]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/foxnews.py
d/patches: Refresh with gbp pq.
[youtubedl] / youtube_dl / extractor / foxnews.py
index 917f76b1effb3a2fff9d4f4c17c1cca348280132..dc0662f74ce5a84d59aa94333ee14d56a592cda2 100644 (file)
@@ -1,14 +1,15 @@
 from __future__ import unicode_literals
 
+import re
+
+from .amp import AMPIE
 from .common import InfoExtractor
-from ..utils import (
-    parse_iso8601,
-    int_or_none,
-)
 
 
-class FoxNewsIE(InfoExtractor):
-    _VALID_URL = r'https?://video\.foxnews\.com/v/(?:video-embed\.html\?video_id=)?(?P<id>\d+)'
+class FoxNewsIE(AMPIE):
+    IE_NAME = 'foxnews'
+    IE_DESC = 'Fox News and Fox Business Video'
+    _VALID_URL = r'https?://(?P<host>video\.(?:insider\.)?fox(?:news|business)\.com)/v/(?:video-embed\.html\?video_id=)?(?P<id>\d+)'
     _TESTS = [
         {
             'url': 'http://video.foxnews.com/v/3937480/frozen-in-time/#sp=show-clips',
@@ -17,11 +18,11 @@ class FoxNewsIE(InfoExtractor):
                 'id': '3937480',
                 'ext': 'flv',
                 'title': 'Frozen in Time',
-                'description': 'Doctors baffled by 16-year-old girl that is the size of a toddler',
+                'description': '16-year-old girl is size of toddler',
                 'duration': 265,
                 'timestamp': 1304411491,
                 'upload_date': '20110503',
-                'thumbnail': 're:^https?://.*\.jpg$',
+                'thumbnail': r're:^https?://.*\.jpg$',
             },
         },
         {
@@ -31,64 +32,109 @@ class FoxNewsIE(InfoExtractor):
                 'id': '3922535568001',
                 'ext': 'mp4',
                 'title': "Rep. Luis Gutierrez on if Obama's immigration plan is legal",
-                'description': "Congressman discusses the president's executive action",
+                'description': "Congressman discusses president's plan",
                 'duration': 292,
                 'timestamp': 1417662047,
                 'upload_date': '20141204',
-                'thumbnail': 're:^https?://.*\.jpg$',
+                'thumbnail': r're:^https?://.*\.jpg$',
+            },
+            'params': {
+                # m3u8 download
+                'skip_download': True,
             },
         },
         {
             'url': 'http://video.foxnews.com/v/video-embed.html?video_id=3937480&d=video.foxnews.com',
             'only_matching': True,
         },
+        {
+            'url': 'http://video.foxbusiness.com/v/4442309889001',
+            'only_matching': True,
+        },
+        {
+            # From http://insider.foxnews.com/2016/08/25/univ-wisconsin-student-group-pushing-silence-certain-words
+            'url': 'http://video.insider.foxnews.com/v/video-embed.html?video_id=5099377331001&autoplay=true&share_url=http://insider.foxnews.com/2016/08/25/univ-wisconsin-student-group-pushing-silence-certain-words&share_title=Student%20Group:%20Saying%20%27Politically%20Correct,%27%20%27Trash%27%20and%20%27Lame%27%20Is%20Offensive&share=true',
+            'only_matching': True,
+        },
     ]
 
