1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-16 17:15:59 +00:00
youtube-dl/youtube_dl/extractor/funk.py

174 lines
6.3 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import itertools
import re
from .common import InfoExtractor
from .nexx import NexxIE
from ..compat import compat_str
from ..utils import (
int_or_none,
try_get,
)
class FunkBaseIE(InfoExtractor):
_HEADERS = {
'Accept': '*/*',
'Accept-Language': 'en-US,en;q=0.9,ru;q=0.8',
'authorization': 'eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJjbGllbnROYW1lIjoid2ViYXBwLXYzMSIsInNjb3BlIjoic3RhdGljLWNvbnRlbnQtYXBpLGN1cmF0aW9uLWFwaSxuZXh4LWNvbnRlbnQtYXBpLXYzMSx3ZWJhcHAtYXBpIn0.mbuG9wS9Yf5q6PqgR4fiaRFIagiHk9JhwoKES7ksVX4',
}
_AUTH = 'eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJjbGllbnROYW1lIjoid2ViYXBwLXYzMSIsInNjb3BlIjoic3RhdGljLWNvbnRlbnQtYXBpLGN1cmF0aW9uLWFwaSxuZXh4LWNvbnRlbnQtYXBpLXYzMSx3ZWJhcHAtYXBpIn0.mbuG9wS9Yf5q6PqgR4fiaRFIagiHk9JhwoKES7ksVX4'
@staticmethod
def _make_headers(referer):
headers = FunkBaseIE._HEADERS.copy()
headers['Referer'] = referer
return headers
def _make_url_result(self, video):
return {
'_type': 'url_transparent',
'url': 'nexx:741:%s' % video['sourceId'],
'ie_key': NexxIE.ie_key(),
'id': video['sourceId'],
'title': video.get('title'),
'description': video.get('description'),
'duration': int_or_none(video.get('duration')),
'season_number': int_or_none(video.get('seasonNr')),
'episode_number': int_or_none(video.get('episodeNr')),
}
class FunkMixIE(FunkBaseIE):
_VALID_URL = r'https?://(?:www\.)?funk\.net/mix/(?P<id>[^/]+)/(?P<alias>[^/?#&]+)'
_TESTS = [{
'url': 'https://www.funk.net/mix/59d65d935f8b160001828b5b/die-realste-kifferdoku-aller-zeiten',
'md5': '8edf617c2f2b7c9847dfda313f199009',
'info_dict': {
'id': '123748',
'ext': 'mp4',
'title': '"Die realste Kifferdoku aller Zeiten"',
'description': 'md5:c97160f5bafa8d47ec8e2e461012aa9d',
'timestamp': 1490274721,
'upload_date': '20170323',
},
}]
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
mix_id = mobj.group('id')
alias = mobj.group('alias')
lists = self._download_json(
'https://www.funk.net/api/v3.1/curation/curatedLists/',
mix_id, headers=self._make_headers(url), query={
'size': 100,
})['_embedded']['curatedListList']
metas = next(
l for l in lists
if mix_id in (l.get('entityId'), l.get('alias')))['videoMetas']
video = next(
meta['videoDataDelegate']
for meta in metas
if try_get(
meta, lambda x: x['videoDataDelegate']['alias'],
compat_str) == alias)
return self._make_url_result(video)
class FunkChannelIE(FunkBaseIE):
_VALID_URL = r'https?://(?:www\.)?funk\.net/channel/(?P<id>[^/]+)/(?P<alias>[^/?#&]+)'
_TESTS = [{
'url': 'https://www.funk.net/channel/ba/die-lustigsten-instrumente-aus-dem-internet-teil-2',
'info_dict': {
'id': '1155821',
'ext': 'mp4',
'title': 'Die LUSTIGSTEN INSTRUMENTE aus dem Internet - Teil 2',
'description': 'md5:a691d0413ef4835588c5b03ded670c1f',
'timestamp': 1514507395,
'upload_date': '20171229',
},
'params': {
'skip_download': True,
},
}, {
# only available via byIdList API
'url': 'https://www.funk.net/channel/informr/martin-sonneborn-erklaert-die-eu',
'info_dict': {
'id': '205067',
'ext': 'mp4',
'title': 'Martin Sonneborn erklärt die EU',
'description': 'md5:050f74626e4ed87edf4626d2024210c0',
'timestamp': 1494424042,
'upload_date': '20170510',
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.funk.net/channel/59d5149841dca100012511e3/mein-erster-job-lovemilla-folge-1/lovemilla/',
'only_matching': True,
}]
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
channel_id = mobj.group('id')
alias = mobj.group('alias')
headers = self._make_headers(url)
video = None
# Id-based channels are currently broken on their side: webplayer
# tries to process them via byChannelAlias endpoint and fails
# predictably.
for page_num in itertools.count():
by_channel_alias = self._download_json(
'https://www.funk.net/api/v3.1/webapp/videos/byChannelAlias/%s'
% channel_id,
'Downloading byChannelAlias JSON page %d' % (page_num + 1),
headers=headers, query={
'filterFsk': 'false',
'sort': 'creationDate,desc',
'size': 100,
'page': page_num,
}, fatal=False)
if not by_channel_alias:
break
video_list = try_get(
by_channel_alias, lambda x: x['_embedded']['videoList'], list)
if not video_list:
break
try:
video = next(r for r in video_list if r.get('alias') == alias)
break
except StopIteration:
pass
if not try_get(
by_channel_alias, lambda x: x['_links']['next']):
break
if not video:
by_id_list = self._download_json(
'https://www.funk.net/api/v3.0/content/videos/byIdList',
channel_id, 'Downloading byIdList JSON', headers=headers,
query={
'ids': alias,
}, fatal=False)
if by_id_list:
video = try_get(by_id_list, lambda x: x['result'][0], dict)
if not video:
results = self._download_json(
'https://www.funk.net/api/v3.0/content/videos/filter',
channel_id, 'Downloading filter JSON', headers=headers, query={
'channelId': channel_id,
'size': 100,
})['result']
video = next(r for r in results if r.get('alias') == alias)
return self._make_url_result(video)