From 355c7bb512a493b0aceeeca3e382da3838a7de69 Mon Sep 17 00:00:00 2001 From: Johan Andersson Date: Wed, 15 Mar 2023 06:53:24 +0100 Subject: [PATCH] urplay: they added support for 1080p --- lib/svtplay_dl/service/urplay.py | 45 ++++++++++---------------------- 1 file changed, 14 insertions(+), 31 deletions(-) diff --git a/lib/svtplay_dl/service/urplay.py b/lib/svtplay_dl/service/urplay.py index e39691c..173f8f3 100644 --- a/lib/svtplay_dl/service/urplay.py +++ b/lib/svtplay_dl/service/urplay.py @@ -1,6 +1,5 @@ # ex:ts=4:sw=4:sts=4:et # -*- tab-width: 4; c-basic-offset: 4; indent-tabs-mode: nil -*- -import copy import json import logging import re @@ -8,10 +7,10 @@ from html import unescape from urllib.parse import urljoin from svtplay_dl.error import ServiceError +from svtplay_dl.fetcher.dash import dashparse from svtplay_dl.fetcher.hls import hlsparse from svtplay_dl.service import OpenGraphThumbMixin from svtplay_dl.service import Service -from svtplay_dl.subtitle import subtitle_probe from svtplay_dl.utils.http import download_thumbnails @@ -28,39 +27,23 @@ class Urplay(Service, OpenGraphThumbMixin): data = unescape(match.group(1)) jsondata = json.loads(data) - res = self.http.get("https://streaming-loadbalancer.ur.se/loadbalancer.json") - loadbalancer = res.json()["redirect"] + vid = jsondata["props"]["pageProps"]["program"]["id"] jsondata = jsondata["props"]["pageProps"]["program"] + res = self.http.get(f"https://media-api.urplay.se/config-streaming/v1/urplay/sources/{vid}") + + if "dash" in res.json()["sources"]: + yield from dashparse( + self.config, + self.http.request("get", res.json()["sources"]["dash"]), + res.json()["sources"]["dash"], + output=self.output, + ) + if "hls" in res.json()["sources"]: + yield from hlsparse(self.config, self.http.request("get", res.json()["sources"]["hls"]), res.json()["sources"]["hls"], output=self.output) + self.outputfilename(jsondata, urldata) - for streaminfo in jsondata["streamingInfo"].keys(): - stream = jsondata["streamingInfo"][streaminfo] - - if streaminfo == "raw": - if "sd" in stream: - url = f"https://{loadbalancer}/{stream['sd']['location']}playlist.m3u8" - yield from hlsparse(self.config, self.http.request("get", url), url, output=self.output) - if "hd" in stream: - url = f"https://{loadbalancer}/{stream['hd']['location']}playlist.m3u8" - yield from hlsparse(self.config, self.http.request("get", url), url, output=self.output) - if "m4A" in stream: - url = f"https://{loadbalancer}/{stream['m4A']['location']}playlist.m3u8" - yield from hlsparse(self.config, self.http.request("get", url), url, output=self.output) - if not (self.config.get("get_all_subtitles")) and streaminfo == "sweComplete": - yield from subtitle_probe(copy.copy(self.config), stream["tt"]["location"].replace(".tt", ".vtt"), output=copy.copy(self.output)) - - if self.config.get("get_all_subtitles") and "tt" in stream: - label = stream["tt"]["language"] - if stream["tt"]["scope"] != "complete": - label = f"{label}-{stream['tt']['scope']}" - yield from subtitle_probe( - copy.copy(self.config), - stream["tt"]["location"].replace(".tt", ".vtt"), - subfix=label, - output=copy.copy(self.output), - ) - def find_all_episodes(self, config): episodes = []