1
0
mirror of https://github.com/spaam/svtplay-dl.git synced 2024-11-24 04:05:39 +01:00

Cosmetics: whitespaces

This commit is contained in:
Johan Andersson 2013-01-12 23:07:41 +01:00
parent 1209e37616
commit 7fe7d5f321

View File

@ -235,7 +235,7 @@ def parsem3u(data):
streaminfo = {}
globdata = {}
data = data.replace("\r","\n")
data = data.replace("\r", "\n")
for l in data.split("\n")[1:]:
if not l:
continue
@ -259,11 +259,11 @@ def parsem3u(data):
return globdata, files
def decode_f4f(fragID, fragData ):
def decode_f4f(fragID, fragData):
start = fragData.find("mdat") + 4
if (fragID > 1):
for dummy in range( 2 ):
tagLen, = struct.unpack_from( ">L", fragData, start )
for dummy in range(2):
tagLen, = struct.unpack_from(">L", fragData, start)
tagLen &= 0x00ffffff
start += tagLen + 11 + 4
return start
@ -376,10 +376,10 @@ def download_hds(options, url, swf=None):
file_d.write(data[number:])
now = time.time()
dt = now - start
et = dt / ( i + 1 ) * total
et = dt / (i + 1) * total
rt = et - dt
td = timedelta( seconds = int( rt ) )
estimated = "Estimated Remaining: " + str( td )
td = timedelta(seconds = int(rt))
estimated = "Estimated Remaining: " + str(td)
i += 1
if options.output != "-":
@ -448,10 +448,10 @@ def download_hls(options, url, baseurl=None):
file_d.write(data)
now = time.time()
dt = now - start
et = dt / ( n + 1 ) * len( files )
et = dt / (n + 1) * len(files)
rt = et - dt
td = timedelta( seconds = int( rt ) )
estimated = "Estimated Remaining: " + str( td )
td = timedelta(seconds = int(rt))
estimated = "Estimated Remaining: " + str(td)
n += 1
if options.output != "-":
@ -469,7 +469,7 @@ def download_http(options, url):
if extension:
options.output = options.output + extension.group(1)
log.info("Outfile: %s", options.output)
file_d = open(options.output,"wb")
file_d = open(options.output, "wb")
else:
file_d = sys.stdout
@ -755,7 +755,7 @@ class Tv4play():
sa = list(ss.getiterator("item"))
else:
sa = list(ss.iter("item"))
if xml.find("live").text:
if xml.find("live").text != "false":
options.live = True
@ -831,47 +831,47 @@ class Svtplay():
else:
download_http(options, test["url"])
class Nrk ( object ) :
def get ( self, options, url ) :
data = get_http_data( url )
match = re.search( r'data-media="(.*manifest.f4m)"', data )
manifest_url = match.group( 1 )
class Nrk(object):
def get(self, options, url):
data = get_http_data(url)
match = re.search(r'data-media="(.*manifest.f4m)"', data)
manifest_url = match.group(1)
if options.hls:
manifest_url = manifest_url.replace( "/z/", "/i/" ).replace( "manifest.f4m", "master.m3u8" )
manifest_url = manifest_url.replace("/z/", "/i/").replace("manifest.f4m", "master.m3u8")
download_hls(options, manifest_url)
else:
manifest_url = "%s?hdcore=2.8.0&g=hejsan" % manifest_url
download_hds(options, manifest_url)
class Dr ( object ) :
def get ( self, options, url ) :
data = get_http_data( url )
match = re.search( r'resource:[ ]*"([^"]*)",', data )
resource_url = match.group( 1 )
resource_data = get_http_data( resource_url )
resource = json.loads( resource_data )
class Dr(object):
def get(self, options, url):
data = get_http_data(url)
match = re.search(r'resource:[ ]*"([^"]*)",', data)
resource_url = match.group(1)
resource_data = get_http_data(resource_url)
resource = json.loads(resource_data)
streams = {}
for stream in resource['links'] :
for stream in resource['links']:
streams[stream['bitrateKbps']] = stream['uri']
if len(streams) == 1:
uri = streams[streams.keys()[0]]
else:
uri = select_quality(options, streams)
# need -v ?
options.other = "-v -y '" + uri.replace( "rtmp://vod.dr.dk/cms/", "" ) + "'"
download_rtmp( options, uri )
options.other = "-v -y '" + uri.replace("rtmp://vod.dr.dk/cms/", "") + "'"
download_rtmp(options, uri)
class Ruv ( object ) :
def get ( self, options, url ) :
data = get_http_data( url )
match = re.search( r'(http://load.cache.is/vodruv.*)"', data )
js_url = match.group( 1 )
js = get_http_data( js_url )
tengipunktur = js.split( '"' )[1]
match = re.search( r"http.*tengipunktur [+] '([:]1935.*)'", data )
m3u8_url = "http://" + tengipunktur + match.group( 1 )
base_url = m3u8_url.rsplit( "/", 1 )[0]
download_hls( options, m3u8_url, base_url )
class Ruv(object):
def get(self, options, url):
data = get_http_data(url)
match = re.search(r'(http://load.cache.is/vodruv.*)"', data)
js_url = match.group(1)
js = get_http_data(js_url)
tengipunktur = js.split('"')[1]
match = re.search(r"http.*tengipunktur [+] '([:]1935.*)'", data)
m3u8_url = "http://" + tengipunktur + match.group(1)
base_url = m3u8_url.rsplit("/", 1)[0]
download_hls(options, m3u8_url, base_url)
def progressbar(total, pos, msg=""):
"""
@ -1023,7 +1023,6 @@ def get_media(url, options):
url = "http://www.kanal5play.se/api/getVideo?format=FLASH&videoId=%s" % match.group(1)
Kanal5().get(options, url)
elif re.findall("kanal9play", url):
data = get_http_data(url)
match = re.search("@videoPlayer\" value=\"(.*)\"", data)
@ -1122,12 +1121,12 @@ def get_media(url, options):
elif re.findall("svtplay.se", url):
Svtplay().get(options, url)
elif re.findall( "tv.nrk.no", url ) :
Nrk().get( options, url )
elif re.findall( "dr.dk/TV", url ) :
Dr().get( options, url )
elif re.findall( "ruv.is", url ) :
Ruv().get( options, url )
elif re.findall("tv.nrk.no", url):
Nrk().get(options, url)
elif re.findall("dr.dk/TV", url):
Dr().get(options, url)
elif re.findall("ruv.is", url):
Ruv().get(options, url)
else:
log.error("That site is not supported. Make a ticket or send a message")
sys.exit(2)
@ -1139,7 +1138,7 @@ def setup_log(silent):
else:
stream = sys.stdout
level = logging.INFO
fmt = logging.Formatter('%(levelname)s %(message)s')
hdlr = logging.StreamHandler(stream)
hdlr.setFormatter(fmt)