mirror of
https://github.com/yt-dlp/yt-dlp
synced 2024-12-26 21:59:08 +01:00
[ie/microsoftembed] Add extractors for dev materials (#9177)
Closes #7112 Authored by: c-basalt
This commit is contained in:
parent
aefede2556
commit
9200bc70c9
3 changed files with 265 additions and 194 deletions
|
@ -1119,12 +1119,15 @@ from .meipai import MeipaiIE
|
||||||
from .melonvod import MelonVODIE
|
from .melonvod import MelonVODIE
|
||||||
from .metacritic import MetacriticIE
|
from .metacritic import MetacriticIE
|
||||||
from .mgtv import MGTVIE
|
from .mgtv import MGTVIE
|
||||||
from .microsoftembed import MicrosoftEmbedIE
|
from .microsoftembed import (
|
||||||
from .microsoftstream import MicrosoftStreamIE
|
MicrosoftBuildIE,
|
||||||
from .microsoftvirtualacademy import (
|
MicrosoftEmbedIE,
|
||||||
MicrosoftVirtualAcademyCourseIE,
|
MicrosoftLearnEpisodeIE,
|
||||||
MicrosoftVirtualAcademyIE,
|
MicrosoftLearnPlaylistIE,
|
||||||
|
MicrosoftLearnSessionIE,
|
||||||
|
MicrosoftMediusIE,
|
||||||
)
|
)
|
||||||
|
from .microsoftstream import MicrosoftStreamIE
|
||||||
from .mildom import (
|
from .mildom import (
|
||||||
MildomClipIE,
|
MildomClipIE,
|
||||||
MildomIE,
|
MildomIE,
|
||||||
|
|
|
@ -1,5 +1,14 @@
|
||||||
|
import re
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..utils import int_or_none, traverse_obj, unified_timestamp
|
from ..utils import (
|
||||||
|
int_or_none,
|
||||||
|
parse_iso8601,
|
||||||
|
traverse_obj,
|
||||||
|
unified_timestamp,
|
||||||
|
url_basename,
|
||||||
|
url_or_none,
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
class MicrosoftEmbedIE(InfoExtractor):
|
class MicrosoftEmbedIE(InfoExtractor):
|
||||||
|
@ -63,3 +72,250 @@ class MicrosoftEmbedIE(InfoExtractor):
|
||||||
'subtitles': subtitles,
|
'subtitles': subtitles,
|
||||||
'thumbnails': thumbnails,
|
'thumbnails': thumbnails,
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
|
class MicrosoftMediusBaseIE(InfoExtractor):
|
||||||
|
@staticmethod
|
||||||
|
def _sub_to_dict(subtitle_list):
|
||||||
|
subtitles = {}
|
||||||
|
for sub in subtitle_list:
|
||||||
|
subtitles.setdefault(sub.pop('tag', 'und'), []).append(sub)
|
||||||
|
return subtitles
|
||||||
|
|
||||||
|
def _extract_ism(self, ism_url, video_id):
|
||||||
|
formats = self._extract_ism_formats(ism_url, video_id)
|
||||||
|
for fmt in formats:
|
||||||
|
if fmt['language'] != 'eng' and 'English' not in fmt['format_id']:
|
||||||
|
fmt['language_preference'] = -10
|
||||||
|
return formats
|
||||||
|
|
||||||
|
|
||||||
|
class MicrosoftMediusIE(MicrosoftMediusBaseIE):
|
||||||
|
_VALID_URL = r'https?://medius\.microsoft\.com/Embed/(?:Video\?id=|video-nc/|VideoDetails/)(?P<id>[\da-f-]+)'
|
||||||
|
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'https://medius.microsoft.com/Embed/video-nc/9640d86c-f513-4889-959e-5dace86e7d2b',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '9640d86c-f513-4889-959e-5dace86e7d2b',
|
||||||
|
'ext': 'ismv',
|
||||||
|
'title': 'Rapidly code, test and ship from secure cloud developer environments',
|
||||||
|
'description': 'md5:33c8e4facadc438613476eea24165f71',
|
||||||
|
'thumbnail': r're:https://mediusimg\.event\.microsoft\.com/video-\d+/thumbnail\.jpg.*',
|
||||||
|
'subtitles': 'count:30',
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
'url': 'https://medius.microsoft.com/Embed/video-nc/81215af5-c813-4dcd-aede-94f4e1a7daa3',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '81215af5-c813-4dcd-aede-94f4e1a7daa3',
