mirror of
https://github.com/spaam/svtplay-dl.git
synced 2024-11-28 06:04:17 +01:00
c781bd1bb8
svtplay: Adds support for --include-clips as well as support to just download one tab of episodes by using ?tab="tab" in url. This should fix https://github.com/spaam/svtplay-dl/issues/549, https://github.com/spaam/svtplay-dl/issues/514 and https://github.com/spaam/svtplay-dl/issues/480
245 lines
9.8 KiB
Python
245 lines
9.8 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 os
|
|
import xml.etree.ElementTree as ET
|
|
import copy
|
|
import json
|
|
import hashlib
|
|
|
|
from svtplay_dl.log import log
|
|
from svtplay_dl.service import Service, OpenGraphThumbMixin
|
|
from svtplay_dl.utils import filenamify, is_py2
|
|
from svtplay_dl.utils.urllib import urlparse, urljoin, parse_qs
|
|
from svtplay_dl.fetcher.hds import hdsparse
|
|
from svtplay_dl.fetcher.hls import hlsparse
|
|
from svtplay_dl.fetcher.dash import dashparse
|
|
from svtplay_dl.subtitle import subtitle
|
|
from svtplay_dl.error import ServiceError
|
|
|
|
|
|
class Svtplay(Service, OpenGraphThumbMixin):
|
|
supported_domains = ['svtplay.se', 'svt.se', 'beta.svtplay.se', 'svtflow.se']
|
|
|
|
def get(self):
|
|
parse = urlparse(self.url)
|
|
if parse.netloc == "www.svtplay.se" or parse.netloc == "svtplay.se":
|
|
if parse.path[:6] != "/video" and parse.path[:6] != "/klipp":
|
|
yield ServiceError("This mode is not supported anymore. need the url with the video")
|
|
return
|
|
|
|
match = re.search("__svtplay'] = ({.*});", self.get_urldata())
|
|
if not match:
|
|
yield ServiceError("Cant find video info.")
|
|
return
|
|
janson = json.loads(match.group(1))["videoTitlePage"]
|
|
|
|
if "live" in janson["video"]:
|
|
self.optionslive = janson["video"]["live"]
|
|
|
|
if self.options.output_auto:
|
|
self.options.service = "svtplay"
|
|
self.options.output = self.outputfilename(janson, self.options.output)
|
|
|
|
if self.exclude():
|
|
yield ServiceError("Excluding video")
|
|
return
|
|
|
|
if "programVersionId" in janson["video"]:
|
|
vid = janson["video"]["programVersionId"]
|
|
else:
|
|
vid = janson["video"]["id"]
|
|
res = self.http.get("http://api.svt.se/videoplayer-api/video/{}".format(vid))
|
|
janson = res.json()
|
|
|
|
if "subtitleReferences" in janson:
|
|
for i in janson["subtitleReferences"]:
|
|
if i["format"] == "websrt" and "url" in i:
|
|
yield subtitle(copy.copy(self.options), "wrst", i["url"])
|
|
|
|
if "videoReferences" in janson:
|
|
if len(janson["videoReferences"]) == 0:
|
|
yield ServiceError("Media doesn't have any associated videos (yet?)")
|
|
return
|
|
|
|
for i in janson["videoReferences"]:
|
|
parse = urlparse(i["url"])
|
|
query = parse_qs(parse.query)
|
|
if i["format"] == "hls":
|
|
streams = hlsparse(self.options, self.http.request("get", i["url"]), i["url"])
|
|
if streams:
|
|
for n in list(streams.keys()):
|
|
yield streams[n]
|
|
if "alt" in query and len(query["alt"]) > 0:
|
|
alt = self.http.get(query["alt"][0])
|
|
if alt:
|
|
streams = hlsparse(self.options, self.http.request("get", alt.request.url), alt.request.url)
|
|
if streams:
|
|
for n in list(streams.keys()):
|
|
yield streams[n]
|
|
if i["format"] == "hds":
|
|
match = re.search(r"\/se\/secure\/", i["url"])
|
|
if not match:
|
|
streams = hdsparse(self.options, self.http.request("get", i["url"], params={"hdcore": "3.7.0"}), i["url"])
|
|
if streams:
|
|
for n in list(streams.keys()):
|
|
yield streams[n]
|
|
if "alt" in query and len(query["alt"]) > 0:
|
|
alt = self.http.get(query["alt"][0])
|
|
if alt:
|
|
streams = hdsparse(self.options, self.http.request("get", alt.request.url, params={"hdcore": "3.7.0"}), alt.request.url)
|
|
if streams:
|
|
for n in list(streams.keys()):
|
|
yield streams[n]
|
|
if i["format"] == "dash264" or i["format"] == "dashhbbtv":
|
|
streams = dashparse(self.options, self.http.request("get", i["url"]), i["url"])
|
|
if streams:
|
|
for n in list(streams.keys()):
|
|
yield streams[n]
|
|
|
|
if "alt" in query and len(query["alt"]) > 0:
|
|
alt = self.http.get(query["alt"][0])
