from __future__ import unicode_literals
-import re
-
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,
)
-from ..utils import ExtractorError
class PrimeShareTVIE(InfoExtractor):
if '>File not exist<' in webpage:
raise ExtractorError('Video %s does not exist' % video_id, expected=True)
- fields = dict(re.findall(r'''(?x)<input\s+
- type="hidden"\s+
- name="([^"]+)"\s+
- (?:id="[^"]+"\s+)?
- value="([^"]*)"
- ''', webpage))
+ fields = self._hidden_inputs(webpage)
headers = {
'Referer': url,
webpage, 'wait time', default=7)) + 1
self._sleep(wait_time, video_id)
- req = compat_urllib_request.Request(
+ req = sanitized_Request(
url, compat_urllib_parse.urlencode(fields), headers)
video_page = self._download_webpage(
req, video_id, 'Downloading video page')