mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-12-24 05:00:10 +09:00
179 lines
8.0 KiB
Python
179 lines
8.0 KiB
Python
# encoding: utf-8
|
||
from __future__ import unicode_literals
|
||
|
||
import re, base64, zlib
|
||
from hashlib import sha1
|
||
from math import pow, sqrt, floor
|
||
from .common import InfoExtractor
|
||
from ..utils import (
|
||
ExtractorError,
|
||
compat_urllib_parse,
|
||
compat_urllib_request,
|
||
bytes_to_intlist,
|
||
intlist_to_bytes,
|
||
unified_strdate,
|
||
clean_html,
|
||
)
|
||
from ..aes import (
|
||
aes_cbc_decrypt,
|
||
inc,
|
||
)
|
||
|
||
class CrunchyrollIE(InfoExtractor):
|
||
_VALID_URL = r'(?:https?://)?(?:(?P<prefix>www|m)\.)?(?P<url>crunchyroll\.com/(?:[^/]*/[^/?&]*?|media/\?id=)(?P<video_id>[0-9]+))(?:[/?&]|$)'
|
||
_TESTS = [{
|
||
'url': 'http://www.crunchyroll.com/wanna-be-the-strongest-in-the-world/episode-1-an-idol-wrestler-is-born-645513',
|
||
'file': '645513.flv',
|
||
#'md5': 'b1639fd6ddfaa43788c85f6d1dddd412',
|
||
'info_dict': {
|
||
'title': 'Wanna be the Strongest in the World Episode 1 – An Idol-Wrestler is Born!',
|
||
'description': 'md5:2d17137920c64f2f49981a7797d275ef',
|
||
'thumbnail': 'http://img1.ak.crunchyroll.com/i/spire1-tmb/20c6b5e10f1a47b10516877d3c039cae1380951166_full.jpg',
|
||
'uploader': 'Yomiuri Telecasting Corporation (YTV)',
|
||
'upload_date': '20131013',
|
||
},
|
||
'params': {
|
||
# rtmp
|
||
'skip_download': True,
|
||
},
|
||
}]
|
||
|
||
_FORMAT_IDS = {
|
||
'360': ('60', '106'),
|
||
'480': ('61', '106'),
|
||
'720': ('62', '106'),
|
||
'1080': ('80', '108'),
|
||
}
|
||
|
||
def _decrypt_subtitles(self, data, iv, id):
|
||
data = bytes_to_intlist(data)
|
||
iv = bytes_to_intlist(iv)
|
||
id = int(id)
|
||
|
||
def obfuscate_key_aux(count, modulo, start):
|
||
output = list(start)
|
||
for _ in range(count):
|
||
output.append(output[-1] + output[-2])
|
||
# cut off start values
|
||
output = output[2:]
|
||
output = list(map(lambda x: x % modulo + 33, output))
|
||
return output
|
||
|
||
def obfuscate_key(key):
|
||
num1 = int(floor(pow(2, 25) * sqrt(6.9)))
|
||
num2 = (num1 ^ key) << 5
|
||
num3 = key ^ num1
|
||
num4 = num3 ^ (num3 >> 3) ^ num2
|
||
prefix = intlist_to_bytes(obfuscate_key_aux(20, 97, (1, 2)))
|
||
shaHash = bytes_to_intlist(sha1(prefix + str(num4).encode('ascii')).digest())
|
||
# Extend 160 Bit hash to 256 Bit
|
||
return shaHash + [0] * 12
|
||
|
||
key = obfuscate_key(id)
|
||
class Counter:
|
||
__value = iv
|
||
def next_value(self):
|
||
temp = self.__value
|
||
self.__value = inc(self.__value)
|
||
return temp
|
||
decrypted_data = intlist_to_bytes(aes_cbc_decrypt(data, key, iv))
|
||
return zlib.decompress(decrypted_data)
|
||
|
||
def _convert_subtitles_to_srt(self, subtitles):
|
||
i=1
|
||
output = ''
|
||
for start, end, text in re.findall(r'<event [^>]*?start="([^"]+)" [^>]*?end="([^"]+)" [^>]*?text="([^"]+)"[^>]*?>', subtitles):
|
||
start = start.replace('.', ',')
|
||
end = end.replace('.', ',')
|
||
text = clean_html(text)
|
||
text = text.replace('\\N', '\n')
|
||
if not text:
|
||
continue
|
||
output += '%d\n%s --> %s\n%s\n\n' % (i, start, end, text)
|
||
i+=1
|
||
return output
|
||
|
||
def _real_extract(self,url):
|
||
mobj = re.match(self._VALID_URL, url)
|
||
video_id = mobj.group('video_id')
|
||
|
||
if mobj.group('prefix') == 'm':
|
||
mobile_webpage = self._download_webpage(url, video_id, 'Downloading mobile webpage')
|
||
webpage_url = self._search_regex(r'<link rel="canonical" href="([^"]+)" />', mobile_webpage, 'webpage_url')
|
||
else:
|
||
webpage_url = 'http://www.' + mobj.group('url')
|
||
|
||
webpage = self._download_webpage(webpage_url, video_id, 'Downloading webpage')
|
||
note_m = self._html_search_regex(r'<div class="showmedia-trailer-notice">(.+?)</div>', webpage, 'trailer-notice', default='')
