1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-11-24 04:56:55 +01:00

added an IE and test for dotsub.

This commit is contained in:
M.Yasoob Khalid 2013-07-09 00:05:52 +05:00
parent 81f0259b9e
commit 13e06d298c
2 changed files with 43 additions and 1 deletions

View File

@ -1,4 +1,3 @@
from .archiveorg import ArchiveOrgIE
from .ard import ARDIE
from .arte import ArteTvIE
@ -12,6 +11,7 @@ from .comedycentral import ComedyCentralIE
from .cspan import CSpanIE
from .dailymotion import DailymotionIE
from .depositfiles import DepositFilesIE
from .dotsub import DotsubIE
from .dreisat import DreiSatIE
from .eighttracks import EightTracksIE
from .escapist import EscapistIE

View File

@ -0,0 +1,42 @@
import re
import json
from .common import InfoExtractor
class DotsubIE(InfoExtractor):
_VALID_URL = r'(?:http://)?(?:www\.)?dotsub\.com/view/([^/]+)'
_TEST = {
u'url': u'http://dotsub.com/view/aed3b8b2-1889-4df5-ae63-ad85f5572f27',
u'file': u'aed3b8b2-1889-4df5-ae63-ad85f5572f27.flv',
u'md5': u'0914d4d69605090f623b7ac329fea66e',
u'info_dict': {
u"title": u"Pyramids of Waste (2010), AKA The Lightbulb Conspiracy - Planned obsolescence documentary",
u"uploader": u"4v4l0n42",
u'description': u'Pyramids of Waste (2010) also known as "The lightbulb conspiracy" is a documentary about how our economic system based on consumerism and planned obsolescence is breaking our planet down.\r\n\r\nSolutions to this can be found at:\r\nhttp://robotswillstealyourjob.com\r\nhttp://www.federicopistono.org\r\n\r\nhttp://opensourceecology.org\r\nhttp://thezeitgeistmovement.com',
u'thumbnail': u'http://dotsub.com/media/aed3b8b2-1889-4df5-ae63-ad85f5572f27/p'
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group(1)
info_url = "https://dotsub.com/api/media/%s/metadata" %(video_id)
webpage = self._download_webpage(info_url, video_id)
info = json.loads(webpage)
video_url = info['mediaURI']
uploader = info['user']
description = info['description']
view_count = info['numberOfViews']
title = info['title']
thumbnail_url = info['screenshotURI']
ext = 'flv'
return [{
'id': video_id,
'url': video_url,
'ext': ext,
'title': title,
'thumbnail': thumbnail_url,
'description': description,
'uploader': uploader,
'view_count': view_count,
}]