@ -87,8 +87,8 @@ class PornHubIE(InfoExtractor):
webpage = self._download_webpage(req, video_id)
error_msg = self._html_search_regex(
r'<div[^>]+class="removed">\s*<div[^>]*>\s*<p>\s*<span>([^<]*)</span>',
webpage, 'error message', default=None)
r'(?s)<div[^>]+class=(["\']).*?\bremoved\b.*?\1[^>]*>(?P<error>.+?)</div>',
webpage, 'error message', default=None, group='error')
if error_msg:
error_msg = re.sub(r'\s+', ' ', error_msg)
raise ExtractorError(