|
@@ -14,7 +14,7 @@ from ..utils import (
|
|
|
|
|
|
class PlayedIE(InfoExtractor):
|
|
|
IE_NAME = 'played.to'
|
|
|
- _VALID_URL = r'https?://played\.to/(?P<id>[a-zA-Z0-9_-]+)'
|
|
|
+ _VALID_URL = r'https?://(?:www\.)?played\.to/(?P<id>[a-zA-Z0-9_-]+)'
|
|
|
|
|
|
_TEST = {
|
|
|
'url': 'http://played.to/j2f2sfiiukgt',
|
|
@@ -27,15 +27,14 @@ class PlayedIE(InfoExtractor):
|
|
|
}
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
- mobj = re.match(self._VALID_URL, url)
|
|
|
- video_id = mobj.group('id')
|
|
|
+ video_id = self._match_id(url)
|
|
|
|
|
|
orig_webpage = self._download_webpage(url, video_id)
|
|
|
- fields = re.findall(r'type="hidden" name="(.+?)"\s* value="?(.+?)">', orig_webpage)
|
|
|
+ fields = re.findall(
|
|
|
+ r'type="hidden" name="([^"]+)"\s+value="([^"]+)">', orig_webpage)
|
|
|
data = dict(fields)
|
|
|
|
|
|
- self.to_screen('%s: Waiting for timeout' % video_id)
|
|
|
- time.sleep(2)
|
|
|
+ self._sleep(2, video_id)
|
|
|
|
|
|
post = compat_urllib_parse.urlencode(data)
|
|
|
headers = {
|
|
@@ -54,4 +53,4 @@ class PlayedIE(InfoExtractor):
|
|
|
'id': video_id,
|
|
|
'title': title,
|
|
|
'url': video_url,
|
|
|
- }
|
|
|
+ }
|