mirror of
https://github.com/spaam/svtplay-dl.git
synced 2024-11-27 21:54:17 +01:00
dr: fix so it work with the new site
This commit is contained in:
parent
2bddaba826
commit
6bde2e76bf
@ -1,20 +1,10 @@
|
||||
# ex:ts=4:sw=4:sts=4:et
|
||||
# -*- tab-width: 4; c-basic-offset: 4; indent-tabs-mode: nil -*-
|
||||
import base64
|
||||
import binascii
|
||||
import copy
|
||||
import hashlib
|
||||
import json
|
||||
import logging
|
||||
import re
|
||||
from urllib.parse import urljoin
|
||||
from urllib.parse import urlparse
|
||||
import uuid
|
||||
|
||||
from cryptography.hazmat.backends import default_backend
|
||||
from cryptography.hazmat.primitives.ciphers import algorithms
|
||||
from cryptography.hazmat.primitives.ciphers import Cipher
|
||||
from cryptography.hazmat.primitives.ciphers import modes
|
||||
from svtplay_dl.error import ServiceError
|
||||
from svtplay_dl.fetcher.hds import hdsparse
|
||||
from svtplay_dl.fetcher.hls import hlsparse
|
||||
from svtplay_dl.service import OpenGraphThumbMixin
|
||||
from svtplay_dl.service import Service
|
||||
@ -27,113 +17,61 @@ class Dr(Service, OpenGraphThumbMixin):
|
||||
def get(self):
|
||||
data = self.get_urldata()
|
||||
|
||||
match = re.search(r'resource:[ ]*"([^"]*)",', data)
|
||||
if match:
|
||||
resource_url = match.group(1)
|
||||
resource_data = self.http.request("get", resource_url).content
|
||||
resource = json.loads(resource_data)
|
||||
streams = self.find_stream(self.config, resource)
|
||||
yield from streams
|
||||
else:
|
||||
match = re.search(r'resource="([^"]*)"', data)
|
||||
if not match:
|
||||
yield ServiceError("Cant find resource info for this video")
|
||||
return
|
||||
if match.group(1)[:4] != "http":
|
||||
resource_url = "http:{}".format(match.group(1))
|
||||
else:
|
||||
resource_url = match.group(1)
|
||||
resource_data = self.http.request("get", resource_url).text
|
||||
resource = json.loads(resource_data)
|
||||
match = re.search("__data = ([^<]+)</script>", data)
|
||||
if not match:
|
||||
yield ServiceError("Cant find info for this video")
|
||||
return
|
||||
janson = json.loads(match.group(1))
|
||||
page = janson["cache"]["page"][list(janson["cache"]["page"].keys())[0]]
|
||||
offers = page["entries"][0]["item"]["offers"]
|
||||
resolution = None
|
||||
vid = None
|
||||
for i in offers:
|
||||
if i["deliveryType"] == "Stream":
|
||||
vid = i["scopes"][0]
|
||||
resolution = i["resolution"]
|
||||
|
||||
if "Links" not in resource:
|
||||
yield ServiceError("Cant access this video. its geoblocked.")
|
||||
return
|
||||
if "SubtitlesList" in resource and len(resource["SubtitlesList"]) > 0:
|
||||
suburl = resource["SubtitlesList"][0]["Uri"]
|
||||
yield subtitle(copy.copy(self.config), "wrst", suburl, output=self.output)
|
||||
if "Data" in resource:
|
||||
streams = self.find_stream(self.config, resource)
|
||||
yield from streams
|
||||
else:
|
||||
for stream in resource["Links"]:
|
||||
uri = stream["Uri"]
|
||||
if uri is None:
|
||||
uri = self._decrypt(stream["EncryptedUri"])
|
||||
deviceid = uuid.uuid1()
|
||||
res = self.http.request(
|
||||
"post",
|
||||
"https://isl.dr-massive.com/api/authorization/anonymous-sso?device=web_browser&ff=idp%2Cldp&lang=da",
|
||||
json={"deviceId": str(deviceid), "scopes": ["Catalog"], "optout": True},
|
||||
)
|
||||
token = res.json()[0]["value"]
|
||||
|
||||
if stream["Target"] == "HDS":
|
||||
streams = hdsparse(copy.copy(self.config), self.http.request("get", uri, params={"hdcore": "3.7.0"}), uri, output=self.output)
|
||||
if streams:
|
||||
for n in list(streams.keys()):
|
||||
yield streams[n]
|
||||
if stream["Target"] == "HLS":
|
||||
streams = hlsparse(self.config, self.http.request("get", uri), uri, output=self.output)
|
||||
url = "https://isl.dr-massive.com/api/account/items/{}/videos?delivery=stream&device=web_browser&ff=idp%2Cldp&lang=da&resolution={}&sub=Anonymous".format(
|
||||
vid, resolution
|
||||
)
|
||||
res = self.http.request("get", url, headers={"authorization": "Bearer {}".format(token)})
|
||||
for video in res.json():
|
||||
if video["accessService"] == "StandardVideo":
|
||||
if video["format"] == "video/hls":
|
||||
res = self.http.request("get", video["url"])
|
||||
if res.status_code > 400:
|
||||
yield ServiceError("Can't play this because the video is geoblocked or not available.")