|
||||||
|
'ext': 'ismv',
|
||||||
|
'title': 'Microsoft Build opening',
|
||||||
|
'description': 'md5:43455096141077a1f23144cab8cec1cb',
|
||||||
|
'thumbnail': r're:https://mediusimg\.event\.microsoft\.com/video-\d+/thumbnail\.jpg.*',
|
||||||
|
'subtitles': 'count:31',
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
'url': 'https://medius.microsoft.com/Embed/VideoDetails/78493569-9b3b-4a85-a409-ee76e789e25c',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '78493569-9b3b-4a85-a409-ee76e789e25c',
|
||||||
|
'ext': 'ismv',
|
||||||
|
'title': ' Anomaly Detection & Root cause at Edge',
|
||||||
|
'description': 'md5:f8f1ad93d7918649bfb97fa081b03b83',
|
||||||
|
'thumbnail': r're:https://mediusdownload.event.microsoft.com/asset.*\.jpg.*',
|
||||||
|
'subtitles': 'count:17',
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
'url': 'https://medius.microsoft.com/Embed/Video?id=0dc69bda-079b-4070-a7db-a8da1a06a9c7',
|
||||||
|
'only_matching': True,
|
||||||
|
}, {
|
||||||
|
'url': 'https://medius.microsoft.com/Embed/video-nc/fe823a91-959c-465b-96d4-8f4db624f72c',
|
||||||
|
'only_matching': True,
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _extract_subtitle(self, webpage, video_id):
|
||||||
|
captions = traverse_obj(
|
||||||
|
self._search_json(r'const\s+captionsConfiguration\s*=', webpage, 'captions', video_id, default=None),
|
||||||
|
('languageList', lambda _, v: url_or_none(v['src']), {
|
||||||
|
'url': 'src',
|
||||||
|
'tag': ('srclang', {str}),
|
||||||
|
'name': ('kind', {str}),
|
||||||
|
})) or [{'url': url, 'tag': url_basename(url).split('.vtt')[0].split('_')[-1]}
|
||||||
|
for url in re.findall(r'var\s+file\s+=\s+\{[^}]+\'(https://[^\']+\.vtt\?[^\']+)', webpage)]
|
||||||
|
|
||||||
|
return self._sub_to_dict(captions)
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(f'https://medius.microsoft.com/Embed/video-nc/{video_id}', video_id)
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': self._og_search_title(webpage),
|
||||||
|
'description': self._og_search_description(webpage),
|
||||||
|
'formats': self._extract_ism(
|
||||||
|
self._search_regex(r'StreamUrl\s*=\s*"([^"]+manifest)"', webpage, 'ism url'), video_id),
|
||||||
|
'thumbnail': self._og_search_thumbnail(webpage),
|
||||||
|
'subtitles': self._extract_subtitle(webpage, video_id),
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
class MicrosoftLearnPlaylistIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'https?://learn\.microsoft\.com/(?:[\w-]+/)?(?P<type>shows|events)/(?P<id>[\w-]+)/?(?:[?#]|$)'
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'https://learn.microsoft.com/en-us/shows/bash-for-beginners',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'bash-for-beginners',
|
||||||
|
'title': 'Bash for Beginners',
|
||||||
|
'description': 'md5:16a91c07222117d1e00912f0dbc02c2c',
|
||||||
|
},
|
||||||
|
'playlist_count': 20,
|
||||||
|
}, {
|
||||||
|
'url': 'https://learn.microsoft.com/en-us/events/build-2022',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'build-2022',
|
||||||
|
'title': 'Microsoft Build 2022 - Events',
|
||||||
|
'description': 'md5:c16b43848027df837b22c6fbac7648d3',
|
||||||
|
},
|
||||||
|
'playlist_count': 201,
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _entries(self, url_base, video_id):
|
||||||
|
skip = 0
|
||||||
|
while True:
|
||||||
|
