mirror of https://github.com/yt-dlp/yt-dlp.git
TumblrIE
I haven't found many videos to test, so it may not work for all.
This commit is contained in:
parent
8cb94542f4
commit
c15e024141
|
@ -337,6 +337,15 @@
|
|||
"title": "11.04.2013 09:23 Uhr - Tagesschau in 100 Sekunden"
|
||||
},
|
||||
"skip": "Requires rtmpdump"
|
||||
},
|
||||
{
|
||||
"name": "Tumblr",
|
||||
"url": "http://birthdayproject2012.tumblr.com/post/17258355236/a-sample-video-from-leeann-if-you-need-an-idea",
|
||||
"file": "17258355236.mp4",
|
||||
"md5": "7c6a514d691b034ccf8567999e9e88a3",
|
||||
"info_dict": {
|
||||
"title": "A sample video from LeeAnn. (If you need an idea..."
|
||||
}
|
||||
}
|
||||
|
||||
]
|
||||
|
|
|
@ -4327,6 +4327,40 @@ class ARDIE(InfoExtractor):
|
|||
info["url"] = stream["video_url"]
|
||||
return [info]
|
||||
|
||||
class TumblrIE(InfoExtractor):
|
||||
_VALID_URL = r'http://(?P<blog_name>.*?).tumblr.com/((post)|(video))/(?P<id>\d*)/(.*?)'
|
||||
|
||||
def _real_extract(self, url):
|
||||
m_url = re.match(self._VALID_URL, url)
|
||||
video_id = m_url.group('id')
|
||||
blog = m_url.group('blog_name')
|
||||
|
||||
url = 'http://%s.tumblr.com/post/%s/' % (blog, video_id)
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
re_video = r'src=\\x22(?P<video_url>http://%s.tumblr.com/video_file/%s/(.*?))\\x22 type=\\x22video/(?P<ext>.*?)\\x22' % (blog, video_id)
|
||||
video = re.search(re_video, webpage)
|
||||
if video is None:
|
||||
self.to_screen("No video founded")
|
||||
return []
|
||||
video_url = video.group('video_url')
|
||||
ext = video.group('ext')
|
||||
|
||||
re_thumb = r'posters(.*?)\[\\x22(?P<thumb>.*?)\\x22' # We pick the first poster
|
||||
thumb = re.search(re_thumb, webpage).group('thumb').replace('\\', '')
|
||||
|
||||
# The only place where you can get a title, it's not complete,
|
||||
# but searching in other places doesn't work for all videos
|
||||
re_title = r'<title>(.*?) - (?P<title>.*?)</title>'
|
||||
title = unescapeHTML(re.search(re_title, webpage).group('title'))
|
||||
|
||||
return [{'id': video_id,
|
||||
'url': video_url,
|
||||
'title': title,
|
||||
'thumbnail': thumb,
|
||||
'ext': ext
|
||||
}]
|
||||
|
||||
|
||||
def gen_extractors():
|
||||
""" Return a list of an instance of every supported extractor.
|
||||
|
@ -4381,6 +4415,7 @@ def gen_extractors():
|
|||
SpiegelIE(),
|
||||
LiveLeakIE(),
|
||||
ARDIE(),
|
||||
TumblrIE(),
|
||||
GenericIE()
|
||||
]
|
||||
|
||||
|
|
Loading…
Reference in New Issue