|
||||
else:
|
||||
streams = hlsparse(self.config, res, video["url"], output=self.output)
|
||||
for n in list(streams.keys()):
|
||||
yield streams[n]
|
||||
yield subtitle(copy.copy(self.config), "wrst", video["subtitles"][0]["link"], output=self.output)
|
||||
|
||||
def find_all_episodes(self, config):
|
||||
episodes = []
|
||||
matches = re.findall(r'<button class="show-more" data-url="([^"]+)" data-partial="([^"]+)"', self.get_urldata())
|
||||
for encpath, enccomp in matches:
|
||||
newstyle = "_" in encpath
|
||||
if newstyle:
|
||||
encbasepath = encpath.split("_")[0]
|
||||
path = base64.b64decode(encbasepath + "===").decode("latin1")
|
||||
data = self.get_urldata()
|
||||
match = re.search("__data = ([^<]+)</script>", data)
|
||||
if not match:
|
||||
logging.error("Can't find video info.")
|
||||
return episodes
|
||||
janson = json.loads(match.group(1))
|
||||
page = janson["cache"]["page"][list(janson["cache"]["page"].keys())[0]]
|
||||
item = page["entries"][0]["item"]
|
||||
if "season" in item:
|
||||
entries = item["season"]["episodes"]["items"]
|
||||
for i in entries:
|
||||
episodes.append("https://www.dr.dk/drtv{}".format(i["watchPath"]))
|
||||
|
||||
if config.get("all_last") != -1:
|
||||
episodes = episodes[: config.get("all_last")]
|
||||
else:
|
||||
path = base64.b64decode(encpath + "===").decode("latin1")
|
||||
|
||||
if "/view/" in path:
|
||||
continue
|
||||
|
||||
params = "offset=0&limit=1000"
|
||||
if newstyle:
|
||||
encparams = base64.b64encode(params.encode("latin1")).decode("latin1").rstrip("=")
|
||||
encpath = "{}_{}".format(encbasepath, encparams)
|
||||
else:
|
||||
path = "{}?{}".format(urlparse(path).path, params)
|
||||
encpath = base64.b64encode(path.encode("latin1")).decode("latin1").rstrip("=")
|
||||
|
||||
url = urljoin("https://www.dr.dk/tv/partial/", "{}/{}".format(enccomp, encpath))
|
||||
data = self.http.request("get", url).content.decode("latin1")
|
||||
|
||||
matches = re.findall(r'"program-link" href="([^"]+)">', data)
|
||||
episodes = [urljoin("https://www.dr.dk/", url) for url in matches]
|
||||
break
|
||||
|
||||
if not episodes:
|
||||
prefix = "/".join(urlparse(self.url).path.rstrip("/").split("/")[:-1])
|
||||
matches = re.findall(r'"program-link" href="([^"]+)">', self.get_urldata())
|
||||
episodes = [urljoin("https://www.dr.dk/", url) for url in matches if url.startswith(prefix)]
|
||||
|
||||
if config.get("all_last") != -1:
|
||||
episodes = episodes[: config.get("all_last")]
|
||||
else:
|
||||
episodes.reverse()
|
||||
|
||||
episodes.reverse()
|
||||
return episodes
|
||||
|
||||
def find_stream(self, config, resource):
|
||||
tempresource = resource["Data"][0]["Assets"]
|
||||
# To find the VideoResource, they have Images as well
|
||||
for resources in tempresource:
|
||||
if resources["Kind"] == "VideoResource":
|
||||
links = resources["Links"]
|
||||
break
|
||||
for i in links:
|
||||
if i["Target"] == "Ios" or i["Target"] == "HLS":
|
||||
streams = hlsparse(config, self.http.request("get", i["Uri"]), i["Uri"], output=self.output)
|
||||
for n in list(streams.keys()):
|
||||
yield streams[n]
|
||||
|
||||
def _decrypt(self, url):
|
||||
n = int(url[2:10], 16)
|
||||
iv_hex = url[10 + n :]
|
||||
data = binascii.a2b_hex(url[10 : 10 + n].encode("ascii"))
|
||||
key = hashlib.sha256(("%s:sRBzYNXBzkKgnjj8pGtkACch" % iv_hex).encode("utf-8")).digest()
|
||||
iv = bytes.fromhex(iv_hex)
|
||||
|
||||
backend = default_backend()
|
||||
cipher = Cipher(algorithms.AES(key), modes.CBC(iv), backend=backend)
|
||||
decryptor = cipher.decryptor()
|
||||
decrypted = decryptor.update(data)
|
||||
return decrypted[: -decrypted[-1]].decode("utf-8")
|
||||
|
Loading…
Reference in New Issue
Block a user