diff options
author | Philipp Hagemeister <phihag@phihag.de> | 2013-07-11 12:02:16 +0200 |
---|---|---|
committer | Philipp Hagemeister <phihag@phihag.de> | 2013-07-11 12:02:16 +0200 |
commit | 3fa95508373bb5813099c2e4ccad95638a506916 (patch) | |
tree | f963f088ffe9278e148441fde42b1fd43c3ea081 /youtube_dl/extractor/ehow.py | |
parent | 117adb0f0f14766189b3e2f635462990637d1274 (diff) | |
parent | b1082f01a681623028a3dc6d2ba0fef6b78a8301 (diff) |
Merge remote-tracking branch 'yasoob/master'
Diffstat (limited to 'youtube_dl/extractor/ehow.py')
-rw-r--r-- | youtube_dl/extractor/ehow.py | 44 |
1 files changed, 44 insertions, 0 deletions
diff --git a/youtube_dl/extractor/ehow.py b/youtube_dl/extractor/ehow.py new file mode 100644 index 000000000..a664b081a --- /dev/null +++ b/youtube_dl/extractor/ehow.py @@ -0,0 +1,44 @@ +import re +from ..utils import compat_urllib_parse +from .common import InfoExtractor + + +class EhowIE(InfoExtractor): + _VALID_URL = r'(?:http://)?(?:www\.)?ehow\.com/([^/]+)' + _TEST = { + u'url': u'http://www.ehow.com/video_12245069_hardwood-flooring-basics.html', + u'file': u'12245069.flv', + u'md5': u'9809b4e3f115ae2088440bcb4efbf371', + u'info_dict': { + u"title": u"Hardwood Flooring Basics", + u"description": u"Hardwood flooring may be time consuming, but its ultimately a pretty straightforward concept. Learn about hardwood flooring basics with help from a hardware flooring business owner in this free video...", + u"uploader": u"Erick Nathan" + } + } + + def _real_extract(self, url): + mobj = re.match(self._VALID_URL, url) + video_id = mobj.group(1).split("_")[1] + webpage = self._download_webpage(url, video_id) + video_url = self._search_regex(r'[^A-Za-z0-9]?(?:file|source)=(http[^\'"&]*)', + webpage, u'video URL') + final_url = compat_urllib_parse.unquote(video_url) + thumbnail_url = self._search_regex(r'<meta property="og:image" content="(.+?)" />', + webpage, u'thumbnail URL') + uploader = self._search_regex(r'<meta name="uploader" content="(.+?)" />', + webpage, u'uploader') + title = self._search_regex(r'<meta property="og:title" content="(.+?)" />', + webpage, u'Video title').replace(' | eHow','') + description = self._search_regex(r'<meta property="og:description" content="(.+?)" />', + webpage, u'video description') + ext = final_url.split('.')[-1] + return [{ + 'id': video_id, + 'url': final_url, + 'ext': ext, + 'title': title, + 'thumbnail': thumbnail_url, + 'description': description, + 'uploader': uploader, + }] + |