[Minds] Add new extractor (#17934)

This commit is contained in:
Tatsh 2021-01-17 08:05:39 -05:00 committed by GitHub
parent d4564afc70
commit de026a6acd
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 169 additions and 0 deletions

View File

@ -651,6 +651,11 @@ from .microsoftvirtualacademy import (
MicrosoftVirtualAcademyIE, MicrosoftVirtualAcademyIE,
MicrosoftVirtualAcademyCourseIE, MicrosoftVirtualAcademyCourseIE,
) )
from .minds import (
MindsIE,
MindsActivityIE,
MindsChannelIE,
)
from .ministrygrid import MinistryGridIE from .ministrygrid import MinistryGridIE
from .minoto import MinotoIE from .minoto import MinotoIE
from .miomio import MioMioIE from .miomio import MioMioIE

View File

@ -0,0 +1,164 @@
# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..compat import compat_str
from ..utils import (int_or_none, sanitized_Request, str_or_none,
unified_strdate)
class MindsIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?minds\.com/media/(?P<id>[0-9]+)'
_TEST = {
'url': 'https://www.minds.com/media/100000000000086822',
'md5': '215a658184a419764852239d4970b045',
'info_dict': {
'id': '100000000000086822',
'ext': 'mp4',
'title': 'Minds intro sequence',
'thumbnail': 'https://cdn-cinemr.minds.com/cinemr_com/334128440657580032/thumbnail-00001.png',
'uploader_id': '100000000000000341',
'description': '<?xml encoding="utf-8" ?>',
'upload_date': '20130524',
'timestamp': 1369404826,
},
'params': {
'skip_download': True,
},
}
def _real_extract(self, url):
video_id = self._match_id(url)
video_api_url = 'https://www.minds.com/api/v1/media/%s' % video_id
token = self._get_cookies(url).get('XSRF-TOKEN')
headers = {
'authority': 'www.minds.com',
'referer': url,
'x-xsrf-token': token.value if token else '',
}
data = self._download_json(video_api_url, video_id, headers=headers,
query={'children': 'false'})
formats = []
owner = data.get('ownerObj', {})
transcodes = data.get('transcodes', {})
# These keys are the width so keep the highest width last
keys = sorted(transcodes.keys())
for format_id in keys:
is_numeric = re.match('^[0-9]+\.mp4', format_id)
video_url = transcodes[format_id]
info = {
'url': video_url,
'format_id': format_id,
'http_headers': headers,
}
if is_numeric:
info['width'] = int(format_id.split('.')[0])
formats.append(info)
uploader_id = str_or_none(owner.get('guid') or
data.get('owner_guid') or
owner.get('legacy_guid') or
owner.get('owner_guid'))
description = str_or_none(data.get('description'))
if description:
description = description.strip()
uploader_url = age_limit = thumbnail = None
if owner.get('username'):
uploader_url = 'https://www.minds.com/%s' % owner.get('username')
if data.get('mature') is True:
age_limit = 18
thumbnail_api_url = data.get('thumbnail_src')
if thumbnail_api_url:
req = sanitized_Request(thumbnail_api_url)
req.get_method = lambda: 'HEAD'
res = self._request_webpage(req, video_id)
if res.headers.get('content-type', '').startswith('image/'):
thumbnail = getattr(res, 'url', None)
tags = data.get('tags', '').strip()
if isinstance(tags, compat_str) and tags:
tags = [x.strip() for x in tags.split(',')]
else:
tags = None
category = data.get('category')
if isinstance(category, compat_str) and category:
category = [category]
else:
category = None
return {
'id': video_id,
'title': data['title'],
'formats': formats,
'description': description,
'license': str_or_none(data.get('license')),
'creator': str_or_none(owner.get('name') or owner.get('username')),
'release_date': unified_strdate(data.get('time_created')),
'timestamp': int_or_none(data.get('time_created')),
'uploader_id': uploader_id,
'uploader_url': uploader_url,
'view_count': int_or_none(data.get('play:count')),
'like_count': int_or_none(data.get('thumbs:up:count')),
'dislike_count': int_or_none(data.get('thumbs:down:count')),
'average_rating': int_or_none(data.get('rating')),
'age_limit': age_limit,
'categories': [str_or_none(data.get('category'))],
'tags': tags,
# As of 20181020 the API is returning `false` for this value both
# at top level and within the entity.comments:count path. The only
# other way to get this is to fetch all comments and count.
'comment_count': int_or_none(data.get('comments:count')),
'thumbnail': thumbnail,
}
class MindsActivityIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?minds\.com/newsfeed/(?P<id>[0-9]+)'
def _real_extract(self, url):
guid = self._match_id(url)
api_url = 'https://www.minds.com/api/v1/newsfeed/single/%s' % guid
token = self._get_cookies(url).get('XSRF-TOKEN')
headers = {
'authority': 'www.minds.com',
'referer': url,
'x-xsrf-token': token.value if token else '',
}
data = self._download_json(api_url, guid, headers=headers)
return self.url_result('https://www.minds.com/media/%s' % data['activity']['entity_guid'])
class MindsChannelIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?minds\.com/(?!newsfeed|media|api)(?P<id>[^/]+)'
def _real_extract(self, url):
channel_name = self._match_id(url)
api_url = 'https://www.minds.com/api/v1/channel/%s' % channel_name
token = self._get_cookies(url).get('XSRF-TOKEN')
headers = {
'authority': 'www.minds.com',
'referer': url,
'x-xsrf-token': token.value if token else '',
}
data = self._download_json(api_url, channel_name, headers=headers)
channel = data.get('channel', {})
params = {'limit': 12, 'offset': ''}
api_url = 'https://www.minds.com/api/v1/newsfeed/personal/%s' % channel['guid']
entries = []
while True:
data = self._download_json(api_url, channel['guid'],
headers=headers, query=params)
activity = data.get('activity', [])
if len(activity) == 0 or not data.get('load-next'):
break
for info in activity:
if info.get('custom_type') != 'video':
continue
entries.append(self.url_result('https://www.minds.com/media/%s' % info['entity_guid']))
params['offset'] = data['load-next']
return self.playlist_result(entries,
playlist_title='%s activity' % channel_name)