|
||
if note_m:
|
||
raise ExtractorError(note_m)
|
||
|
||
video_title = self._html_search_regex(r'<h1[^>]*>(.+?)</h1>', webpage, 'video_title', flags=re.DOTALL)
|
||
video_title = re.sub(r' {2,}', ' ', video_title)
|
||
video_description = self._html_search_regex(r'"description":"([^"]+)', webpage, 'video_description', default='')
|
||
if not video_description:
|
||
video_description = None
|
||
video_upload_date = self._html_search_regex(r'<div>Availability for free users:(.+?)</div>', webpage, 'video_upload_date', fatal=False, flags=re.DOTALL)
|
||
if video_upload_date:
|
||
video_upload_date = unified_strdate(video_upload_date)
|
||
video_uploader = self._html_search_regex(r'<div>\s*Publisher:(.+?)</div>', webpage, 'video_uploader', fatal=False, flags=re.DOTALL)
|
||
|
||
playerdata_url = compat_urllib_parse.unquote(self._html_search_regex(r'"config_url":"([^"]+)', webpage, 'playerdata_url'))
|
||
playerdata_req = compat_urllib_request.Request(playerdata_url)
|
||
playerdata_req.data = compat_urllib_parse.urlencode({'current_page': webpage_url})
|
||
playerdata_req.add_header('Content-Type', 'application/x-www-form-urlencoded')
|
||
playerdata = self._download_webpage(playerdata_req, video_id, note='Downloading media info')
|
||
|
||
stream_id = self._search_regex(r'<media_id>([^<]+)', playerdata, 'stream_id')
|
||
video_thumbnail = self._search_regex(r'<episode_image_url>([^<]+)', playerdata, 'thumbnail', fatal=False)
|
||
|
||
formats = []
|
||
for fmt in re.findall(r'\?p([0-9]{3,4})=1', webpage):
|
||
stream_quality, stream_format = self._FORMAT_IDS[fmt]
|
||
video_format = fmt+'p'
|
||
streamdata_req = compat_urllib_request.Request('http://www.crunchyroll.com/xml/')
|
||
# urlencode doesn't work!
|
||
streamdata_req.data = 'req=RpcApiVideoEncode%5FGetStreamInfo&video%5Fencode%5Fquality='+stream_quality+'&media%5Fid='+stream_id+'&video%5Fformat='+stream_format
|
||
streamdata_req.add_header('Content-Type', 'application/x-www-form-urlencoded')
|
||
streamdata_req.add_header('Content-Length', str(len(streamdata_req.data)))
|
||
streamdata = self._download_webpage(streamdata_req, video_id, note='Downloading media info for '+video_format)
|
||
video_url = self._search_regex(r'<host>([^<]+)', streamdata, 'video_url')
|
||
video_play_path = self._search_regex(r'<file>([^<]+)', streamdata, 'video_play_path')
|
||
formats.append({
|
||
'url': video_url,
|
||
'play_path': video_play_path,
|
||
'ext': 'flv',
|
||
'format': video_format,
|
||
'format_id': video_format,
|
||
})
|
||
|
||
subtitles = {}
|
||
for sub_id, sub_name in re.findall(r'\?ssid=([0-9]+)" title="([^"]+)', webpage):
|
||
sub_page = self._download_webpage('http://www.crunchyroll.com/xml/?req=RpcApiSubtitle_GetXml&subtitle_script_id='+sub_id,\
|
||
video_id, note='Downloading subtitles for '+sub_name)
|
||
id = self._search_regex(r'id=\'([0-9]+)', sub_page, 'subtitle_id', fatal=False)
|
||
iv = self._search_regex(r'<iv>([^<]+)', sub_page, 'subtitle_iv', fatal=False)
|
||
data = self._search_regex(r'<data>([^<]+)', sub_page, 'subtitle_data', fatal=False)
|
||
if not id or not iv or not data:
|
||
continue
|
||
id = int(id)
|
||
iv = base64.b64decode(iv)
|
||
data = base64.b64decode(data)
|
||
|
||
subtitle = self._decrypt_subtitles(data, iv, id).decode('utf-8')
|
||
lang_code = self._search_regex(r'lang_code=\'([^\']+)', subtitle, 'subtitle_lang_code', fatal=False)
|
||
if not lang_code:
|
||
continue
|
||
subtitles[lang_code] = self._convert_subtitles_to_srt(subtitle)
|
||
|
||
return {
|
||
'id': video_id,
|
||
'title': video_title,
|
||
'description': video_description,
|
||
'thumbnail': video_thumbnail,
|
||
'uploader': video_uploader,
|
||
'upload_date': video_upload_date,
|
||
'subtitles': subtitles,
|
||
'formats': formats,
|
||
}
|