title': 'Grime - Hashtags S2E4',
'description': 'md5:5546aa612958c08a98faaad4abce484d',
'duration': 904,
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.redbull.com/int-en/tv/video/AP-1UWHCAR9S1W11/rob-meets-sam-gaze?playlist=playlists::3f81040a-2f31-4832-8e2e-545b1d39d173',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/us-en/videos/AP-1YM9QCYE52111',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/us-en/events/AP-1XV2K61Q51W11/live/AP-1XUJ86FDH1W11',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/films/AP-1ZSMAW8FH2111',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/episodes/AP-1TQWK7XE11W11',
'only_matching': True,
}]
def extract_info(self, video_id):
session = self._download_json(
'https://api.redbull.tv/v3/session', video_id,
note='Downloading access token', query={
'category': 'personal_computer',
'os_family': 'http',
})
if session.get('code') == 'error':
raise ExtractorError('%s said: %s' % (
self.IE_NAME, session['message']))
token = session['token']
try:
video = self._download_json(
'https://api.redbull.tv/v3/products/' + video_id,
video_id, note='Downloading video information',
headers={'Authorization': token}
)
except ExtractorError as e:
if isinstance(e.cause, compat_HTTPError) and e.cause.code == 404:
error_message = self._parse_json(
e.cause.read().decode(), video_id)['error']
raise ExtractorError('%s said: %s' % (
self.IE_NAME, error_message), expected=True)
raise
title = video['title'].strip()
formats = self._extract_m3u8_formats(
'https://dms.redbull.tv/v3/%s/%s/playlist.m3u8' % (video_id, token),
video_id, 'mp4', entry_protocol='m3u8_native', m3u8_id='hls')
self._sort_formats(formats)
subtitles = {}
for resource in video.get('resources', []):
if resource.startswith('closed_caption_'):
splitted_resource = resource.split('_')
if splitted_resource[2]:
subtitles.setdefault('en', []).append({
'url': 'https://resources.redbull.tv/%s/%s' % (video_id, resource),
'ext': splitted_resource[2],
})
subheading = video.get('subheading')
if subheading:
title += ' - %s' % subheading
return {
'id': video_id,
'title': title,
'description': video.get('long_description') or video.get(
'short_description'),
'duration': float_or_none(video.get('duration'), scale=1000),
'formats': formats,
'subtitles': subtitles,
}
def _real_extract(self, url):
video_id = self._match_id(url)
return self.extract_info(video_id)
class RedBullEmbedIE(RedBullTVIE):
_VALID_URL = r'https?://(?:www\.)?redbull\.com/embed/(?P<id>rrn:content:[^:]+:[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12}:[a-z]{2}-[A-Z]{2,3})'
_TESTS = [{
# HLS manifest accessible only using assetId
'url': 'https://www.redbull.com/embed/rrn:content:episode-videos:f3021f4f-3ed4-51ac-915a-11987126e405:en-INT',
'only_matching': True,
}]
_VIDEO_ESSENSE_TMPL = '''... on %s {
videoEssence {
attributes
}
}'''
def _real_extract(self, url):
rrn_id = self._match_id(url)
asset_id = self._download_json(
'https://edge-graphql.crepo-production.redbullaws.com/v1/graphql',
rrn_id, headers={'API-KEY': 'e90a1ff11335423998b100c929ecc866'},
query={
'query': '''{
resource(id: "%s", enforceGeoBlocking: false) {
%s
%s
}
}''' % (rrn_id, self._VIDEO_ESSENSE_TMPL % 'LiveVideo', self._VIDEO_ESSENSE_TMPL % 'VideoResource'),
})['data']['resource']['videoEssence']['attributes']['assetId']
return self.extract_info(asset_id)
class RedBullTVRrnContentIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?redbull\.com/(?P<region>[a-z]{2,3})-(?P<lang>[a-z]{2})/tv/(?:video|live|film)/(?P<id>rrn:content:[^:]+:[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})'
_TESTS = [{
'url': 'https://www.redbull.com/int-en/tv/video/rrn:content:live-videos:e3e6feb4-e95f-50b7-962a-c70f8fd13c73/mens-dh-finals-fort-william',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/tv/video/rrn:content:videos:a36a0f36-ff1b-5db8-a69d-ee11a14bf48b/tn-ts-style?playlist=rrn:content:event-profiles:83f05926-5de8-5389-b5e4-9bb312d715e8:extras',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/tv/film/rrn:content:films:d1f4d00e-4c04-5d19-b510-a805ffa2ab83/follow-me',
'only_matching': True,
}]
def _real_extract(self, url):
region, lang, rrn_id = re.search(self._VALID_URL, url).groups()
rrn_id += ':%s-%s' % (lang, region.upper())
return self.url_result(
'https://www.redbull.com/embed/' + rrn_id,
RedBullEmbedIE.ie_key(), rrn_id)
class RedBullIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?redbull\.com/(?P<region>[a-z]{2,3})-(?P<lang>[a-z]{2})/(?P<type>(?:episode|film|(?:(?:recap|trailer)-)?video)s|live)/(?!AP-|rrn:content:)(?P<id>[^/?#&]+)'
_TESTS = [{
'url': 'https://www.redbull.com/int-en/episodes/grime-hashtags-s02-e04',
'md5': 'db8271a7200d40053a1809ed0dd574ff',
'info_dict': {
'id': 'AA-1MT8DQWA91W14',
'ext': 'mp4',
'title': 'Grime - Hashtags S2E4',
'description': 'md5:5546aa612958c08a98faaad4abce484d',
},
}, {
'url': 'https://www.redbull.com/int-en/films/kilimanjaro-mountain-of-greatness',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/recap-videos/uci-mountain-bike-world-cup-2017-mens-xco-finals-from-vallnord',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/trailer-videos/kings-of-content',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/videos/tnts-style-red-bull-dance-your-style-s1-e12',
'only_matching': True,
}, {
'url': 'https://www.redbull.com/int-en/live/mens-dh-finals-fort-william',
'only_matching': True,
}, {
# only available on the int-en website so a fallback is need for the API
# https://www.redbull.com/v3/api/graphql/v1/v3/query/en-GB>en-INT?filter[uriSlug]=fia-wrc-saturday-recap-estonia&rb3Schema=v1:hero
'url': 'https://www.redbull.com/gb-en/live/fia-wrc-saturday-recap-estonia',
'only_matching': True,
}]
_INT_FALLBACK_LIST = ['de', 'en', 'es', 'fr']
_LAT_FALLBACK_MAP = ['ar', 'bo', 'car', 'cl', 'co', 'mx', 'pe']
def _real_extract(self, url):
region, lang, filter_type, display_id = re.search(self._VALID_URL, url).groups()
if filter_type == 'episodes':
filter_type = 'episode-videos'
elif filter_type == 'live':
filter_type = 'live-videos'
regions = [region.upper()]
if region != 'int':
if region in self._LAT_FALLBACK_MAP:
regions.append('LAT')
if lang in self._INT_FALLBACK_LIST:
regions.append('INT')
locale = '>'.join(['%s-%s' % (lang, reg) for reg in regions])
rrn_id = self._download_json(
'https://www.redbull.com/v3/api/graphql/v1/v3/query/' + locale,
display_id, query={
'filter[type]': filter_type,
'filter[uriSlug]': display_id,
'rb3Schema': 'v1:hero',
})['data']['id']
return self.url_result(
'https://www.redbull.com/embed/' + rrn_id,
RedBullEmbedIE.ie_key(), rrn_id)