mirror of
https://github.com/spaam/svtplay-dl.git
synced 2024-11-24 12:15:40 +01:00
99 lines
3.6 KiB
Python
99 lines
3.6 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 copy
|
|
import xml.etree.ElementTree as ET
|
|
import time
|
|
|
|
|
|
from svtplay_dl.output import progress_stream, output, ETA, progressbar
|
|
from svtplay_dl.utils.urllib import urljoin
|
|
from svtplay_dl.error import UIException, ServiceError
|
|
from svtplay_dl.fetcher import VideoRetriever
|
|
|
|
|
|
class DASHException(UIException):
|
|
def __init__(self, url, message):
|
|
self.url = url
|
|
super(DASHException, self).__init__(message)
|
|
|
|
|
|
class LiveDASHException(DASHException):
|
|
def __init__(self, url):
|
|
super(LiveDASHException, self).__init__(
|
|
url, "This is a live DASH stream, and they are not supported.")
|
|
|
|
|
|
def dashparse(options, res, url):
|
|
streams = {}
|
|
|
|
if res.status_code == 403 or res.status_code == 404:
|
|
streams[0] = ServiceError("Can't read DASH playlist. {0}".format(res.status_code))
|
|
return streams
|
|
xml = ET.XML(res.text)
|
|
baseurl = urljoin(url, xml.find("{urn:mpeg:dash:schema:mpd:2011}BaseURL").text)
|
|
videofiles = xml.findall(".//{urn:mpeg:dash:schema:mpd:2011}AdaptationSet[@contentType='video']/{urn:mpeg:dash:schema:mpd:2011}Representation")
|
|
|
|
audiofiles = xml.findall(".//{urn:mpeg:dash:schema:mpd:2011}AdaptationSet[@contentType='audio']/{urn:mpeg:dash:schema:mpd:2011}Representation")
|
|
for i in audiofiles:
|
|
audiourl = urljoin(baseurl, i.find("{urn:mpeg:dash:schema:mpd:2011}BaseURL").text)
|
|
audiobitrate = float(i.attrib["bandwidth"]) / 1000
|
|
for n in videofiles:
|
|
bitrate = float(n.attrib["bandwidth"])/1000 + audiobitrate
|
|
videourl = urljoin(baseurl, n.find("{urn:mpeg:dash:schema:mpd:2011}BaseURL").text)
|
|
options.other = "mp4"
|
|
streams[int(bitrate)] = DASH(copy.copy(options), videourl, bitrate, cookies=res.cookies, audio=audiourl)
|
|
|
|
return streams
|
|
|
|
|
|
class DASH(VideoRetriever):
|
|
def name(self):
|
|
return "dash"
|
|
|
|
def download(self):
|
|
if self.options.live and not self.options.force:
|
|
raise LiveDASHException(self.url)
|
|
|
|
if self.audio:
|
|
self._download(self.audio, audio=True)
|
|
self._download(self.url)
|
|
|
|
|
|
def _download(self, url, audio=False):
|
|
cookies = self.kwargs["cookies"]
|
|
data = self.http.request("get", url, cookies=cookies, headers={'Range': 'bytes=0-8192'})
|
|
try:
|
|
total_size = data.headers['Content-Range']
|
|
total_size = total_size[total_size.find("/")+1:]
|
|
except KeyError:
|
|
total_size = 0
|
|
total_size = int(total_size)
|
|
bytes_so_far = 8192
|
|
if audio:
|
|
file_d = output(copy.copy(self.options), "m4a")
|
|
else:
|
|
file_d = output(self.options, self.options.other)
|
|
if hasattr(file_d, "read") is False:
|
|
return
|
|
file_d.write(data.content)
|
|
eta = ETA(total_size)
|
|
while bytes_so_far < total_size:
|
|
old = bytes_so_far + 1
|
|
bytes_so_far = old + 1000000
|
|
if bytes_so_far > total_size:
|
|
bytes_so_far = total_size
|
|
|
|
bytes_range = "bytes=%s-%s" % (old, bytes_so_far)
|
|
|
|
data = self.http.request("get", url, cookies=cookies, headers={'Range': bytes_range})
|
|
file_d.write(data.content)
|
|
if self.options.output != "-" and not self.options.silent:
|
|
eta.update(old)
|
|
progressbar(total_size, old, ''.join(["ETA: ", str(eta)]))
|
|
|
|
if self.options.output != "-":
|
|
file_d.close()
|
|
progress_stream.write('\n')
|
|
self.finished = True
|