[mit] Modernize
authorPhilipp Hagemeister <phihag@phihag.de>
Tue, 25 Feb 2014 23:06:31 +0000 (00:06 +0100)
committerPhilipp Hagemeister <phihag@phihag.de>
Tue, 25 Feb 2014 23:06:31 +0000 (00:06 +0100)
youtube_dl/extractor/mit.py

index 76b717fe5dbac08b8b103a1e44192a6fbf6d2a55..f50e36cb4fcc82a45d3a5f379beadb2764e2faf9 100644 (file)
@@ -1,24 +1,28 @@
+from __future__ import unicode_literals
+
 import re
 import json
 
 from .common import InfoExtractor
 from ..utils import (
+    compat_urlparse,
     clean_html,
     get_element_by_id,
 )
 
 
 class TechTVMITIE(InfoExtractor):
-    IE_NAME = u'techtv.mit.edu'
+    IE_NAME = 'techtv.mit.edu'
     _VALID_URL = r'https?://techtv\.mit\.edu/(videos|embeds)/(?P<id>\d+)'
 
     _TEST = {
-        u'url': u'http://techtv.mit.edu/videos/25418-mit-dna-learning-center-set',
-        u'file': u'25418.mp4',
-        u'md5': u'1f8cb3e170d41fd74add04d3c9330e5f',
-        u'info_dict': {
-            u'title': u'MIT DNA Learning Center Set',
-            u'description': u'md5:82313335e8a8a3f243351ba55bc1b474',
+        'url': 'http://techtv.mit.edu/videos/25418-mit-dna-learning-center-set',
+        'md5': '1f8cb3e170d41fd74add04d3c9330e5f',
+        'info_dict': {
+            'id': '25418',
+            'ext': 'mp4',
+            'title': 'MIT DNA Learning Center Set',
+            'description': 'md5:82313335e8a8a3f243351ba55bc1b474',
         },
     }
 
@@ -27,12 +31,12 @@ class TechTVMITIE(InfoExtractor):
         video_id = mobj.group('id')
         raw_page = self._download_webpage(
             'http://techtv.mit.edu/videos/%s' % video_id, video_id)
-        clean_page = re.compile(u'<!--.*?-->', re.S).sub(u'', raw_page)
+        clean_page = re.compile(r'<!--.*?-->', re.S).sub('', raw_page)
 
-        base_url = self._search_regex(r'ipadUrl: \'(.+?cloudfront.net/)',
-            raw_page, u'base url')
-        formats_json = self._search_regex(r'bitrates: (\[.+?\])', raw_page,
-            u'video formats')
+        base_url = self._search_regex(
+            r'ipadUrl: \'(.+?cloudfront.net/)', raw_page, 'base url')
+        formats_json = self._search_regex(
+            r'bitrates: (\[.+?\])', raw_page, 'video formats')
         formats_mit = json.loads(formats_json)
         formats = [
             {
@@ -48,28 +52,32 @@ class TechTVMITIE(InfoExtractor):
 
         title = get_element_by_id('edit-title', clean_page)
         description = clean_html(get_element_by_id('edit-description', clean_page))
-        thumbnail = self._search_regex(r'playlist:.*?url: \'(.+?)\'',
-            raw_page, u'thumbnail', flags=re.DOTALL)
+        thumbnail = self._search_regex(
+            r'playlist:.*?url: \'(.+?)\'',
+            raw_page, 'thumbnail', flags=re.DOTALL)
 
-        return {'id': video_id,
-                'title': title,
-                'formats': formats,
-                'description': description,
-                'thumbnail': thumbnail,
-                }
+        return {
+            'id': video_id,
+            'title': title,
+            'formats': formats,
+            'description': description,
+            'thumbnail': thumbnail,
+        }
 
 
 class MITIE(TechTVMITIE):
-    IE_NAME = u'video.mit.edu'
+    IE_NAME = 'video.mit.edu'
     _VALID_URL = r'https?://video\.mit\.edu/watch/(?P<title>[^/]+)'
 
     _TEST = {
-        u'url': u'http://video.mit.edu/watch/the-government-is-profiling-you-13222/',
-        u'file': u'21783.mp4',
-        u'md5': u'7db01d5ccc1895fc5010e9c9e13648da',
-        u'info_dict': {
-            u'title': u'The Government is Profiling You',
-            u'description': u'md5:ad5795fe1e1623b73620dbfd47df9afd',
+        'url': 'http://video.mit.edu/watch/the-government-is-profiling-you-13222/',
+        'file': '.mp4',
+        'md5': '7db01d5ccc1895fc5010e9c9e13648da',
+        'info_dict': {
+            'id': '21783',
+            'ext': 'mp4',
+            'title': 'The Government is Profiling You',
+            'description': 'md5:ad5795fe1e1623b73620dbfd47df9afd',
         },
     }
 
@@ -77,7 +85,6 @@ class MITIE(TechTVMITIE):
         mobj = re.match(self._VALID_URL, url)
         page_title = mobj.group('title')
         webpage = self._download_webpage(url, page_title)
-        self.to_screen('%s: Extracting %s url' % (page_title, TechTVMITIE.IE_NAME))
-        embed_url = self._search_regex(r'<iframe .*?src="(.+?)"', webpage,
-            u'embed url')
+        embed_url = self._search_regex(
+            r'<iframe .*?src="(.+?)"', webpage, 'embed url')
         return self.url_result(embed_url, ie='TechTVMIT')