Added new host: allmyvideos.net
parent
64892c0b79
commit
38349518f1
@ -0,0 +1,51 @@
|
||||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
compat_urllib_parse,
|
||||
compat_urllib_request,
|
||||
)
|
||||
|
||||
|
||||
class AllmyvideosIE(InfoExtractor):
|
||||
IE_NAME = 'allmyvideos.net'
|
||||
_VALID_URL = r'https?://allmyvideos\.net/(?P<id>[a-zA-Z0-9_-]+)'
|
||||
|
||||
_TEST = {
|
||||
'url': 'http://allmyvideos.net/jih3nce3x6wn',
|
||||
'md5': '8f26c1e7102556a0d7f24306d32c2092',
|
||||
'info_dict': {
|
||||
'id': 'jih3nce3x6wn',
|
||||
'ext': 'mp4',
|
||||
'title': 'youtube-dl test video',
|
||||
},
|
||||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
video_id = mobj.group('id')
|
||||
|
||||
orig_webpage = self._download_webpage(url, video_id)
|
||||
fields = re.findall(r'type="hidden" name="(.+?)"\s* value="?(.+?)">', orig_webpage)
|
||||
data = {}
|
||||
for name, value in fields:
|
||||
data[name] = value
|
||||
|
||||
post = compat_urllib_parse.urlencode(data)
|
||||
headers = {
|
||||
b'Content-Type': b'application/x-www-form-urlencoded',
|
||||
}
|
||||
req = compat_urllib_request.Request(url, post, headers)
|
||||
webpage = self._download_webpage(req, video_id, note='Downloading video page ...')
|
||||
|
||||
#Could be several links with different quality
|
||||
links = re.findall(r'"file" : "?(.+?)",', webpage)
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': data['fname'][:len(data['fname'])-4], #Remove .mp4 extension
|
||||
'url': links[len(links)-1] #Choose the higher quality link
|
||||
}
|
Loading…
Reference in New Issue