1
0
mirror of https://github.com/spaam/svtplay-dl.git synced 2024-11-24 12:15:40 +01:00
svtplay-dl/lib/svtplay_dl/service/aftonbladet.py
2017-09-18 14:47:34 +02:00

74 lines
2.6 KiB
Python

# ex:ts=4:sw=4:sts=4:et
# -*- tab-width: 4; c-basic-offset: 4; indent-tabs-mode: nil -*-
from __future__ import absolute_import
import re
import json
from svtplay_dl.service import Service
from svtplay_dl.utils import decode_html_entities
from svtplay_dl.utils.urllib import urlparse
from svtplay_dl.error import ServiceError
from svtplay_dl.fetcher.hls import hlsparse
class Aftonbladettv(Service):
supported_domains = ['tv.aftonbladet.se', "svd.se"]
def get(self):
data = self.get_urldata()
if self.exclude():
yield ServiceError("Excluding video")
return
apiurl = None
match = re.search('data-player-config="([^"]+)"', data)
if not match:
match = re.search('data-svpPlayer-video="([^"]+)"', data)
if not match:
yield ServiceError("Can't find video info")
return
data = json.loads(decode_html_entities(match.group(1)))
videoId = data["playerOptions"]["id"]
apiurl = data["playerOptions"]["api"]
vendor = data["playerOptions"]["vendor"]
self.options.live = data["live"]
if not self.options.live:
dataurl = "{0}{1}/assets/{2}?appName=svp-player".format(apiurl, vendor, videoId)
data = self.http.request("get", dataurl).text
data = json.loads(data)
streams = hlsparse(self.options, self.http.request("get", data["streamUrls"]["hls"]), data["streamUrls"]["hls"])
if streams:
for n in list(streams.keys()):
yield streams[n]
class Aftonbladet(Service):
supported_domains = ["aftonbladet.se"]
def get(self):
data = self.get_urldata()
if self.exclude():
yield ServiceError("Excluding video")
return
match = re.search('window.FLUX_STATE = ({.*})</script>', data)
if not match:
yield ServiceError("Can't find video info")
return
janson = json.loads(match.group(1))
articleid = janson["article"]["currentArticleId"]
components = janson["articles"][articleid]["article"]["components"]
for i in components:
if "components" in i:
for n in i["components"]:
if "type" in n and n["type"] == "video":
streams = hlsparse(self.options, self.http.request("get", n["videoAsset"]["streamUrls"]["hls"]),
n["videoAsset"]["streamUrls"]["hls"])
if streams:
for n in list(streams.keys()):
yield streams[n]