1
0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-07 11:42:43 +01:00

Merge remote-tracking branch 'yasoob/master'

This commit is contained in:
Philipp Hagemeister 2014-08-22 01:38:59 +02:00
commit a200f4cee2
2 changed files with 49 additions and 0 deletions

View File

@ -69,6 +69,7 @@
from .dotsub import DotsubIE
from .dreisat import DreiSatIE
from .drtv import DRTVIE
from .dump import DumpIE
from .defense import DefenseGouvFrIE
from .discovery import DiscoveryIE
from .divxstage import DivxStageIE

View File

@ -0,0 +1,48 @@
# encoding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
)
class DumpIE(InfoExtractor):
_VALID_URL = r'^https?://(?:www\.)?dump\.com/(?P<id>[a-zA-Z0-9]+)/'
_TEST = {
u'url': u'http://www.dump.com/oneus/',
u'file': u'oneus.flv',
u'md5': u'ad71704d1e67dfd9e81e3e8b42d69d99',
u'info_dict': {
u"title": u"He's one of us.",
},
}
def _real_extract(self, url):
m = re.match(self._VALID_URL, url)
video_id = m.group('id')
# Note: There is an easier-to-parse configuration at
# http://www.aparat.com/video/video/config/videohash/%video_id
# but the URL in there does not work
webpage = self._download_webpage(url, video_id)
try:
video_url = re.findall(r'file","(.+?.flv)"', webpage)[-1]
except IndexError:
raise ExtractorError(u'No video URL found')
thumb = re.findall('<meta property="og:image" content="(.+?)"',webpage)[0]
title = self._search_regex(r'<b>([^"]+)</b>', webpage, u'title')
return {
'id': video_id,
'title': title,
'url': video_url,
'ext': 'flv',
'thumbnail': thumb,
}