]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/brightcove.py
Update changelog.
[youtubedl] / youtube_dl / extractor / brightcove.py
index f85acbb5db3dcb8e68b1e6e19f4eb91095fa6cfd..66fe0ac9ade6fad80d77f0429c136c2d022af16d 100644 (file)
+# encoding: utf-8
+
 import re
 import json
+import xml.etree.ElementTree
 
 from .common import InfoExtractor
+from ..utils import (
+    compat_urllib_parse,
+    find_xpath_attr,
+    compat_urlparse,
+    compat_str,
+    compat_urllib_request,
+
+    ExtractorError,
+)
+
 
 class BrightcoveIE(InfoExtractor):
-    _VALID_URL = r'http://.*brightcove\.com/.*\?(?P<query>.*videoPlayer=(?P<id>\d*).*)'
+    _VALID_URL = r'https?://.*brightcove\.com/(services|viewer).*\?(?P<query>.*)'
+    _FEDERATED_URL_TEMPLATE = 'http://c.brightcove.com/services/viewer/htmlFederated?%s'
+    _PLAYLIST_URL_TEMPLATE = 'http://c.brightcove.com/services/json/experience/runtime/?command=get_programming_for_experience&playerKey=%s'
+
+    _TESTS = [
+        {
+            # From http://www.8tv.cat/8aldia/videos/xavier-sala-i-martin-aquesta-tarda-a-8-al-dia/
+            u'url': u'http://c.brightcove.com/services/viewer/htmlFederated?playerID=1654948606001&flashID=myExperience&%40videoPlayer=2371591881001',
+            u'file': u'2371591881001.mp4',
+            u'md5': u'8eccab865181d29ec2958f32a6a754f5',
+            u'note': u'Test Brightcove downloads and detection in GenericIE',
+            u'info_dict': {
+                u'title': u'Xavier Sala i Martín: “Un banc que no presta és un banc zombi que no serveix per a res”',
+                u'uploader': u'8TV',
+                u'description': u'md5:a950cc4285c43e44d763d036710cd9cd',
+            }
+        },
+        {
+            # From http://medianetwork.oracle.com/video/player/1785452137001
+            u'url': u'http://c.brightcove.com/services/viewer/htmlFederated?playerID=1217746023001&flashID=myPlayer&%40videoPlayer=1785452137001',
+            u'file': u'1785452137001.flv',
+            u'info_dict': {
+                u'title': u'JVMLS 2012: Arrays 2.0 - Opportunities and Challenges',
+                u'description': u'John Rose speaks at the JVM Language Summit, August 1, 2012.',
+                u'uploader': u'Oracle',
+            },
+        },
+        {
+            # From http://mashable.com/2013/10/26/thermoelectric-bracelet-lets-you-control-your-body-temperature/
+            u'url': u'http://c.brightcove.com/services/viewer/federated_f9?&playerID=1265504713001&publisherID=AQ%7E%7E%2CAAABBzUwv1E%7E%2CxP-xFHVUstiMFlNYfvF4G9yFnNaqCw_9&videoID=2750934548001',
+            u'info_dict': {
+                u'id': u'2750934548001',
+                u'ext': u'mp4',
+                u'title': u'This Bracelet Acts as a Personal Thermostat',
+                u'description': u'md5:547b78c64f4112766ccf4e151c20b6a0',
+                u'uploader': u'Mashable',
+            },
+        },
+    ]
+
+    @classmethod
+    def _build_brighcove_url(cls, object_str):
+        """
+        Build a Brightcove url from a xml string containing
+        <object class="BrightcoveExperience">{params}</object>
+        """
+
+        # Fix up some stupid HTML, see https://github.com/rg3/youtube-dl/issues/1553
+        object_str = re.sub(r'(<param name="[^"]+" value="[^"]+")>',
+                            lambda m: m.group(1) + '/>', object_str)
+        # Fix up some stupid XML, see https://github.com/rg3/youtube-dl/issues/1608
+        object_str = object_str.replace(u'<--', u'<!--')
+
+        object_doc = xml.etree.ElementTree.fromstring(object_str)
+        assert u'BrightcoveExperience' in object_doc.attrib['class']
+        params = {'flashID': object_doc.attrib['id'],
+                  'playerID': find_xpath_attr(object_doc, './param', 'name', 'playerID').attrib['value'],
+                  }
+        def find_param(name):
+            node = find_xpath_attr(object_doc, './param', 'name', name)
+            if node is not None:
+                return node.attrib['value']
+            return None
+        playerKey = find_param('playerKey')
+        # Not all pages define this value
+        if playerKey is not None:
+            params['playerKey'] = playerKey
+        # The three fields hold the id of the video
+        videoPlayer = find_param('@videoPlayer') or find_param('videoId') or find_param('videoID')
+        if videoPlayer is not None:
+            params['@videoPlayer'] = videoPlayer
+        linkBase = find_param('linkBaseURL')
+        if linkBase is not None:
+            params['linkBaseURL'] = linkBase
+        data = compat_urllib_parse.urlencode(params)
+        return cls._FEDERATED_URL_TEMPLATE % data
+
+    @classmethod
+    def _extract_brightcove_url(cls, webpage):
+        """Try to extract the brightcove url from the wepbage, returns None
+        if it can't be found
+        """
+        m_brightcove = re.search(
+            r'<object[^>]+?class=([\'"])[^>]*?BrightcoveExperience.*?\1.+?</object>',
+            webpage, re.DOTALL)
+        if m_brightcove is not None:
+            return cls._build_brighcove_url(m_brightcove.group())
+        else:
+            return None
 
