yt-dlp/yt_dlp/extractor/sharepoint.py

113 lines
5.4 KiB
Python

import json
import urllib.parse
from .common import InfoExtractor
from ..utils import determine_ext, int_or_none, url_or_none
from ..utils.traversal import traverse_obj
class SharePointIE(InfoExtractor):
_BASE_URL_RE = r'https?://[\w-]+\.sharepoint\.com/'
_VALID_URL = [
rf'{_BASE_URL_RE}:v:/[a-z]/(?:[^/?#]+/)*(?P<id>[^/?#]{{46}})/?(?:$|[?#])',
rf'{_BASE_URL_RE}(?!:v:)(?:[^/?#]+/)*stream\.aspx\?(?:[^#]+&)?id=(?P<id>[^&#]+)',
]
_TESTS = [{
'url': 'https://lut-my.sharepoint.com/:v:/g/personal/juha_eerola_student_lab_fi/EUrAmrktb4ZMhUcY9J2PqMEBD_9x_l0DyYWVgAvp-TTOMw?e=ZpQOOw',
'md5': '2950821d0d4937a0a76373782093b435',
'info_dict': {
'id': '01EQRS7EKKYCNLSLLPQZGIKRYY6SOY7KGB',
'display_id': 'EUrAmrktb4ZMhUcY9J2PqMEBD_9x_l0DyYWVgAvp-TTOMw',
'ext': 'mp4',
'title': 'CmvpJST',
'duration': 54.567,
'thumbnail': r're:https://.+/thumbnail',
'uploader_id': '8dcec565-a956-4b91-95e5-bacfb8bc015f',
},
}, {
'url': 'https://greaternyace.sharepoint.com/:v:/s/acementornydrive/ETski5eAfNVEoPRZUAyy1wEBpLgVFYWso5bjbZjfBLlPUg?e=PQUfVb',
'md5': 'c496a01644223273bff12e93e501afd1',
'info_dict': {
'id': '01QI4AVTZ3ESFZPAD42VCKB5CZKAGLFVYB',
'display_id': 'ETski5eAfNVEoPRZUAyy1wEBpLgVFYWso5bjbZjfBLlPUg',
'ext': 'mp4',
'title': '930103681233985536',
'duration': 3797.326,
'thumbnail': r're:https://.+/thumbnail',
},
}, {
'url': 'https://lut-my.sharepoint.com/personal/juha_eerola_student_lab_fi/_layouts/15/stream.aspx?id=%2Fpersonal%2Fjuha_eerola_student_lab_fi%2FDocuments%2FM-DL%2FCmvpJST.mp4&ga=1&referrer=StreamWebApp.Web&referrerScenario=AddressBarCopied.view',
'info_dict': {
'id': '01EQRS7EKKYCNLSLLPQZGIKRYY6SOY7KGB',
'display_id': '/personal/juha_eerola_student_lab_fi/Documents/M-DL/CmvpJST.mp4',
'ext': 'mp4',
'title': 'CmvpJST',
'duration': 54.567,
'thumbnail': r're:https://.+/thumbnail',
'uploader_id': '8dcec565-a956-4b91-95e5-bacfb8bc015f',
},
'skip': 'Session cookies needed',
}, {
'url': 'https://izoobasisschool.sharepoint.com/:v:/g/Eaqleq8COVBIvIPvod0U27oBypC6aWOkk8ptuDpmJ6arHw',
'only_matching': True,
}, {
'url': 'https://uskudaredutr-my.sharepoint.com/:v:/g/personal/songul_turkaydin_uskudar_edu_tr/EbTf-VRUIbtGuIN73tx1MuwBCHBOmNcWNqSLw61Fd2_o0g?e=n5Vkof',
'only_matching': True,
}, {
'url': 'https://epam-my.sharepoint.com/:v:/p/dzmitry_tamashevich/Ec4ZOs-rATZHjFYZWVxjczEB649FCoYFKDV_x3RxZiWAGA?e=4hswgA',
'only_matching': True,
}, {
'url': 'https://microsoft.sharepoint.com/:v:/t/MicrosoftSPARKRecordings-MSFTInternal/EWCyeqByVWBAt8wDvNZdV-UB0BvU5YVbKm0UHgdrUlI6dg?e=QbPck6',
'only_matching': True,
}]
def _real_extract(self, url):
display_id = urllib.parse.unquote(self._match_id(url))
webpage, urlh = self._download_webpage_handle(url, display_id)
if urllib.parse.urlparse(urlh.url).hostname == 'login.microsoftonline.com':
self.raise_login_required(
'Session cookies are required for this URL and can be passed '
'with the --cookies option. The --cookies-from-browser option will not work', method=None)
video_data = self._search_json(r'g_fileInfo\s*=', webpage, 'player config', display_id)
video_id = video_data['VroomItemId']
parsed_url = urllib.parse.urlparse(video_data['.transformUrl'])
base_media_url = urllib.parse.urlunparse(parsed_url._replace(
path=urllib.parse.urljoin(f'{parsed_url.path}/', '../videomanifest'),
query=urllib.parse.urlencode({
**urllib.parse.parse_qs(parsed_url.query),
'cTag': video_data['.ctag'],
'action': 'Access',
'part': 'index',
}, doseq=True)))
# Web player adds more params to the format URLs but we still get all formats without them
formats = self._extract_mpd_formats(
base_media_url, video_id, mpd_id='dash', query={'format': 'dash'}, fatal=False)
for hls_type in ('hls', 'hls-vnext'):
formats.extend(self._extract_m3u8_formats(
base_media_url, video_id, 'mp4', m3u8_id=hls_type,
query={'format': hls_type}, fatal=False, quality=-2))
if video_url := traverse_obj(video_data, ('downloadUrl', {url_or_none})):
formats.append({
'url': video_url,
'ext': determine_ext(video_data.get('extension') or video_data.get('name')),
'quality': 1,
'format_id': 'source',
'filesize': int_or_none(video_data.get('size')),
'vcodec': 'none' if video_data.get('isAudio') is True else None,
})
return {
'id': video_id,
'formats': formats,
'title': video_data.get('title') or video_data.get('displayName'),
'display_id': display_id,
'uploader_id': video_data.get('authorId'),
'duration': traverse_obj(video_data, (
'MediaServiceFastMetadata', {json.loads}, 'media', 'duration', {lambda x: x / 10000000})),
'thumbnail': url_or_none(video_data.get('thumbnailUrl')),
}