playlist_info = self._download_json(url_base, video_id, f'Downloading entries {skip}', query={
|
||||||
|
'locale': 'en-us',
|
||||||
|
'$skip': skip,
|
||||||
|
})
|
||||||
|
url_paths = traverse_obj(playlist_info, ('results', ..., 'url', {str}))
|
||||||
|
for url_path in url_paths:
|
||||||
|
yield self.url_result(f'https://learn.microsoft.com/en-us{url_path}')
|
||||||
|
skip += len(url_paths)
|
||||||
|
if skip >= playlist_info.get('count', 0) or not url_paths:
|
||||||
|
break
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
playlist_id, playlist_type = self._match_valid_url(url).group('id', 'type')
|
||||||
|
webpage = self._download_webpage(url, playlist_id)
|
||||||
|
|
||||||
|
metainfo = {
|
||||||
|
'title': self._og_search_title(webpage),
|
||||||
|
'description': self._og_search_description(webpage),
|
||||||
|
}
|
||||||
|
sub_type = 'episodes' if playlist_type == 'shows' else 'sessions'
|
||||||
|
|
||||||
|
url_base = f'https://learn.microsoft.com/api/contentbrowser/search/{playlist_type}/{playlist_id}/{sub_type}'
|
||||||
|
return self.playlist_result(self._entries(url_base, playlist_id), playlist_id, **metainfo)
|
||||||
|
|
||||||
|
|
||||||
|
class MicrosoftLearnEpisodeIE(MicrosoftMediusBaseIE):
|
||||||
|
_VALID_URL = r'https?://learn\.microsoft\.com/(?:[\w-]+/)?shows/[\w-]+/(?P<id>[^?#/]+)'
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'https://learn.microsoft.com/en-us/shows/bash-for-beginners/what-is-the-difference-between-a-terminal-and-a-shell-2-of-20-bash-for-beginners/',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'd44e1a03-a0e5-45c2-9496-5c9fa08dc94c',
|
||||||
|
'ext': 'ismv',
|
||||||
|
'title': 'What is the Difference Between a Terminal and a Shell? (Part 2 of 20)',
|
||||||
|
'description': 'md5:7bbbfb593d21c2cf2babc3715ade6b88',
|
||||||
|
'timestamp': 1676339547,
|
||||||
|
'upload_date': '20230214',
|
||||||
|
'thumbnail': r're:https://learn\.microsoft\.com/video/media/.*\.png',
|
||||||
|
'subtitles': 'count:14',
|
||||||
|
},
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
|
||||||
|
entry_id = self._html_search_meta('entryId', webpage, 'entryId', fatal=True)
|
||||||
|
video_info = self._download_json(
|
||||||
|
f'https://learn.microsoft.com/api/video/public/v1/entries/{entry_id}', video_id)
|
||||||
|
return {
|
||||||
|
'id': entry_id,
|
||||||
|
'formats': self._extract_ism(video_info['publicVideo']['adaptiveVideoUrl'], video_id),
|
||||||
|
'subtitles': self._sub_to_dict(traverse_obj(video_info, (
|
||||||
|
'publicVideo', 'captions', lambda _, v: url_or_none(v['url']), {
|
||||||
|
'tag': ('language', {str}),
|
||||||
|
'url': 'url',
|
||||||
|
}))),
|
||||||
|
'title': self._og_search_title(webpage),
|
||||||
|
'description': self._og_search_description(webpage),
|
||||||
|
**traverse_obj(video_info, {
|
||||||
|
'timestamp': ('createTime', {parse_iso8601}),
|
||||||
|
'thumbnails': ('publicVideo', 'thumbnailOtherSizes', ..., {'url': {url_or_none}}),
|
||||||
|
}),
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
class MicrosoftLearnSessionIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'https?://learn\.microsoft\.com/(?:[\w-]+/)?events/[\w-]+/(?P<id>[^?#/]+)'
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'https://learn.microsoft.com/en-us/events/build-2022/ts01-rapidly-code-test-ship-from-secure-cloud-developer-environments',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '9640d86c-f513-4889-959e-5dace86e7d2b',
