from __future__ import unicode_literals
from .common import InfoExtractor
-from ..compat import (
- compat_urllib_parse,
- compat_urllib_request,
+from ..utils import (
+ ExtractorError,
+ sanitized_Request,
+ urlencode_postdata,
)
-from ..utils import ExtractorError
class PrimeShareTVIE(InfoExtractor):
webpage, 'wait time', default=7)) + 1
self._sleep(wait_time, video_id)
- req = compat_urllib_request.Request(
- url, compat_urllib_parse.urlencode(fields), headers)
+ req = sanitized_Request(
+ url, urlencode_postdata(fields), headers)
video_page = self._download_webpage(
req, video_id, 'Downloading video page')