mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-27 14:15:37 +01:00
[youku:show] Add new extractor
This commit is contained in:
parent
6da8d7de69
commit
411c590a1f
|
@ -1013,7 +1013,10 @@ from .yesjapan import YesJapanIE
|
|||
from .yinyuetai import YinYueTaiIE
|
||||
from .ynet import YnetIE
|
||||
from .youjizz import YouJizzIE
|
||||
from .youku import YoukuIE
|
||||
from .youku import (
|
||||
YoukuIE,
|
||||
YoukuShowIE,
|
||||
)
|
||||
from .youporn import YouPornIE
|
||||
from .yourupload import YourUploadIE
|
||||
from .youtube import (
|
||||
|
|
|
@ -2,7 +2,9 @@
|
|||
from __future__ import unicode_literals
|
||||
|
||||
import base64
|
||||
import itertools
|
||||
import random
|
||||
import re
|
||||
import string
|
||||
import time
|
||||
|
||||
|
@ -13,6 +15,7 @@ from ..compat import (
|
|||
)
|
||||
from ..utils import (
|
||||
ExtractorError,
|
||||
get_element_by_attribute,
|
||||
sanitized_Request,
|
||||
)
|
||||
|
||||
|
@ -285,3 +288,52 @@ class YoukuIE(InfoExtractor):
|
|||
'title': title,
|
||||
'entries': entries,
|
||||
}
|
||||
|
||||
|
||||
class YoukuShowIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?youku\.com/show_page/id_(?P<id>[0-9a-z]+)\.html'
|
||||
IE_NAME = 'youku:show'
|
||||
|
||||
_TEST = {
|
||||
'url': 'http://www.youku.com/show_page/id_zc7c670be07ff11e48b3f.html',
|
||||
'info_dict': {
|
||||
'id': 'zc7c670be07ff11e48b3f',
|
||||
'title': '花千骨 未删减版',
|
||||
'description': 'md5:578d4f2145ae3f9128d9d4d863312910',
|
||||
},
|
||||
'playlist_count': 50,
|
||||
}
|
||||
|
||||
_PAGE_SIZE = 40
|
||||
|
||||
def _find_videos_in_page(self, webpage):
|
||||
videos = re.findall(
|
||||
r'<li><a[^>]+href="(?P<url>https?://v\.youku\.com/[^"]+)"[^>]+title="(?P<title>[^"]+)"', webpage)
|
||||
return [
|
||||
self.url_result(video_url, YoukuIE.ie_key(), title)
|
||||
for video_url, title in videos]
|
||||
|
||||
def _real_extract(self, url):
|
||||
show_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url, show_id)
|
||||
|
||||
entries = self._find_videos_in_page(webpage)
|
||||
|
||||
playlist_title = self._html_search_regex(
|
||||
r'<span[^>]+class="name">([^<]+)</span>', webpage, 'playlist title', fatal=False)
|
||||
detail_div = get_element_by_attribute('class', 'detail', webpage) or ''
|
||||
playlist_description = self._html_search_regex(
|
||||
r'<span[^>]+style="display:none"[^>]*>([^<]+)</span>',
|
||||
detail_div, 'playlist description', fatal=False)
|
||||
|
||||
for idx in itertools.count(1):
|
||||
episodes_page = self._download_webpage(
|
||||
'http://www.youku.com/show_episode/id_%s.html' % show_id,
|
||||
show_id, query={'divid': 'reload_%d' % (idx * self._PAGE_SIZE + 1)},
|
||||
note='Downloading episodes page %d' % idx)
|
||||
new_entries = self._find_videos_in_page(episodes_page)
|
||||
entries.extend(new_entries)
|
||||
if len(new_entries) < self._PAGE_SIZE:
|
||||
break
|
||||
|
||||
return self.playlist_result(entries, show_id, playlist_title, playlist_description)
|
||||
|
|
Loading…
Reference in New Issue
Block a user