|
||||||
|
'ext': 'ismv',
|
||||||
|
'title': 'Rapidly code, test and ship from secure cloud developer environments - Events',
|
||||||
|
'description': 'md5:f26c1a85d41c1cffd27a0279254a25c3',
|
||||||
|
'timestamp': 1653408600,
|
||||||
|
'upload_date': '20220524',
|
||||||
|
'thumbnail': r're:https://mediusimg\.event\.microsoft\.com/video-\d+/thumbnail\.jpg.*',
|
||||||
|
},
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
|
||||||
|
metainfo = {
|
||||||
|
'title': self._og_search_title(webpage),
|
||||||
|
'description': self._og_search_description(webpage),
|
||||||
|
'timestamp': parse_iso8601(self._html_search_meta('startDate', webpage, 'startDate')),
|
||||||
|
}
|
||||||
|
|
||||||
|
return self.url_result(
|
||||||
|
self._html_search_meta('externalVideoUrl', webpage, 'videoUrl', fatal=True),
|
||||||
|
url_transparent=True, ie=MicrosoftMediusIE, **metainfo)
|
||||||
|
|
||||||
|
|
||||||
|
class MicrosoftBuildIE(InfoExtractor):
|
||||||
|
_VALID_URL = [
|
||||||
|
r'https?://build\.microsoft\.com/[\w-]+/sessions/(?P<id>[\da-f-]+)',
|
||||||
|
r'https?://build\.microsoft\.com/[\w-]+/(?P<id>sessions)/?(?:[?#]|$)',
|
||||||
|
]
|
||||||
|
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'https://build.microsoft.com/en-US/sessions/b49feb31-afcd-4217-a538-d3ca1d171198?source=sessions',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'aee55fb5-fcf9-4b38-b764-a3527cb57554',
|
||||||
|
'ext': 'ismv',
|
||||||
|
'title': 'Microsoft Build opening keynote',
|
||||||
|
'description': 'md5:d38338f336ef4b6ef9ad2a7466a76655',
|
||||||
|
'timestamp': 1716307200,
|
||||||
|
'upload_date': '20240521',
|
||||||
|
'thumbnail': r're:https://mediusimg\.event\.microsoft\.com/video-\d+/thumbnail\.jpg.*',
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
'url': 'https://build.microsoft.com/en-US/sessions',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'sessions',
|
||||||
|
},
|
||||||
|
'playlist_mincount': 418,
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
|
||||||
|
entries = [
|
||||||
|
self.url_result(
|
||||||
|
video_info['onDemand'], ie=MicrosoftMediusIE, url_transparent=True, **traverse_obj(video_info, {
|
||||||
|
'id': ('sessionId', {str}),
|
||||||
|
'title': ('title', {str}),
|
||||||
|
'description': ('description', {str}),
|
||||||
|
'timestamp': ('startDateTime', {parse_iso8601}),
|
||||||
|
}))
|
||||||
|
for video_info in self._download_json(
|
||||||
|
'https://api-v2.build.microsoft.com/api/session/all/en-US', video_id, 'Downloading video info')
|
||||||
|
]
|
||||||
|
if video_id == 'sessions':
|
||||||
|
return self.playlist_result(entries, video_id)
|
||||||
|
else:
|
||||||
|
return traverse_obj(entries, (lambda _, v: v['id'] == video_id), get_all=False)
|
||||||
|
|
|
@ -1,188 +0,0 @@
|
||||||
import re
|
|
||||||
|
|
||||||
from .common import InfoExtractor
|
|
||||||
from ..utils import (
|
|
||||||
int_or_none,
|
|
||||||
parse_duration,
|
|
||||||
smuggle_url,
|
|
||||||
unsmuggle_url,
|
|
||||||
xpath_text,
|
|
||||||
)
|
|
||||||
|
|
||||||
|
|
||||||
class MicrosoftVirtualAcademyBaseIE(InfoExtractor):
|
|
||||||
def _extract_base_url(self, course_id, display_id):
|
|
||||||
return self._download_json(
|
|
||||||
f'https://api-mlxprod.microsoft.com/services/products/anonymous/{course_id}',
|
|
||||||
display_id, 'Downloading course base URL')
|
|
||||||
|
|
||||||
def _extract_chapter_and_title(self, title):
