mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-17 20:37:21 +01:00
46 lines
1.2 KiB
Python
46 lines
1.2 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
|
|
class CrackedIE(InfoExtractor):
|
|
_VALID_URL = r'http?://.*?\.cracked\.com/video_+(?P<id>.*)_.*'
|
|
_TEST = {
|
|
'url': 'http://www.cracked.com/video_18803_4-social-criticisms-hidden-in-sonic-hedgehog-games.html',
|
|
|
|
'info_dict': {
|
|
'id': '18803',
|
|
'ext': 'mp4',
|
|
'title': "4 Social Criticisms Hidden in 'Sonic the Hedgehog' Games | Cracked.com",
|
|
'height': 375,
|
|
'width': 666,
|
|
|
|
|
|
}
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
video_id = mobj.group('id')
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
title = self._search_regex(r'<title>(.*?)</title>',webpage,'title')
|
|
video_url = self._search_regex(r'var CK_vidSrc = "+(.*)"',webpage,'url')
|
|
width = self._search_regex(r'width="(.*?)"',webpage,'width')
|
|
height = re.findall(r'height="(.*?)"',webpage)[1]
|
|
|
|
|
|
|
|
|
|
return {
|
|
'url':video_url,
|
|
'id': video_id,
|
|
'ext':'mp4',
|
|
'title':title,
|
|
'height':int(height),
|
|
'width':int(width)
|
|
|
|
|
|
} |