aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/slideshare.py
diff options
context:
space:
mode:
authorJaime Marquínez Ferrándiz <jaime.marquinez.ferrandiz@gmail.com>2014-02-09 14:22:56 +0100
committerJaime Marquínez Ferrándiz <jaime.marquinez.ferrandiz@gmail.com>2014-02-09 14:23:19 +0100
commit1afe753462f0293122dc7a9b534b4f5cdb1e5c4e (patch)
tree133b44c05d94d1114428c3a98dedea4563343eee /youtube_dl/extractor/slideshare.py
parent524c2c716a55d16b3cf4f2ed344e3a063bb63ce8 (diff)
downloadyoutube-dl-1afe753462f0293122dc7a9b534b4f5cdb1e5c4e.tar.xz
[slideshare] Fix description extraction and modernize
The ‘og:description’ property doesn’t contain the full description
Diffstat (limited to 'youtube_dl/extractor/slideshare.py')
-rw-r--r--youtube_dl/extractor/slideshare.py23
1 files changed, 14 insertions, 9 deletions
diff --git a/youtube_dl/extractor/slideshare.py b/youtube_dl/extractor/slideshare.py
index afc3001b5..9c62825cc 100644
--- a/youtube_dl/extractor/slideshare.py
+++ b/youtube_dl/extractor/slideshare.py
@@ -1,3 +1,5 @@
+from __future__ import unicode_literals
+
import re
import json
@@ -12,11 +14,12 @@ class SlideshareIE(InfoExtractor):
_VALID_URL = r'https?://www\.slideshare\.net/[^/]+?/(?P<title>.+?)($|\?)'
_TEST = {
- u'url': u'http://www.slideshare.net/Dataversity/keynote-presentation-managing-scale-and-complexity',
- u'file': u'25665706.mp4',
- u'info_dict': {
- u'title': u'Managing Scale and Complexity',
- u'description': u'This was a keynote presentation at the NoSQL Now! 2013 Conference & Expo (http://www.nosqlnow.com). This presentation was given by Adrian Cockcroft from Netflix',
+ 'url': 'http://www.slideshare.net/Dataversity/keynote-presentation-managing-scale-and-complexity',
+ 'info_dict': {
+ 'id': '25665706',
+ 'ext': 'mp4',
+ 'title': 'Managing Scale and Complexity',
+ 'description': 'This was a keynote presentation at the NoSQL Now! 2013 Conference & Expo (http://www.nosqlnow.com). This presentation was given by Adrian Cockcroft from Netflix.',
},
}
@@ -26,15 +29,17 @@ class SlideshareIE(InfoExtractor):
webpage = self._download_webpage(url, page_title)
slideshare_obj = self._search_regex(
r'var slideshare_object = ({.*?}); var user_info =',
- webpage, u'slideshare object')
+ webpage, 'slideshare object')
info = json.loads(slideshare_obj)
- if info['slideshow']['type'] != u'video':
- raise ExtractorError(u'Webpage type is "%s": only video extraction is supported for Slideshare' % info['slideshow']['type'], expected=True)
+ if info['slideshow']['type'] != 'video':
+ raise ExtractorError('Webpage type is "%s": only video extraction is supported for Slideshare' % info['slideshow']['type'], expected=True)
doc = info['doc']
bucket = info['jsplayer']['video_bucket']
ext = info['jsplayer']['video_extension']
video_url = compat_urlparse.urljoin(bucket, doc + '-SD.' + ext)
+ description = self._html_search_regex(
+ r'<p class="description.*?"[^>]*>(.*?)</p>', webpage, 'description')
return {
'_type': 'video',
@@ -43,5 +48,5 @@ class SlideshareIE(InfoExtractor):
'ext': ext,
'url': video_url,
'thumbnail': info['slideshow']['pin_image_url'],
- 'description': self._og_search_description(webpage),
+ 'description': description,
}