     def _real_extract(self, url):
-        video_id = self._match_id(url)
-
-        video = self._download_json(
-            'http://video.foxnews.com/v/feed/video/%s.js?template=fox' % video_id, video_id)
-
-        item = video['channel']['item']
-        title = item['title']
-        description = item['description']
-        timestamp = parse_iso8601(item['dc-date'])
-
-        media_group = item['media-group']
-        duration = None
-        formats = []
-        for media in media_group['media-content']:
-            attributes = media['@attributes']
-            video_url = attributes['url']
-            if video_url.endswith('.f4m'):
-                formats.extend(self._extract_f4m_formats(video_url + '?hdcore=3.4.0&plugin=aasp-3.4.0.132.124', video_id))
-            elif video_url.endswith('.m3u8'):
-                formats.extend(self._extract_m3u8_formats(video_url, video_id, 'flv'))
-            elif not video_url.endswith('.smil'):
-                duration = int_or_none(attributes.get('duration'))
-                formats.append({
-                    'url': video_url,
-                    'format_id': media['media-category']['@attributes']['label'],
-                    'preference': 1,
-                    'vbr': int_or_none(attributes.get('bitrate')),
-                    'filesize': int_or_none(attributes.get('fileSize'))
-                })
-        self._sort_formats(formats)
-
-        media_thumbnail = media_group['media-thumbnail']['@attributes']
-        thumbnails = [{
-            'url': media_thumbnail['url'],
-            'width': int_or_none(media_thumbnail.get('width')),
-            'height': int_or_none(media_thumbnail.get('height')),
-        }] if media_thumbnail else []
+        host, video_id = re.match(self._VALID_URL, url).groups()
+
+        info = self._extract_feed_info(
+            'http://%s/v/feed/video/%s.js?template=fox' % (host, video_id))
+        info['id'] = video_id
+        return info
+
+
+class FoxNewsArticleIE(InfoExtractor):
+    _VALID_URL = r'https?://(?:www\.)?foxnews\.com/(?!v)([^/]+/)+(?P<id>[a-z-]+)'
+    IE_NAME = 'foxnews:article'
+
+    _TEST = {
+        'url': 'http://www.foxnews.com/politics/2016/09/08/buzz-about-bud-clinton-camp-denies-claims-wore-earpiece-at-forum.html',
+        'md5': '62aa5a781b308fdee212ebb6f33ae7ef',
+        'info_dict': {
+            'id': '5116295019001',
+            'ext': 'mp4',
+            'title': 'Trump and Clinton asked to defend positions on Iraq War',
+            'description': 'Veterans react on \'The Kelly File\'',
+            'timestamp': 1473299755,
+            'upload_date': '20160908',
+        },
+    }
+
+    def _real_extract(self, url):
+        display_id = self._match_id(url)
+        webpage = self._download_webpage(url, display_id)
+
+        video_id = self._html_search_regex(
+            r'data-video-id=([\'"])(?P<id>[^\'"]+)\1',
+            webpage, 'video ID', group='id')
+        return self.url_result(
+            'http://video.foxnews.com/v/' + video_id,
+            FoxNewsIE.ie_key())
+
+
+class FoxNewsInsiderIE(InfoExtractor):
+    _VALID_URL = r'https?://insider\.foxnews\.com/([^/]+/)+(?P<id>[a-z-]+)'
+    IE_NAME = 'foxnews:insider'
+
+    _TEST = {
+        'url': 'http://insider.foxnews.com/2016/08/25/univ-wisconsin-student-group-pushing-silence-certain-words',
+        'md5': 'a10c755e582d28120c62749b4feb4c0c',
+        'info_dict': {
+            'id': '5099377331001',
+            'display_id': 'univ-wisconsin-student-group-pushing-silence-certain-words',
+            'ext': 'mp4',
+            'title': 'Student Group: Saying \'Politically Correct,\' \'Trash\' and \'Lame\' Is Offensive',
+            'description': 'Is campus censorship getting out of control?',
+            'timestamp': 1472168725,
+            'upload_date': '20160825',
+            'thumbnail': r're:^https?://.*\.jpg$',
+        },
+        'params': {
+            # m3u8 download
+            'skip_download': True,
+        },
+        'add_ie': [FoxNewsIE.ie_key()],
+    }
+
+    def _real_extract(self, url):
+        display_id = self._match_id(url)
+
+        webpage = self._download_webpage(url, display_id)
+
+        embed_url = self._html_search_meta('embedUrl', webpage, 'embed URL')
+
+        title = self._og_search_title(webpage)
+        description = self._og_search_description(webpage)
 
         return {
-            'id': video_id,
+            '_type': 'url_transparent',
+            'ie_key': FoxNewsIE.ie_key(),
+            'url': embed_url,
+            'display_id': display_id,
             'title': title,
             'description': description,
-            'duration': duration,
-            'timestamp': timestamp,
-            'formats': formats,
-            'thumbnails': thumbnails,
         }