from __future__ import unicode_literals
-import json
import re
+import json
from .common import InfoExtractor
+from ..utils import str_to_int
class NineGagIE(InfoExtractor):
IE_NAME = '9gag'
- _VALID_URL = r'^https?://(?:www\.)?9gag\.tv/v/(?P<id>[0-9]+)'
+ _VALID_URL = r'''(?x)^https?://(?:www\.)?9gag\.tv/
+ (?:
+ v/(?P<numid>[0-9]+)|
+ p/(?P<id>[a-zA-Z0-9]+)/(?P<display_id>[^?#/]+)
+ )
+ '''
- _TEST = {
+ _TESTS = [{
"url": "http://9gag.tv/v/1912",
- "file": "1912.mp4",
"info_dict": {
+ "id": "1912",
+ "ext": "mp4",
"description": "This 3-minute video will make you smile and then make you feel untalented and insignificant. Anyway, you should share this awesomeness. (Thanks, Dino!)",
- "title": "\"People Are Awesome 2013\" Is Absolutely Awesome"
+ "title": "\"People Are Awesome 2013\" Is Absolutely Awesome",
+ "view_count": int,
+ "thumbnail": "re:^https?://",
},
'add_ie': ['Youtube']
- }
+ },
+ {
+ 'url': 'http://9gag.tv/p/KklwM/alternate-banned-opening-scene-of-gravity?ref=fsidebar',
+ 'info_dict': {
+ 'id': 'KklwM',
+ 'ext': 'mp4',
+ 'display_id': 'alternate-banned-opening-scene-of-gravity',
+ "description": "While Gravity was a pretty awesome movie already, YouTuber Krishna Shenoi came up with a way to improve upon it, introducing a much better solution to Sandra Bullock's seemingly endless tumble in space. The ending is priceless.",
+ 'title': "Banned Opening Scene Of \"Gravity\" That Changes The Whole Movie",
+ },
+ }]
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('id')
+ video_id = mobj.group('numid') or mobj.group('id')
+ display_id = mobj.group('display_id') or video_id
+
+ webpage = self._download_webpage(url, display_id)
- webpage = self._download_webpage(url, video_id)
- data_json = self._html_search_regex(r'''(?x)
- <div\s*id="tv-video"\s*data-video-source="youtube"\s*
- data-video-meta="([^"]+)"''', webpage, 'video metadata')
+ post_view = json.loads(self._html_search_regex(
+ r'var postView = new app\.PostView\({\s*post:\s*({.+?}),', webpage, 'post view'))
- data = json.loads(data_json)
+ youtube_id = post_view['videoExternalId']
+ title = post_view['title']
+ description = post_view['description']
+ view_count = str_to_int(post_view['externalView'])
+ thumbnail = post_view.get('thumbnail_700w') or post_view.get('ogImageUrl') or post_view.get('thumbnail_300w')
return {
'_type': 'url_transparent',
- 'url': data['youtubeVideoId'],
+ 'url': youtube_id,
'ie_key': 'Youtube',
'id': video_id,
- 'title': data['title'],
- 'description': data['description'],
- 'view_count': int(data['view_count']),
- 'like_count': int(data['statistic']['like']),
- 'dislike_count': int(data['statistic']['dislike']),
- 'thumbnail': data['thumbnail_url'],
+ 'display_id': display_id,
+ 'title': title,
+ 'description': description,
+ 'view_count': view_count,
+ 'thumbnail': thumbnail,
}