mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-07 19:52:40 +01:00
commit
b8d61935a0
@ -46,6 +46,7 @@
|
||||
from .clipfish import ClipfishIE
|
||||
from .cliphunter import CliphunterIE
|
||||
from .clipsyndicate import ClipsyndicateIE
|
||||
from .cloudy import CloudyIE
|
||||
from .clubic import ClubicIE
|
||||
from .cmt import CMTIE
|
||||
from .cnet import CNETIE
|
||||
|
67
youtube_dl/extractor/cloudy.py
Normal file
67
youtube_dl/extractor/cloudy.py
Normal file
@ -0,0 +1,67 @@
|
||||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
ExtractorError,
|
||||
compat_parse_qs,
|
||||
compat_urllib_parse,
|
||||
)
|
||||
|
||||
|
||||
class CloudyIE(InfoExtractor):
|
||||
_VALID_URL = r'''(?x)
|
||||
https?://(?:www\.)?cloudy\.ec/
|
||||
(?:v/|embed\.php\?id=)
|
||||
(?P<id>[A-Za-z0-9]+)
|
||||
'''
|
||||
_API_URL = 'http://www.cloudy.ec/api/player.api.php?%s'
|
||||
_TEST = {
|
||||
'url': 'https://www.cloudy.ec/v/af511e2527aac',
|
||||
'md5': '5cb253ace826a42f35b4740539bedf07',
|
||||
'info_dict': {
|
||||
'id': 'af511e2527aac',
|
||||
'ext': 'flv',
|
||||
'title': 'Funny Cats and Animals Compilation june 2013',
|
||||
}
|
||||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
video_id = mobj.group('id')
|
||||
|
||||
url = 'http://www.cloudy.ec/embed.php?id=%s' % video_id
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
file_key = self._search_regex(
|
||||
r'filekey\s*=\s*"([^"]+)"', webpage, 'file_key')
|
||||
data_url = self._API_URL % compat_urllib_parse.urlencode({
|
||||
'file': video_id,
|
||||
'key': file_key,
|
||||
})
|
||||
player_data = self._download_webpage(
|
||||
data_url, video_id, 'Downloading player data')
|
||||
data = compat_parse_qs(player_data)
|
||||
|
||||
if 'error' in data:
|
||||
raise ExtractorError(
|
||||
'%s error: %s' % (self.IE_NAME, ' '.join(data['error_msg'])),
|
||||
expected=True)
|
||||
|
||||
title = data.get('title', [None])[0]
|
||||
if title:
|
||||
title = title.replace('&asdasdas', '').strip()
|
||||
|
||||
formats = []
|
||||
formats.append({
|
||||
'format_id': 'sd',
|
||||
'url': data.get('url', [None])[0],
|
||||
})
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': title,
|
||||
'formats': formats,
|
||||
}
|
Loading…
Reference in New Issue
Block a user