toypics.net support
authorfiocfun <fiocfun@gmail.com>
Fri, 21 Mar 2014 18:31:41 +0000 (00:31 +0600)
committerfiocfun <fiocfun@gmail.com>
Fri, 21 Mar 2014 22:07:44 +0000 (04:07 +0600)
youtube_dl/extractor/__init__.py
youtube_dl/extractor/toypics.py [new file with mode: 0644]

index 9cf73272fbcf2ab5eee3961c9437779d4f707541..06eac56e5fd8e8382f0fb42de8a6080b5bc8cb1b 100644 (file)
@@ -236,6 +236,7 @@ from .theplatform import ThePlatformIE
 from .thisav import ThisAVIE
 from .tinypic import TinyPicIE
 from .toutv import TouTvIE
+from .toypics import ToypicsIE
 from .traileraddict import TrailerAddictIE
 from .trilulilu import TriluliluIE
 from .trutube import TruTubeIE
diff --git a/youtube_dl/extractor/toypics.py b/youtube_dl/extractor/toypics.py
new file mode 100644 (file)
index 0000000..3cbfe2e
--- /dev/null
@@ -0,0 +1,55 @@
+from .common import InfoExtractor
+from math import ceil
+import re
+
+class ToypicsIE(InfoExtractor):
+    _VALID_URL = r'(?:http://)?videos\.toypics\.net/.*'
+    _TEST = {
+        'url': 'http://videos.toypics.net/view/514/chancebulged,-2-1/',
+        #'md5': '8a8b546956bbd0e769dbe28f6e80abb3', == $head -c10K 12929646011616163504.mp4 |md5sum //no idea why it fails
+        'info_dict': {
+            'id': '514',
+            'ext': 'mp4',
+            'title': 'Chance-Bulge\'d, 2',
+            'age_limit': 18
+        }
+    }
+    PAGINATED=8
+
+    def _real_extract(self, url):
+        mobj = re.match(r'(http://)?videos\.toypics\.net/(?P<username>[^/?]+)$', url)
+        if not mobj:
+            return self.extract_one(url)
+        return [self.extract_one(u) for u in self.process_paginated(url,
+            r'public/">Public Videos \((?P<videos_count>[0-9]+)\)</a></li>',
+            r'<p class="video-entry-title">\n\s*<a href="(http://videos.toypics.net/view/[^"]+)">'
+        )]
+
+    def process_paginated(self, profile_url, re_total, re_video_page):
+        profile_page = self._download_webpage(profile_url, 'profile' , 'getting profile page: '+profile_url)
+        videos_count = self._html_search_regex(re_total, profile_page, 'videos count')
+        lst = []
+        for n in xrange(1,int(ceil(float(videos_count)/self.PAGINATED)) +1):
+            lpage_url = profile_url +'/public/%d'%n
+            lpage = self._download_webpage(lpage_url, 'page %d'%n)
+            lst.extend(re.findall(re_video_page, lpage))
+        return lst
+
+    def extract_one(self,url):
+        mobj = re.match(r'(http://)?videos\.toypics\.net/view/(?P<videoid>[0-9]+)/.*', url)
+        video_id = mobj.group('videoid')
+        page = self._download_webpage(url, video_id, 'getting page: '+url)
+        video_url = self._html_search_regex(
+            r'src:\s+"(http://static[0-9]+\.toypics\.net/flvideo/[^"]+)"', page, 'video URL')
+        title = self._html_search_regex(
+            r'<title>Toypics - ([^<]+)</title>', page, 'title')
+        username = self._html_search_regex(
+            r'toypics.net/([^/"]+)" class="user-name">', page, 'username')
+        return {
+            'id': video_id,
+            'url': video_url,
+            'ext': video_url[-3:],
+            'title': title,
+            'uploader': username,
+            'age_limit': 18
+        }