mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-24 19:57:23 +01:00
123 lines
4.5 KiB
Python
123 lines
4.5 KiB
Python
# encoding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
compat_urllib_parse,
|
|
compat_urllib_request,
|
|
compat_urlparse,
|
|
compat_str,
|
|
|
|
ExtractorError,
|
|
unified_strdate,
|
|
)
|
|
|
|
|
|
class NiconicoIE(InfoExtractor):
|
|
IE_NAME = 'niconico'
|
|
IE_DESC = 'ニコニコ動画'
|
|
|
|
_TEST = {
|
|
'url': 'http://www.nicovideo.jp/watch/sm22312215',
|
|
'md5': 'd1a75c0823e2f629128c43e1212760f9',
|
|
'info_dict': {
|
|
'id': 'sm22312215',
|
|
'ext': 'mp4',
|
|
'title': 'Big Buck Bunny',
|
|
'uploader': 'takuya0301',
|
|
'uploader_id': '2698420',
|
|
'upload_date': '20131123',
|
|
'description': '(c) copyright 2008, Blender Foundation / www.bigbuckbunny.org',
|
|
},
|
|
'params': {
|
|
'username': 'ydl.niconico@gmail.com',
|
|
'password': 'youtube-dl',
|
|
},
|
|
}
|
|
|
|
_VALID_URL = r'^https?://(?:www\.|secure\.)?nicovideo\.jp/watch/([a-z][a-z][0-9]+)(?:.*)$'
|
|
_NETRC_MACHINE = 'niconico'
|
|
|
|
def _real_initialize(self):
|
|
self._login()
|
|
|
|
def _login(self):
|
|
(username, password) = self._get_login_info()
|
|
if username is None:
|
|
# Login is required
|
|
raise ExtractorError('No login info available, needed for using %s.' % self.IE_NAME, expected=True)
|
|
|
|
# Log in
|
|
login_form_strs = {
|
|
'mail': username,
|
|
'password': password,
|
|
}
|
|
# Convert to UTF-8 *before* urlencode because Python 2.x's urlencode
|
|
# chokes on unicode
|
|
login_form = dict((k.encode('utf-8'), v.encode('utf-8')) for k, v in login_form_strs.items())
|
|
login_data = compat_urllib_parse.urlencode(login_form).encode('utf-8')
|
|
request = compat_urllib_request.Request(
|
|
'https://secure.nicovideo.jp/secure/login', login_data)
|
|
login_results = self._download_webpage(
|
|
request, None, note='Logging in', errnote='Unable to log in')
|
|
if re.search(r'(?i)<h1 class="mb8p4">Log in error</h1>', login_results) is not None:
|
|
self._downloader.report_warning('unable to log in: bad username or password')
|
|
return False
|
|
return True
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
video_id = mobj.group(1)
|
|
|
|
# Get video webpage. We are not actually interested in it, but need
|
|
# the cookies in order to be able to download the info webpage
|
|
self._download_webpage('http://www.nicovideo.jp/watch/' + video_id, video_id)
|
|
|
|
video_info = self._download_xml(
|
|
'http://ext.nicovideo.jp/api/getthumbinfo/' + video_id, video_id,
|
|
note='Downloading video info page')
|
|
|
|
# Get flv info
|
|
flv_info_webpage = self._download_webpage(
|
|
'http://flapi.nicovideo.jp/api/getflv?v=' + video_id,
|
|
video_id, 'Downloading flv info')
|
|
video_real_url = compat_urlparse.parse_qs(flv_info_webpage)['url'][0]
|
|
|
|
# Start extracting information
|
|
video_title = video_info.find('.//title').text
|
|
video_extension = video_info.find('.//movie_type').text
|
|
video_format = video_extension.upper()
|
|
video_thumbnail = video_info.find('.//thumbnail_url').text
|
|
video_description = video_info.find('.//description').text
|
|
video_uploader_id = video_info.find('.//user_id').text
|
|
video_upload_date = unified_strdate(video_info.find('.//first_retrieve').text.split('+')[0])
|
|
video_view_count = video_info.find('.//view_counter').text
|
|
video_webpage_url = video_info.find('.//watch_url').text
|
|
|
|
# uploader
|
|
video_uploader = video_uploader_id
|
|
url = 'http://seiga.nicovideo.jp/api/user/info?id=' + video_uploader_id
|
|
try:
|
|
user_info = self._download_xml(
|
|
url, video_id, note='Downloading user information')
|
|
video_uploader = user_info.find('.//nickname').text
|
|
except ExtractorError as err:
|
|
self._downloader.report_warning('Unable to download user info webpage: %s' % compat_str(err))
|
|
|
|
return {
|
|
'id': video_id,
|
|
'url': video_real_url,
|
|
'title': video_title,
|
|
'ext': video_extension,
|
|
'format': video_format,
|
|
'thumbnail': video_thumbnail,
|
|
'description': video_description,
|
|
'uploader': video_uploader,
|
|
'upload_date': video_upload_date,
|
|
'uploader_id': video_uploader_id,
|
|
'view_count': video_view_count,
|
|
'webpage_url': video_webpage_url,
|
|
}
|