|
|
||||||
if not title:
|
|
||||||
return None, None
|
|
||||||
m = re.search(r'(?P<chapter>\d+)\s*\|\s*(?P<title>.+)', title)
|
|
||||||
return (int(m.group('chapter')), m.group('title')) if m else (None, title)
|
|
||||||
|
|
||||||
|
|
||||||
class MicrosoftVirtualAcademyIE(MicrosoftVirtualAcademyBaseIE):
|
|
||||||
IE_NAME = 'mva'
|
|
||||||
IE_DESC = 'Microsoft Virtual Academy videos'
|
|
||||||
_VALID_URL = rf'(?:{IE_NAME}:|https?://(?:mva\.microsoft|(?:www\.)?microsoftvirtualacademy)\.com/[^/]+/training-courses/[^/?#&]+-)(?P<course_id>\d+)(?::|\?l=)(?P<id>[\da-zA-Z]+_\d+)'
|
|
||||||
|
|
||||||
_TESTS = [{
|
|
||||||
'url': 'https://mva.microsoft.com/en-US/training-courses/microsoft-azure-fundamentals-virtual-machines-11788?l=gfVXISmEB_6804984382',
|
|
||||||
'md5': '7826c44fc31678b12ad8db11f6b5abb9',
|
|
||||||
'info_dict': {
|
|
||||||
'id': 'gfVXISmEB_6804984382',
|
|
||||||
'ext': 'mp4',
|
|
||||||
'title': 'Course Introduction',
|
|
||||||
'formats': 'mincount:3',
|
|
||||||
'subtitles': {
|
|
||||||
'en': [{
|
|
||||||
'ext': 'ttml',
|
|
||||||
}],
|
|
||||||
},
|
|
||||||
},
|
|
||||||
}, {
|
|
||||||
'url': 'mva:11788:gfVXISmEB_6804984382',
|
|
||||||
'only_matching': True,
|
|
||||||
}]
|
|
||||||
|
|
||||||
def _real_extract(self, url):
|
|
||||||
url, smuggled_data = unsmuggle_url(url, {})
|
|
||||||
|
|
||||||
mobj = self._match_valid_url(url)
|
|
||||||
course_id = mobj.group('course_id')
|
|
||||||
video_id = mobj.group('id')
|
|
||||||
|
|
||||||
base_url = smuggled_data.get('base_url') or self._extract_base_url(course_id, video_id)
|
|
||||||
|
|
||||||
settings = self._download_xml(
|
|
||||||
f'{base_url}/content/content_{video_id}/videosettings.xml?v=1',
|
|
||||||
video_id, 'Downloading video settings XML')
|
|
||||||
|
|
||||||
_, title = self._extract_chapter_and_title(xpath_text(
|
|
||||||
settings, './/Title', 'title', fatal=True))
|
|
||||||
|
|
||||||
formats = []
|
|
||||||
|
|
||||||
for sources in settings.findall('.//MediaSources'):
|
|
||||||
sources_type = sources.get('videoType')
|
|
||||||
for source in sources.findall('./MediaSource'):
|
|
||||||
video_url = source.text
|
|
||||||
if not video_url or not video_url.startswith('http'):
|
|
||||||
continue
|
|
||||||
if sources_type == 'smoothstreaming':
|
|
||||||
formats.extend(self._extract_ism_formats(
|
|
||||||
video_url, video_id, 'mss', fatal=False))
|
|
||||||
continue
|
|
||||||
video_mode = source.get('videoMode')
|
|
||||||
height = int_or_none(self._search_regex(
|
|
||||||
r'^(\d+)[pP]$', video_mode or '', 'height', default=None))
|
|
||||||
codec = source.get('codec')
|
|
||||||
acodec, vcodec = [None] * 2
|
|
||||||
if codec:
|
|
||||||
codecs = codec.split(',')
|
|
||||||
if len(codecs) == 2:
|
|
||||||
acodec, vcodec = codecs
|
|
||||||
elif len(codecs) == 1:
|
|
||||||
vcodec = codecs[0]
|
|
||||||
formats.append({
|
|
||||||
'url': video_url,
|
|
||||||
'format_id': video_mode,
|
|
||||||
'height': height,
|
|
||||||
'acodec': acodec,
|
|
||||||
'vcodec': vcodec,
|
|
||||||
})
|
|
||||||
|
|
||||||
subtitles = {}
|
|
||||||
for source in settings.findall('.//MarkerResourceSource'):
|
|
||||||
subtitle_url = source.text
|
|
||||||
if not subtitle_url:
|
|
||||||
continue
|
|
||||||
subtitles.setdefault('en', []).append({
|
|
||||||
'url': f'{base_url}/{subtitle_url}',
|
|
||||||
'ext': source.get('type'),
|
|
||||||
})
|
|
||||||
|
|
||||||
return {
|
|
||||||
