mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-15 18:34:22 +00:00
[youku] PEP8
This commit is contained in:
parent
8268e94cd4
commit
aed473ccf9
@ -11,6 +11,7 @@ from ..compat import compat_urllib_parse
|
|||||||
|
|
||||||
bytes_is_str = (bytes == str) # for compatible
|
bytes_is_str = (bytes == str) # for compatible
|
||||||
|
|
||||||
|
|
||||||
class YoukuIE(InfoExtractor):
|
class YoukuIE(InfoExtractor):
|
||||||
IE_NAME = 'youku'
|
IE_NAME = 'youku'
|
||||||
_VALID_URL = r'''(?x)
|
_VALID_URL = r'''(?x)
|
||||||
@ -21,13 +22,13 @@ class YoukuIE(InfoExtractor):
|
|||||||
'''
|
'''
|
||||||
|
|
||||||
_TEST = {
|
_TEST = {
|
||||||
'url': 'http://v.youku.com/v_show/id_XMTc1ODE5Njcy.html',
|
'url': 'http://v.youku.com/v_show/id_XMTc1ODE5Njcy.html',
|
||||||
'md5': '5f3af4192eabacc4501508d54a8cabd7',
|
'md5': '5f3af4192eabacc4501508d54a8cabd7',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': 'XMTc1ODE5Njcy',
|
'id': 'XMTc1ODE5Njcy',
|
||||||
'title': '★Smile﹗♡ Git Fresh -Booty Music舞蹈.',
|
'title': '★Smile﹗♡ Git Fresh -Booty Music舞蹈.',
|
||||||
'ext': 'flv'
|
'ext': 'flv'
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
def construct_video_urls(self, data1, data2):
|
def construct_video_urls(self, data1, data2):
|
||||||
@ -36,7 +37,7 @@ class YoukuIE(InfoExtractor):
|
|||||||
ls = list(range(256))
|
ls = list(range(256))
|
||||||
t = 0
|
t = 0
|
||||||
for i in range(256):
|
for i in range(256):
|
||||||
t = (t + ls[i] + ord(s1[i%len(s1)])) % 256
|
t = (t + ls[i] + ord(s1[i % len(s1)])) % 256
|
||||||
ls[i], ls[t] = ls[t], ls[i]
|
ls[i], ls[t] = ls[t], ls[i]
|
||||||
s = '' if not bytes_is_str else b''
|
s = '' if not bytes_is_str else b''
|
||||||
x, y = 0, 0
|
x, y = 0, 0
|
||||||
@ -45,16 +46,16 @@ class YoukuIE(InfoExtractor):
|
|||||||
x = (x + ls[y]) % 256
|
x = (x + ls[y]) % 256
|
||||||
ls[x], ls[y] = ls[y], ls[x]
|
ls[x], ls[y] = ls[y], ls[x]
|
||||||
if isinstance(s2[i], int):
|
if isinstance(s2[i], int):
|
||||||
s += chr(s2[i] ^ ls[(ls[x]+ls[y]) % 256])
|
s += chr(s2[i] ^ ls[(ls[x] + ls[y]) % 256])
|
||||||
else:
|
else:
|
||||||
s += chr(ord(s2[i]) ^ ls[(ls[x]+ls[y]) % 256])
|
s += chr(ord(s2[i]) ^ ls[(ls[x] + ls[y]) % 256])
|
||||||
return s
|
return s
|
||||||
|
|
||||||
sid, token = yk_t(
|
sid, token = yk_t(
|
||||||
'becaf9be',
|
'becaf9be',
|
||||||
base64.b64decode(bytes(data2['ep'], 'ascii')) \
|
base64.b64decode(bytes(data2['ep'], 'ascii'))
|
||||||
if not bytes_is_str \
|
if not bytes_is_str
|
||||||
else base64.b64decode(data2['ep'])
|
else base64.b64decode(data2['ep'])
|
||||||
).split('_')
|
).split('_')
|
||||||
|
|
||||||
# get oip
|
# get oip
|
||||||
@ -89,13 +90,13 @@ class YoukuIE(InfoExtractor):
|
|||||||
fileid = get_fileid(format, n)
|
fileid = get_fileid(format, n)
|
||||||
ep_t = yk_t(
|
ep_t = yk_t(
|
||||||
'bf7e5f01',
|
'bf7e5f01',
|
||||||
bytes('%s_%s_%s' % (sid, fileid, token), 'ascii') \
|
bytes('%s_%s_%s' % (sid, fileid, token), 'ascii')
|
||||||
if not bytes_is_str \
|
if not bytes_is_str
|
||||||
else ('%s_%s_%s' % (sid, fileid, token))
|
else ('%s_%s_%s' % (sid, fileid, token))
|
||||||
)
|
)
|
||||||
ep = base64.b64encode(
|
ep = base64.b64encode(
|
||||||
bytes(ep_t, 'latin') \
|
