0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-25 02:45:12 +00:00
yt-dlp/yt_dlp/extractor/zoom.py
2023-11-11 19:51:54 +00:00

164 lines
6.9 KiB
Python

from .common import InfoExtractor
from ..utils import (
ExtractorError,
int_or_none,
js_to_json,
parse_filesize,
parse_resolution,
str_or_none,
traverse_obj,
url_basename,
urlencode_postdata,
urljoin,
)
class ZoomIE(InfoExtractor):
IE_NAME = 'zoom'
_VALID_URL = r'(?P<base_url>https?://(?:[^.]+\.)?zoom\.us/)rec(?:ording)?/(?P<type>play|share)/(?P<id>[\w.-]+)'
_TESTS = [{
'url': 'https://economist.zoom.us/rec/play/dUk_CNBETmZ5VA2BwEl-jjakPpJ3M1pcfVYAPRsoIbEByGsLjUZtaa4yCATQuOL3der8BlTwxQePl_j0.EImBkXzTIaPvdZO5',
'md5': 'ab445e8c911fddc4f9adc842c2c5d434',
'info_dict': {
'id': 'dUk_CNBETmZ5VA2BwEl-jjakPpJ3M1pcfVYAPRsoIbEByGsLjUZtaa4yCATQuOL3der8BlTwxQePl_j0.EImBkXzTIaPvdZO5',
'ext': 'mp4',
'title': 'China\'s "two sessions" and the new five-year plan',
},
'skip': 'Recording requires email authentication to access',
}, {
# play URL
'url': 'https://ffgolf.zoom.us/rec/play/qhEhXbrxq1Zoucx8CMtHzq1Z_2YZRPVCqWK_K-2FkEGRsSLDeOX8Tu4P6jtjZcRry8QhIbvKZdtr4UNo.QcPn2debFskI9whJ',
'md5': '2c4b1c4e5213ebf9db293e88d9385bee',
'info_dict': {
'id': 'qhEhXbrxq1Zoucx8CMtHzq1Z_2YZRPVCqWK_K-2FkEGRsSLDeOX8Tu4P6jtjZcRry8QhIbvKZdtr4UNo.QcPn2debFskI9whJ',
'ext': 'mp4',
'title': 'Prépa AF2023 - Séance 5 du 11 avril - R20/VM/GO',
},
}, {
# share URL
'url': 'https://us02web.zoom.us/rec/share/hkUk5Zxcga0nkyNGhVCRfzkA2gX_mzgS3LpTxEEWJz9Y_QpIQ4mZFOUx7KZRZDQA.9LGQBdqmDAYgiZ_8',
'md5': '90fdc7cfcaee5d52d1c817fc03c43c9b',
'info_dict': {
'id': 'hkUk5Zxcga0nkyNGhVCRfzkA2gX_mzgS3LpTxEEWJz9Y_QpIQ4mZFOUx7KZRZDQA.9LGQBdqmDAYgiZ_8',
'ext': 'mp4',
'title': 'Timea Andrea Lelik\'s Personal Meeting Room',
},
'skip': 'This recording has expired',
}, {
# view_with_share URL
'url': 'https://cityofdetroit.zoom.us/rec/share/VjE-5kW3xmgbEYqR5KzRgZ1OFZvtMtiXk5HyRJo5kK4m5PYE6RF4rF_oiiO_9qaM.UTAg1MI7JSnF3ZjX',
'md5': 'bdc7867a5934c151957fb81321b3c024',
'info_dict': {
'id': 'VjE-5kW3xmgbEYqR5KzRgZ1OFZvtMtiXk5HyRJo5kK4m5PYE6RF4rF_oiiO_9qaM.UTAg1MI7JSnF3ZjX',
'ext': 'mp4',
'title': 'February 2022 Detroit Revenue Estimating Conference',
'duration': 7299,
'formats': 'mincount:3',
},
}]
def _get_page_data(self, webpage, video_id):
return self._search_json(
r'window\.__data__\s*=', webpage, 'data', video_id, transform_source=js_to_json)
def _get_real_webpage(self, url, base_url, video_id, url_type):
webpage = self._download_webpage(url, video_id, note=f'Downloading {url_type} webpage')
try:
form = self._form_hidden_inputs('password_form', webpage)
except ExtractorError:
return webpage
password = self.get_param('videopassword')
if not password:
raise ExtractorError(
'This video is protected by a passcode, use the --video-password option', expected=True)
is_meeting = form.get('useWhichPasswd') == 'meeting'
