0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-16 01:25:06 +00:00
yt-dlp/yt_dlp/extractor/cybrary.py
sepro add96eb9f8
[cleanup] Add more ruff rules (#10149)
Authored by: seproDev

Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com>
Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
2024-06-12 01:09:58 +02:00

144 lines
6.1 KiB
Python

from .common import InfoExtractor
from ..utils import (
ExtractorError,
smuggle_url,
str_or_none,
traverse_obj,
urlencode_postdata,
)
class CybraryBaseIE(InfoExtractor):
_API_KEY = 'AIzaSyCX9ru6j70PX2My1Eq6Q1zoMAhuTdXlzSw'
_ENDPOINTS = {
'course': 'https://app.cybrary.it/courses/api/catalog/browse/course/{}',
'course_enrollment': 'https://app.cybrary.it/courses/api/catalog/{}/enrollment',
'enrollment': 'https://app.cybrary.it/courses/api/enrollment/{}',
'launch': 'https://app.cybrary.it/courses/api/catalog/{}/launch',
'vimeo_oembed': 'https://vimeo.com/api/oembed.json?url=https://vimeo.com/{}',
}
_NETRC_MACHINE = 'cybrary'
_TOKEN = None
def _perform_login(self, username, password):
CybraryBaseIE._TOKEN = self._download_json(
f'https://identitytoolkit.googleapis.com/v1/accounts:signInWithPassword?key={self._API_KEY}',
None, data=urlencode_postdata({'email': username, 'password': password, 'returnSecureToken': True}),
note='Logging in')['idToken']
def _real_initialize(self):
if not self._TOKEN:
self.raise_login_required(method='password')
def _call_api(self, endpoint, item_id):
return self._download_json(
self._ENDPOINTS[endpoint].format(item_id), item_id,
note=f'Downloading {endpoint} JSON metadata',
headers={'Authorization': f'Bearer {self._TOKEN}'})
def _get_vimeo_id(self, activity_id):
launch_api = self._call_api('launch', activity_id)
if launch_api.get('url'):
return self._search_regex(r'https?://player\.vimeo\.com/video/(?P<vimeo_id>[0-9]+)', launch_api['url'], 'vimeo_id')
return traverse_obj(launch_api, ('vendor_data', 'content', ..., 'videoId'), get_all=False)
class CybraryIE(CybraryBaseIE):
_VALID_URL = r'https?://app\.cybrary\.it/immersive/(?P<enrollment>[0-9]+)/activity/(?P<id>[0-9]+)'
_TESTS = [{
'url': 'https://app.cybrary.it/immersive/12487950/activity/63102',
'md5': '9ae12d37e555cb2ed554223a71a701d0',
'info_dict': {
'id': '646609770',
'ext': 'mp4',
'title': 'Getting Started',
'thumbnail': 'https://i.vimeocdn.com/video/1301817996-76a268f0c56cff18a5cecbbdc44131eb9dda0c80eb0b3a036_1280',
'series_id': '63111',
'uploader_url': 'https://vimeo.com/user30867300',
'duration': 88,
'uploader_id': 'user30867300',
'series': 'Cybrary Orientation',
'uploader': 'Cybrary',
'chapter': 'Cybrary Orientation Series',
'chapter_id': '63110',
},
'expected_warnings': ['No authenticators for vimeo'],
}, {
'url': 'https://app.cybrary.it/immersive/12747143/activity/52686',
'md5': '62f26547dccc59c44363e2a13d4ad08d',
'info_dict': {
'id': '445638073',
'ext': 'mp4',
'title': 'Azure Virtual Network IP Addressing',
'thumbnail': 'https://i.vimeocdn.com/video/936667051-1647ace66c627d4a2382185e0dae8deb830309bfddd53f8b2367b2f91e92ed0e-d_1280',
'series_id': '52733',
'uploader_url': 'https://vimeo.com/user30867300',
'duration': 426,
'uploader_id': 'user30867300',
'series': 'AZ-500: Microsoft Azure Security Technologies',
'uploader': 'Cybrary',
'chapter': 'Implement Network Security',
'chapter_id': '52693',
},
'expected_warnings': ['No authenticators for vimeo'],
}]
def _real_extract(self, url):
activity_id, enrollment_id = self._match_valid_url(url).group('id', 'enrollment')
course = self._call_api('enrollment', enrollment_id)['content']
activity = traverse_obj(course, ('learning_modules', ..., 'activities', lambda _, v: int(activity_id) == v['id']), get_all=False)
if activity.get('type') not in ['Video Activity', 'Lesson Activity']:
raise ExtractorError('The activity is not a video', expected=True)
module = next((m for m in course.get('learning_modules') or []
if int(activity_id) in traverse_obj(m, ('activities', ..., 'id'))), None)
vimeo_id = self._get_vimeo_id(activity_id)
return {
'_type': 'url_transparent',
'series': traverse_obj(course, ('content_description', 'title')),
'series_id': str_or_none(traverse_obj(course, ('content_description', 'id'))),
'id': vimeo_id,
'chapter': module.get('title'),
'chapter_id': str_or_none(module.get('id')),
'title': activity.get('title'),
'url': smuggle_url(f'https://player.vimeo.com/video/{vimeo_id}', {'referer': 'https://api.cybrary.it'}),
}
class CybraryCourseIE(CybraryBaseIE):
_VALID_URL = r'https?://app\.cybrary\.it/browse/course/(?P<id>[\w-]+)/?(?:$|[#?])'
_TESTS = [{
'url': 'https://app.cybrary.it/browse/course/az-500-microsoft-azure-security-technologies',
'info_dict': {
'id': '898',
'title': 'AZ-500: Microsoft Azure Security Technologies',
'description': 'md5:69549d379c0fc1dec92926d4e8b6fbd4',
},
'playlist_count': 59,
}, {
'url': 'https://app.cybrary.it/browse/course/cybrary-orientation',
'info_dict': {
'id': '1245',
'title': 'Cybrary Orientation',
'description': 'md5:9e69ff66b32fe78744e0ad4babe2e88e',
},
'playlist_count': 4,
}]
def _real_extract(self, url):
course_id = self._match_id(url)
course = self._call_api('course', course_id)
enrollment_info = self._call_api('course_enrollment', course['id'])
entries = [self.url_result(
f'https://app.cybrary.it/immersive/{enrollment_info["id"]}/activity/{activity["id"]}')
for activity in traverse_obj(course, ('content_item', 'learning_modules', ..., 'activities', ...))]
return self.playlist_result(
entries,
traverse_obj(course, ('content_item', 'id'), expected_type=str_or_none),
course.get('title'), course.get('short_description'))