Merge branch 'webofstories' of https://github.com/dufferzafar/youtube-dl into dufferzafar-webofstories
commit
3f19b9b7c1
|
@ -713,7 +713,10 @@ from .wdr import (
|
|||
WDRMobileIE,
|
||||
WDRMausIE,
|
||||
)
|
||||
from .webofstories import WebOfStoriesIE
|
||||
from .webofstories import (
|
||||
WebOfStoriesIE,
|
||||
WebOfStoriesPlaylistIE,
|
||||
)
|
||||
from .weibo import WeiboIE
|
||||
from .wimp import WimpIE
|
||||
from .wistia import WistiaIE
|
||||
|
|
|
@ -1,6 +1,8 @@
|
|||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import int_or_none
|
||||
|
||||
|
@ -98,3 +100,26 @@ class WebOfStoriesIE(InfoExtractor):
|
|||
'description': description,
|
||||
'duration': duration,
|
||||
}
|
||||
|
||||
|
||||
class WebOfStoriesPlaylistIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?webofstories\.com/playAll/(?P<id>[^/]+)'
|
||||
_TESTS = []
|
||||
|
||||
def _real_extract(self, url):
|
||||
playlist_id = self._match_id(url)
|
||||
|
||||
webpage = self._download_webpage(url, playlist_id)
|
||||
|
||||
entries = [
|
||||
self.url_result('http://www.webofstories.com/play/%s' % video_number, 'WebOfStories')
|
||||
for video_number in set(re.findall('href="/playAll/%s\?sId=(\d+)"' % playlist_id, webpage))
|
||||
]
|
||||
|
||||
title = self._html_search_regex(
|
||||
r'<title>([^<]+)\s*-\s*Web\sof\sStories</title>', webpage, 'title')
|
||||
|
||||
description = self._html_search_meta(
|
||||
'description', webpage, 'description')
|
||||
|
||||
return self.playlist_result(entries, playlist_id, title, description)
|
||||
|
|
Loading…
Reference in New Issue