1
0
mirror of https://github.com/spaam/svtplay-dl.git synced 2024-11-28 06:04:17 +01:00
svtplay-dl/lib/svtplay_dl/service/__init__.py

103 lines
2.9 KiB
Python
Raw Normal View History

2013-03-02 21:26:28 +01:00
# ex:ts=4:sw=4:sts=4:et
# -*- tab-width: 4; c-basic-offset: 4; indent-tabs-mode: nil -*-
from __future__ import absolute_import
2013-03-23 16:11:36 +01:00
import re
2014-01-05 17:28:00 +01:00
from svtplay_dl.utils.urllib import urlparse
2013-03-23 16:11:36 +01:00
class Service(object):
supported_domains = []
supported_domains_re = []
def handles(self, url):
urlp = urlparse(url)
# Apply supported_domains_re regexp to the netloc. This
# is meant for 'dynamic' domains, e.g. containing country
# information etc.
for domain_re in [re.compile(x) for x in self.supported_domains_re]:
if domain_re.match(urlp.netloc):
return True
if urlp.netloc in self.supported_domains:
return True
# For every listed domain, try with www. subdomain as well.
if urlp.netloc in ['www.'+x for x in self.supported_domains]:
return True
return False
from svtplay_dl.service.aftonbladet import Aftonbladet
from svtplay_dl.service.dr import Dr
from svtplay_dl.service.expressen import Expressen
from svtplay_dl.service.hbo import Hbo
from svtplay_dl.service.justin import Justin
from svtplay_dl.service.kanal5 import Kanal5
from svtplay_dl.service.mtvservices import Mtvservices
from svtplay_dl.service.nrk import Nrk
from svtplay_dl.service.qbrick import Qbrick
from svtplay_dl.service.ruv import Ruv
from svtplay_dl.service.radioplay import Radioplay
from svtplay_dl.service.sr import Sr
from svtplay_dl.service.svtplay import Svtplay
from svtplay_dl.service.tv4play import Tv4play
from svtplay_dl.service.urplay import Urplay
from svtplay_dl.service.viaplay import Viaplay
from svtplay_dl.service.vimeo import Vimeo
2013-03-23 16:11:36 +01:00
from svtplay_dl.utils import get_http_data
sites = [
Aftonbladet(),
Dr(),
Expressen(),
Hbo(),
Justin(),
Kanal5(),
Mtvservices(),
Nrk(),
Qbrick(),
Ruv(),
Radioplay(),
Sr(),
Svtplay(),
Tv4play(),
Urplay(),
Viaplay(),
Vimeo()]
class Generic(object):
''' Videos embed in sites '''
2013-03-23 16:11:36 +01:00
def get(self, url):
data = get_http_data(url)
2013-11-14 22:43:39 +01:00
match = re.search(r"src=\"(http://www.svt.se/wd.*)\" height", data)
stream = None
if match:
url = match.group(1)
for i in sites:
if i.handles(url):
return url, i
match = re.search(r"src=\"(http://player.vimeo.com/video/[0-9]+)\" ", data)
if match:
for i in sites:
if i.handles(match.group(1)):
return match.group(1), i
match = re.search(r"tv4video.swf\?vid=(\d+)", data)
2013-03-24 14:55:14 +01:00
if match:
url = "http://www.tv4play.se/?video_id=%s" % match.group(1)
for i in sites:
if i.handles(url):
2013-03-24 14:55:14 +01:00
return url, i
return url, stream
def service_handler(url):
handler = None
for i in sites:
if i.handles(url):
handler = i
break
return handler