|
|
if alt:
|
|
streams = dashparse(self.options, self.http.request("get", alt.request.url), alt.request.url)
|
|
if streams:
|
|
for n in list(streams.keys()):
|
|
yield streams[n]
|
|
|
|
def _last_chance(self, videos, page, maxpage=2):
|
|
if page > maxpage:
|
|
return videos
|
|
|
|
res = self.http.get("http://www.svtplay.se/sista-chansen?sida=%s" % page)
|
|
match = re.search("__svtplay'] = ({.*});", res.text)
|
|
if not match:
|
|
return videos
|
|
|
|
dataj = json.loads(match.group(1))
|
|
pages = dataj["gridPage"]["pagination"]["totalPages"]
|
|
|
|
for i in dataj["gridPage"]["content"]:
|
|
videos.append(i["contentUrl"])
|
|
page += 1
|
|
self._last_chance(videos, page, pages)
|
|
return videos
|
|
|
|
def _genre(self, jansson):
|
|
videos = []
|
|
for i in jansson["clusterPage"]["clips"]:
|
|
videos.append(i["contentUrl"])
|
|
return videos
|
|
|
|
def find_all_episodes(self, options):
|
|
parse = urlparse(self._url)
|
|
|
|
if len(parse.path) > 7 and parse.path[-7:] == "rss.xml":
|
|
match = self.url
|
|
else:
|
|
match = re.search(r'<link rel="alternate" type="application/rss\+xml" [^>]*href="([^"]+)"',
|
|
self.get_urldata())
|
|
if match:
|
|
match = match.group(1)
|
|
|
|
if match is None:
|
|
videos = []
|
|
tab = None
|
|
match = re.search("__svtplay'] = ({.*});", self.get_urldata())
|
|
if re.search("sista-chansen", parse.path):
|
|
videos = self._last_chance(videos, 1)
|
|
elif not match:
|
|
log.error("Couldn't retrieve episode list")
|
|
return
|
|
else:
|
|
dataj = json.loads(match.group(1))
|
|
if re.search("/genre", parse.path):
|
|
videos = self._genre(dataj)
|
|
else:
|
|
if parse.query:
|
|
match = re.search("tab=(.+)",parse.query)
|
|
if(match):
|
|
tab = match.group(1)
|
|
|
|
items = dataj["videoTitlePage"]["relatedVideosTabs"]
|
|
for i in items:
|
|
if tab:
|
|
if i["slug"] == tab:
|
|
for n in i["videos"]:
|
|
parse = urlparse(n["contentUrl"])
|
|
if parse.path not in videos:
|
|
videos.append(parse.path)
|
|
|
|
else:
|
|
if "sasong" in i["slug"] or "senast" in i["slug"]:
|
|
for n in i["videos"]:
|
|
parse = urlparse(n["contentUrl"])
|
|
if parse.path not in videos:
|
|
videos.append(parse.path)
|
|
|
|
if self.options.include_clips:
|
|
if i["slug"] == "klipp":
|
|
for n in i["videos"]:
|
|
parse = urlparse(n["contentUrl"])
|
|
if parse.path not in videos:
|
|
videos.append(parse.path)
|
|
|
|
episodes = [urljoin("http://www.svtplay.se", x) for x in videos]
|
|
else:
|
|
data = self.http.request("get", match).content
|
|
xml = ET.XML(data)
|
|
|
|
episodes = [x.text for x in xml.findall(".//item/link")]
|
|
episodes_new = []
|
|
n = 1
|
|
for i in episodes:
|
|
episodes_new.append(i)
|
|
if n == options.all_last:
|
|
break
|
|
n += 1
|
|
return sorted(episodes_new)
|
|
|
|
def outputfilename(self, data, filename):
|
|
directory = os.path.dirname(filename)
|
|
name = None
|
|
if data["video"]["programTitle"]:
|
|
name = filenamify(data["video"]["programTitle"])
|
|
other = filenamify(data["video"]["title"])
|
|
|
|
if "programVersionId" in data["video"]:
|
|
vid = str(data["video"]["programVersionId"])
|
|
else:
|
|
vid = str(data["video"]["id"])
|
|
if is_py2:
|
|
id = hashlib.sha256(vid).hexdigest()[:7]
|
|
else:
|
|
id = hashlib.sha256(vid.encode("utf-8")).hexdigest()[:7]
|
|
|
|
if name == other:
|
|
other = None
|
|
elif name == None:
|
|
name = other
|
|
other = None
|
|
season = self.seasoninfo(data)
|
|
title = name
|
|
if season:
|
|
title += ".%s" % season
|
|
if other:
|
|
title += ".%s" % other
|
|
title += "-%s-svtplay" % id
|
|
title = filenamify(title)
|
|
if len(directory):
|
|
output = os.path.join(directory, title)
|
|
else:
|
|
output = title
|
|
return output
|
|
|
|
def seasoninfo(self, data):
|
|
if "season" in data["video"] and data["video"]["season"]:
|
|
season = "{:02d}".format(data["video"]["season"])
|
|
episode = "{:02d}".format(data["video"]["episodeNumber"])
|
|
if int(season) == 0 and int(episode) == 0:
|
|
return None
|
|
return "S%sE%s" % (season, episode)
|
|
else:
|
|
return None
|