|
|
- import re
-
- from .common import InfoExtractor
-
- class StatigramIE(InfoExtractor):
- _VALID_URL = r'(?:http://)?(?:www\.)?statigr\.am/p/([^/]+)'
- _TEST = {
- u'url': u'http://statigr.am/p/484091715184808010_284179915',
- u'file': u'484091715184808010_284179915.mp4',
- u'md5': u'deda4ff333abe2e118740321e992605b',
- u'info_dict': {
- u"uploader_id": u"videoseconds",
- u"title": u"Instagram photo by @videoseconds"
- }
- }
-
- def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group(1)
- webpage = self._download_webpage(url, video_id)
- html_title = self._html_search_regex(
- r'<title>(.+?)</title>',
- webpage, u'title')
- title = re.sub(r'(?: *\(Videos?\))? \| Statigram$', '', html_title)
- uploader_id = self._html_search_regex(
- r'@([^ ]+)', title, u'uploader name', fatal=False)
- ext = 'mp4'
-
- return [{
- 'id': video_id,
- 'url': self._og_search_video_url(webpage),
- 'ext': ext,
- 'title': title,
- 'thumbnail': self._og_search_thumbnail(webpage),
- 'uploader_id' : uploader_id
- }]
|