'id': video_id,
|
|
||||||
'title': title,
|
|
||||||
'subtitles': subtitles,
|
|
||||||
'formats': formats,
|
|
||||||
}
|
|
||||||
|
|
||||||
|
|
||||||
class MicrosoftVirtualAcademyCourseIE(MicrosoftVirtualAcademyBaseIE):
|
|
||||||
IE_NAME = 'mva:course'
|
|
||||||
IE_DESC = 'Microsoft Virtual Academy courses'
|
|
||||||
_VALID_URL = rf'(?:{IE_NAME}:|https?://(?:mva\.microsoft|(?:www\.)?microsoftvirtualacademy)\.com/[^/]+/training-courses/(?P<display_id>[^/?#&]+)-)(?P<id>\d+)'
|
|
||||||
|
|
||||||
_TESTS = [{
|
|
||||||
'url': 'https://mva.microsoft.com/en-US/training-courses/microsoft-azure-fundamentals-virtual-machines-11788',
|
|
||||||
'info_dict': {
|
|
||||||
'id': '11788',
|
|
||||||
'title': 'Microsoft Azure Fundamentals: Virtual Machines',
|
|
||||||
},
|
|
||||||
'playlist_count': 36,
|
|
||||||
}, {
|
|
||||||
# with emphasized chapters
|
|
||||||
'url': 'https://mva.microsoft.com/en-US/training-courses/developing-windows-10-games-with-construct-2-16335',
|
|
||||||
'info_dict': {
|
|
||||||
'id': '16335',
|
|
||||||
'title': 'Developing Windows 10 Games with Construct 2',
|
|
||||||
},
|
|
||||||
'playlist_count': 10,
|
|
||||||
}, {
|
|
||||||
'url': 'https://www.microsoftvirtualacademy.com/en-US/training-courses/microsoft-azure-fundamentals-virtual-machines-11788',
|
|
||||||
'only_matching': True,
|
|
||||||
}, {
|
|
||||||
'url': 'mva:course:11788',
|
|
||||||
'only_matching': True,
|
|
||||||
}]
|
|
||||||
|
|
||||||
@classmethod
|
|
||||||
def suitable(cls, url):
|
|
||||||
return False if MicrosoftVirtualAcademyIE.suitable(url) else super().suitable(url)
|
|
||||||
|
|
||||||
def _real_extract(self, url):
|
|
||||||
mobj = self._match_valid_url(url)
|
|
||||||
course_id = mobj.group('id')
|
|
||||||
display_id = mobj.group('display_id')
|
|
||||||
|
|
||||||
base_url = self._extract_base_url(course_id, display_id)
|
|
||||||
|
|
||||||
manifest = self._download_json(
|
|
||||||
f'{base_url}/imsmanifestlite.json',
|
|
||||||
display_id, 'Downloading course manifest JSON')['manifest']
|
|
||||||
|
|
||||||
organization = manifest['organizations']['organization'][0]
|
|
||||||
|
|
||||||
entries = []
|
|
||||||
for chapter in organization['item']:
|
|
||||||
chapter_number, chapter_title = self._extract_chapter_and_title(chapter.get('title'))
|
|
||||||
chapter_id = chapter.get('@identifier')
|
|
||||||
for item in chapter.get('item', []):
|
|
||||||
item_id = item.get('@identifier')
|
|
||||||
if not item_id:
|
|
||||||
continue
|
|
||||||
metadata = item.get('resource', {}).get('metadata') or {}
|
|
||||||
if metadata.get('learningresourcetype') != 'Video':
|
|
||||||
continue
|
|
||||||
_, title = self._extract_chapter_and_title(item.get('title'))
|
|
||||||
duration = parse_duration(metadata.get('duration'))
|
|
||||||
description = metadata.get('description')
|
|
||||||
entries.append({
|
|
||||||
'_type': 'url_transparent',
|
|
||||||
'url': smuggle_url(
|
|
||||||
f'mva:{course_id}:{item_id}', {'base_url': base_url}),
|
|
||||||
'title': title,
|
|
||||||
'description': description,
|
|
||||||
'duration': duration,
|
|
||||||
'chapter': chapter_title,
|
|
||||||
'chapter_number': chapter_number,
|
|
||||||
'chapter_id': chapter_id,
|
|
||||||
})
|
|
||||||
|
|
||||||
title = organization.get('title') or manifest.get('metadata', {}).get('title')
|
|
||||||
|
|
||||||
return self.playlist_result(entries, course_id, title)
|
|
Loading…
Reference in a new issue