     def _real_extract(self, url):
+        # Change the 'videoId' and others field to '@videoPlayer'
+        url = re.sub(r'(?<=[?&])(videoI(d|D)|bctid)', '%40videoPlayer', url)
+        # Change bckey (used by bcove.me urls) to playerKey
+        url = re.sub(r'(?<=[?&])bckey', 'playerKey', url)
         mobj = re.match(self._VALID_URL, url)
-        query = mobj.group('query')
-        video_id = mobj.group('id')
+        query_str = mobj.group('query')
+        query = compat_urlparse.parse_qs(query_str)
 
-        request_url = 'http://c.brightcove.com/services/viewer/htmlFederated?%s' % query
-        webpage = self._download_webpage(request_url, video_id)
+        videoPlayer = query.get('@videoPlayer')
+        if videoPlayer:
+            return self._get_video_info(videoPlayer[0], query_str, query)
+        else:
+            player_key = query['playerKey']
+            return self._get_playlist_info(player_key[0])
+
+    def _get_video_info(self, video_id, query_str, query):
+        request_url = self._FEDERATED_URL_TEMPLATE % query_str
+        req = compat_urllib_request.Request(request_url)
+        linkBase = query.get('linkBaseURL')
+        if linkBase is not None:
+            req.add_header('Referer', linkBase[0])
+        webpage = self._download_webpage(req, video_id)
 
         self.report_extraction(video_id)
         info = self._search_regex(r'var experienceJSON = ({.*?});', webpage, 'json')
         info = json.loads(info)['data']
         video_info = info['programmedContent']['videoPlayer']['mediaDTO']
-        renditions = video_info['renditions']
-        renditions = sorted(renditions, key=lambda r: r['size'])
-        best_format = renditions[-1]
-        
-        return {'id': video_id,
-                'title': video_info['displayName'],
-                'url': best_format['defaultURL'], 
-                'ext': 'mp4',
-                'description': video_info.get('shortDescription'),
-                'thumbnail': video_info.get('videoStillURL') or video_info.get('thumbnailURL'),
-                'uploader': video_info.get('publisherName'),
-                }
+
+        return self._extract_video_info(video_info)
+
+    def _get_playlist_info(self, player_key):
+        playlist_info = self._download_webpage(self._PLAYLIST_URL_TEMPLATE % player_key,
+                                               player_key, u'Downloading playlist information')
+
+        json_data = json.loads(playlist_info)
+        if 'videoList' not in json_data:
+            raise ExtractorError(u'Empty playlist')
+        playlist_info = json_data['videoList']
+        videos = [self._extract_video_info(video_info) for video_info in playlist_info['mediaCollectionDTO']['videoDTOs']]
+
+        return self.playlist_result(videos, playlist_id=playlist_info['id'],
+                                    playlist_title=playlist_info['mediaCollectionDTO']['displayName'])
+
+    def _extract_video_info(self, video_info):
+        info = {
+            'id': compat_str(video_info['id']),
+            'title': video_info['displayName'],
+            'description': video_info.get('shortDescription'),
+            'thumbnail': video_info.get('videoStillURL') or video_info.get('thumbnailURL'),
+            'uploader': video_info.get('publisherName'),
+        }
+
+        renditions = video_info.get('renditions')
+        if renditions:
+            renditions = sorted(renditions, key=lambda r: r['size'])
+            info['formats'] = [{
+                'url': rend['defaultURL'],
+                'height': rend.get('frameHeight'),
+                'width': rend.get('frameWidth'),
+            } for rend in renditions]
+        elif video_info.get('FLVFullLengthURL') is not None:
+            info.update({
+                'url': video_info['FLVFullLengthURL'],
+            })
+        else:
+            raise ExtractorError(u'Unable to extract video url for %s' % info['id'])
+        return info