validation = self._download_json(
base_url + 'rec/validate%s_passwd' % ('_meet' if is_meeting else ''),
video_id, 'Validating passcode', 'Wrong passcode', data=urlencode_postdata({
'id': form[('meet' if is_meeting else 'file') + 'Id'],
'passwd': password,
'action': form.get('action'),
}))
if not validation.get('status'):
raise ExtractorError(validation['errorMessage'], expected=True)
return self._download_webpage(url, video_id, note=f'Re-downloading {url_type} webpage')
def _real_extract(self, url):
base_url, url_type, video_id = self._match_valid_url(url).group('base_url', 'type', 'id')
query = {}
if url_type == 'share':
webpage = self._get_real_webpage(url, base_url, video_id, 'share')
meeting_id = self._get_page_data(webpage, video_id)['meetingId']
redirect_path = self._download_json(
f'{base_url}nws/recording/1.0/play/share-info/{meeting_id}',
video_id, note='Downloading share info JSON')['result']['redirectUrl']
url = urljoin(base_url, redirect_path)
query['continueMode'] = 'true'
webpage = self._get_real_webpage(url, base_url, video_id, 'play')
file_id = self._get_page_data(webpage, video_id)['fileId']
if not file_id:
# When things go wrong, file_id can be empty string
raise ExtractorError('Unable to extract file ID')
data = self._download_json(
f'{base_url}nws/recording/1.0/play/info/{file_id}', video_id, query=query,
note='Downloading play info JSON')['result']
subtitles = {}
for _type in ('transcript', 'cc', 'chapter'):
if data.get('%sUrl' % _type):
subtitles[_type] = [{
'url': urljoin(base_url, data['%sUrl' % _type]),
'ext': 'vtt',
}]
formats = []
if data.get('viewMp4Url'):
formats.append({
'format_note': 'Camera stream',
'url': data['viewMp4Url'],
'width': int_or_none(traverse_obj(data, ('viewResolvtions', 0))),
'height': int_or_none(traverse_obj(data, ('viewResolvtions', 1))),
'format_id': 'view',
'ext': 'mp4',
'filesize_approx': parse_filesize(str_or_none(traverse_obj(data, ('recording', 'fileSizeInMB')))),
'preference': 0
})
if data.get('shareMp4Url'):
formats.append({
'format_note': 'Screen share stream',
'url': data['shareMp4Url'],
'width': int_or_none(traverse_obj(data, ('shareResolvtions', 0))),
'height': int_or_none(traverse_obj(data, ('shareResolvtions', 1))),
'format_id': 'share',
'ext': 'mp4',
'preference': -1
})
view_with_share_url = data.get('viewMp4WithshareUrl')
if view_with_share_url:
formats.append({
**parse_resolution(self._search_regex(
r'_(\d+x\d+)\.mp4', url_basename(view_with_share_url), 'resolution', default=None)),
'format_note': 'Screen share with camera',
'url': view_with_share_url,
'format_id': 'view_with_share',
'ext': 'mp4',
'preference': 1
})
return {
'id': video_id,
'title': str_or_none(traverse_obj(data, ('meet', 'topic'))),
'duration': int_or_none(data.get('duration')),
'subtitles': subtitles,
'formats': formats,
'http_headers': {
'Referer': base_url,
},
}