bytes(ep_t, 'latin')
|
||||||
if not bytes_is_str \
|
if not bytes_is_str
|
||||||
else ep_t
|
else ep_t
|
||||||
).decode()
|
).decode()
|
||||||
return ep
|
return ep
|
||||||
@ -121,9 +122,9 @@ class YoukuIE(InfoExtractor):
|
|||||||
video_url = \
|
video_url = \
|
||||||
'http://k.youku.com/player/getFlvPath/' + \
|
'http://k.youku.com/player/getFlvPath/' + \
|
||||||
'sid/' + sid + \
|
'sid/' + sid + \
|
||||||
'_' + str(int(n)+1).zfill(2) + \
|
'_' + str(int(n) + 1).zfill(2) + \
|
||||||
'/st/' + self.parse_ext_l(format) + \
|
'/st/' + self.parse_ext_l(format) + \
|
||||||
'/fileid/' + get_fileid(format, n) + '?' + \
|
'/fileid/' + get_fileid(format, n) + '?' + \
|
||||||
compat_urllib_parse.urlencode(param)
|
compat_urllib_parse.urlencode(param)
|
||||||
video_urls.append(video_url)
|
video_urls.append(video_url)
|
||||||
video_urls_dict[format] = video_urls
|
video_urls_dict[format] = video_urls
|
||||||
@ -132,34 +133,34 @@ class YoukuIE(InfoExtractor):
|
|||||||
|
|
||||||
def get_hd(self, fm):
|
def get_hd(self, fm):
|
||||||
hd_id_dict = {
|
hd_id_dict = {
|
||||||
'flv' : '0',
|
'flv': '0',
|
||||||
'mp4' : '1',
|
'mp4': '1',
|
||||||
'hd2' : '2',
|
'hd2': '2',
|
||||||
'hd3' : '3',
|
'hd3': '3',
|
||||||
'3gp' : '0',
|
'3gp': '0',
|
||||||
'3gphd' : '1'
|
'3gphd': '1'
|
||||||
}
|
}
|
||||||
return hd_id_dict[fm]
|
return hd_id_dict[fm]
|
||||||
|
|
||||||
def parse_ext_l(self, fm):
|
def parse_ext_l(self, fm):
|
||||||
ext_dict = {
|
ext_dict = {
|
||||||
'flv' : 'flv',
|
'flv': 'flv',
|
||||||
'mp4' : 'mp4',
|
'mp4': 'mp4',
|
||||||
'hd2' : 'flv',
|
'hd2': 'flv',
|
||||||
'hd3' : 'flv',
|
'hd3': 'flv',
|
||||||
'3gp' : 'flv',
|
'3gp': 'flv',
|
||||||
'3gphd' : 'mp4'
|
'3gphd': 'mp4'
|
||||||
}
|
}
|
||||||
return ext_dict[fm]
|
return ext_dict[fm]
|
||||||
|
|
||||||
def get_format_name(self, fm):
|
def get_format_name(self, fm):
|
||||||
_dict = {
|
_dict = {
|
||||||
'3gp' : 'h6',
|
'3gp': 'h6',
|
||||||
'3gphd' : 'h5',
|
'3gphd': 'h5',
|
||||||
'flv' : 'h4',
|
'flv': 'h4',
|
||||||
'mp4' : 'h3',
|
'mp4': 'h3',
|
||||||
'hd2' : 'h2',
|
'hd2': 'h2',
|
||||||
'hd3' : 'h1'
|
'hd3': 'h1'
|
||||||
}
|
}
|
||||||
return _dict[fm]
|
return _dict[fm]
|
||||||
|
|
||||||
@ -194,10 +195,9 @@ class YoukuIE(InfoExtractor):
|
|||||||
# construct info
|
# construct info
|
||||||
entries = []
|
entries = []
|
||||||
for fm in data1['streamtypes']:
|
for fm in data1['streamtypes']:
|
||||||
#formats = []
|
|
||||||
video_urls = video_urls_dict[fm]
|
video_urls = video_urls_dict[fm]
|
||||||
for i in range(len(video_urls)):
|
for i in range(len(video_urls)):
|
||||||
if len(entries) < i+1:
|
if len(entries) < i + 1:
|
||||||
entries.append({'formats': []})
|
entries.append({'formats': []})
|
||||||
entries[i]['formats'].append(
|
entries[i]['formats'].append(
|
||||||
{
|
{
|
||||||
@ -211,7 +211,7 @@ class YoukuIE(InfoExtractor):
|
|||||||
for i in range(len(entries)):
|
for i in range(len(entries)):
|
||||||
entries[i].update(
|
entries[i].update(
|
||||||
{
|
{
|
||||||
'id': '_part%d' % (i+1),
|
'id': '_part%d' % (i + 1),
|
||||||
'title': title,
|
'title': title,
|
||||||
}
|
}
|
||||||
)
|
)
|
||||||
|
Loading…
Reference in New Issue
Block a user