class ComCarCoffIE(InfoExtractor):
- _VALID_URL = r'https?://(?:www\.)?comediansincarsgettingcoffee\.com/(?P<id>[a-z0-9\-]+)/?'
+ _VALID_URL = r'http://(?:www\.)?comediansincarsgettingcoffee\.com/(?P<id>[a-z0-9\-]*)'
_TESTS = [{
'url': 'http://comediansincarsgettingcoffee.com/miranda-sings-happy-thanksgiving-miranda/',
'info_dict': {
def _real_extract(self, url):
display_id = self._match_id(url)
+ if not display_id:
+ display_id = 'comediansincarsgettingcoffee.com'
webpage = self._download_webpage(url, display_id)
full_data = json.loads(self._search_regex(