1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-12-28 12:02:58 +08:00
youtube-dl/youtube_dl/extractor/funnyordie.py
2013-07-13 11:29:08 +02:00

38 lines
1.4 KiB
Python

import re
from .common import InfoExtractor
class FunnyOrDieIE(InfoExtractor):
_VALID_URL = r'^(?:https?://)?(?:www\.)?funnyordie\.com/videos/(?P<id>[0-9a-f]+)/.*$'
_TEST = {
u'url': u'http://www.funnyordie.com/videos/0732f586d7/heart-shaped-box-literal-video-version',
u'file': u'0732f586d7.mp4',
u'md5': u'f647e9e90064b53b6e046e75d0241fbd',
u'info_dict': {
u"description": u"Lyrics changed to match the video. Spoken cameo by Obscurus Lupa (from ThatGuyWithTheGlasses.com). Based on a concept by Dustin McLean (DustFilms.com). Performed, edited, and written by David A. Scott.",
u"title": u"Heart-Shaped Box: Literal Video Version"
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
webpage = self._download_webpage(url, video_id)
video_url = self._html_search_regex(r'<video[^>]*>\s*<source[^>]*>\s*<source src="(?P<url>[^"]+)"',
webpage, u'video URL', flags=re.DOTALL)
title = self._html_search_regex((r"<h1 class='player_page_h1'.*?>(?P<title>.*?)</h1>",
r'<title>(?P<title>[^<]+?)</title>'), webpage, 'title', flags=re.DOTALL)
info = {
'id': video_id,
'url': video_url,
'ext': 'mp4',
'title': title,
'description': self._og_search_description(webpage),
}
return [info]