# coding=utf-8
from __future__ import unicode_literals
-import re
-
from .common import InfoExtractor
from ..utils import (
int_or_none,
}
def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('id')
+ video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
uploader_id = self._html_search_regex(
r'<li>from: <a href="/user/view/user/\S*?/uid/(\d*)',
webpage, 'video uploader_id', fatal=False)
- upload_date = self._html_search_regex(
+ upload_date = unified_strdate(self._html_search_regex(
r'<li>uploaded: (.*?)</li>', webpage, 'video upload_date',
- fatal=False)
- if upload_date is not None:
- upload_date = unified_strdate(upload_date)
+ fatal=False))
view_count_str = self._html_search_regex(
r'<li>Hits: ([0-9]+?)</li>', webpage, 'video view_count',
fatal=False)