import os.path
from .common import InfoExtractor
-from ..compat import (
- compat_urllib_parse,
- compat_urllib_request,
-)
+from ..compat import compat_urllib_parse
from ..utils import (
ExtractorError,
+ sanitized_Request,
)
if m_error:
raise ExtractorError(m_error.group('msg'), expected=True)
- fields = re.findall(
- r'type="hidden" name="([^"]+)"\s+value="([^"]+)">', orig_webpage)
- data = dict(fields)
+ data = self._hidden_inputs(orig_webpage)
self._sleep(2, video_id)
headers = {
b'Content-Type': b'application/x-www-form-urlencoded',
}
- req = compat_urllib_request.Request(url, post, headers)
+ req = sanitized_Request(url, post, headers)
webpage = self._download_webpage(
req, video_id, note='Downloading video page ...')