|
|
|
@ -5,7 +5,7 @@ from .youtube import YoutubeIE
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class WimpIE(InfoExtractor):
|
|
|
|
|
_VALID_URL = r'http://(?:www\.)?wimp\.com/(?P<id>[^/]+)/'
|
|
|
|
|
_VALID_URL = r'http://(?:www\.)?wimp\.com/(?P<id>[^/]+)'
|
|
|
|
|
_TESTS = [{
|
|
|
|
|
'url': 'http://www.wimp.com/maruexhausted/',
|
|
|
|
|
'md5': 'ee21217ffd66d058e8b16be340b74883',
|
|
|
|
@ -28,18 +28,23 @@ class WimpIE(InfoExtractor):
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
|
|
video_id = self._match_id(url)
|
|
|
|
|
|
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
|
video_url = self._search_regex(
|
|
|
|
|
[r"[\"']file[\"']\s*[:,]\s*[\"'](.+?)[\"']", r"videoId\s*:\s*[\"']([^\"']+)[\"']"],
|
|
|
|
|
webpage, 'video URL')
|
|
|
|
|
if YoutubeIE.suitable(video_url):
|
|
|
|
|
self.to_screen('Found YouTube video')
|
|
|
|
|
|
|
|
|
|
youtube_id = self._search_regex(
|
|
|
|
|
r"videoId\s*:\s*[\"']([0-9A-Za-z_-]{11})[\"']",
|
|
|
|
|
webpage, 'video URL', default=None)
|
|
|
|
|
if youtube_id:
|
|
|
|
|
return {
|
|
|
|
|
'_type': 'url',
|
|
|
|
|
'url': video_url,
|
|
|
|
|
'url': youtube_id,
|
|
|
|
|
'ie_key': YoutubeIE.ie_key(),
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
video_url = self._search_regex(
|
|
|
|
|
r'<video[^>]+>\s*<source src=(["\'])(?P<url>.+?)\1',
|
|
|
|
|
webpage, 'video URL', group='url')
|
|
|
|
|
|
|
|
|
|
return {
|
|
|
|
|
'id': video_id,
|
|
|
|
|
'url': video_url,
|
|
|
|
|