0
0
Fork 0

[youtube] Extract base for entry list extractors and support multi page lists of playlists

main
Sergey M․ 2015-11-22 05:01:01 +06:00
parent 82c4d7b0ce
commit 061a75edd6
1 changed files with 17 additions and 11 deletions

View File

@ -178,15 +178,13 @@ class YoutubeBaseInfoExtractor(InfoExtractor):
return
class YoutubePlaylistBaseInfoExtractor(InfoExtractor):
# Extract the video ids from the playlist pages
class YoutubeEntryListBaseInfoExtractor(InfoExtractor):
# Extract entries from page with "Load more" button
def _entries(self, page, playlist_id):
more_widget_html = content_html = page
for page_num in itertools.count(1):
for video_id, video_title in self.extract_videos_from_page(content_html):
yield self.url_result(
video_id, 'Youtube', video_id=video_id,
video_title=video_title)
for entry in self._process_page(content_html):
yield entry
mobj = re.search(r'data-uix-load-more-href="/?(?P<more>[^"]+)"', more_widget_html)
if not mobj:
@ -203,6 +201,12 @@ class YoutubePlaylistBaseInfoExtractor(InfoExtractor):
break
more_widget_html = more['load_more_widget_html']
class YoutubePlaylistBaseInfoExtractor(YoutubeEntryListBaseInfoExtractor):
def _process_page(self, content):
for video_id, video_title in self.extract_videos_from_page(content):
yield self.url_result(video_id, 'Youtube', video_id, video_title)
def extract_videos_from_page(self, page):
ids_in_page = []
titles_in_page = []
@ -224,15 +228,17 @@ class YoutubePlaylistBaseInfoExtractor(InfoExtractor):
return zip(ids_in_page, titles_in_page)
class YoutubePlaylistsBaseInfoExtractor(InfoExtractor):
class YoutubePlaylistsBaseInfoExtractor(YoutubeEntryListBaseInfoExtractor):
def _process_page(self, content):
for playlist_id in re.findall(r'href="/?playlist\?list=(.+?)"', content):
yield self.url_result(
'https://www.youtube.com/playlist?list=%s' % playlist_id, 'YoutubePlaylist')
def _real_extract(self, url):
playlist_id = self._match_id(url)
webpage = self._download_webpage(url, playlist_id)
entries = [
self.url_result(compat_urlparse.urljoin(url, playlist), 'YoutubePlaylist')
for playlist in re.findall(r'href="(/playlist\?list=.+?)"', webpage)]
title = self._og_search_title(webpage, fatal=False)
return self.playlist_result(entries, playlist_id, title)
return self.playlist_result(self._entries(webpage, playlist_id), playlist_id, title)
class YoutubeIE(YoutubeBaseInfoExtractor):