mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-14 04:13:20 +00:00
3d2623a898
Authored by: coletdjnz, pukkandan
81 lines
2.6 KiB
Python
81 lines
2.6 KiB
Python
import re
|
|
import urllib.parse
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
ExtractorError,
|
|
determine_ext,
|
|
url_or_none,
|
|
urlencode_postdata,
|
|
)
|
|
|
|
|
|
class OwnCloudIE(InfoExtractor):
|
|
_INSTANCES_RE = '|'.join((
|
|
r'(?:[^\.]+\.)?sciebo\.de',
|
|
r'cloud\.uni-koblenz-landau\.de',
|
|
))
|
|
_VALID_URL = rf'https?://(?:{_INSTANCES_RE})/s/(?P<id>[\w.-]+)'
|
|
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://ruhr-uni-bochum.sciebo.de/s/wWhqZzh9jTumVFN',
|
|
'info_dict': {
|
|
'id': 'wWhqZzh9jTumVFN',
|
|
'ext': 'mp4',
|
|
'title': 'CmvpJST.mp4',
|
|
},
|
|
},
|
|
{
|
|
'url': 'https://ruhr-uni-bochum.sciebo.de/s/WNDuFu0XuFtmm3f',
|
|
'info_dict': {
|
|
'id': 'WNDuFu0XuFtmm3f',
|
|
'ext': 'mp4',
|
|
'title': 'CmvpJST.mp4',
|
|
},
|
|
'params': {
|
|
'videopassword': '12345',
|
|
},
|
|
},
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage, urlh = self._download_webpage_handle(url, video_id)
|
|
|
|
if re.search(r'<label[^>]+for="password"', webpage):
|
|
webpage = self._verify_video_password(webpage, urlh.url, video_id)
|
|
|
|
hidden_inputs = self._hidden_inputs(webpage)
|
|
title = hidden_inputs.get('filename')
|
|
parsed_url = urllib.parse.urlparse(url)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'url': url_or_none(hidden_inputs.get('downloadURL')) or parsed_url._replace(
|
|
path=urllib.parse.urljoin(parsed_url.path, 'download')).geturl(),
|
|
'ext': determine_ext(title),
|
|
}
|
|
|
|
def _verify_video_password(self, webpage, url, video_id):
|
|
password = self.get_param('videopassword')
|
|
if password is None:
|
|
raise ExtractorError(
|
|
'This video is protected by a password, use the --video-password option',
|
|
expected=True)
|
|
|
|
validation_response = self._download_webpage(
|
|
url, video_id, 'Validating Password', 'Wrong password?',
|
|
data=urlencode_postdata({
|
|
'requesttoken': self._hidden_inputs(webpage)['requesttoken'],
|
|
'password': password,
|
|
}))
|
|
|
|
if re.search(r'<label[^>]+for="password"', validation_response):
|
|
warning = self._search_regex(
|
|
r'<div[^>]+class="warning">([^<]*)</div>', validation_response,
|
|
'warning', default='The password is wrong')
|
|
raise ExtractorError(f'Opening the video failed, {self.IE_NAME} said: {warning!r}', expected=True)
|
|
return validation_response
|