]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/spiegel.py
New upstream version 2020.06.16.1
[youtubedl] / youtube_dl / extractor / spiegel.py
index ec1b603889754af70d516e70c123be7d2604387a..4df7f4ddce4b701b66decee671e48ecb56af96fa 100644 (file)
@@ -4,12 +4,16 @@ from __future__ import unicode_literals
 import re
 
 from .common import InfoExtractor
+from .nexx import (
+    NexxIE,
+    NexxEmbedIE,
+)
 from .spiegeltv import SpiegeltvIE
 from ..compat import compat_urlparse
 from ..utils import (
-    extract_attributes,
-    unified_strdate,
-    get_element_by_attribute,
+    parse_duration,
+    strip_or_none,
+    unified_timestamp,
 )
 
 
@@ -17,88 +21,72 @@ class SpiegelIE(InfoExtractor):
     _VALID_URL = r'https?://(?:www\.)?spiegel\.de/video/[^/]*-(?P<id>[0-9]+)(?:-embed|-iframe)?(?:\.html)?(?:#.*)?$'
     _TESTS = [{
         'url': 'http://www.spiegel.de/video/vulkan-tungurahua-in-ecuador-ist-wieder-aktiv-video-1259285.html',
-        'md5': '2c2754212136f35fb4b19767d242f66e',
+        'md5': 'b57399839d055fccfeb9a0455c439868',
         'info_dict': {
-            'id': '1259285',
+            'id': '563747',
             'ext': 'mp4',
             'title': 'Vulkanausbruch in Ecuador: Der "Feuerschlund" ist wieder aktiv',
             'description': 'md5:8029d8310232196eb235d27575a8b9f4',
             'duration': 49,
             'upload_date': '20130311',
+            'timestamp': 1362994320,
         },
     }, {
         'url': 'http://www.spiegel.de/video/schach-wm-videoanalyse-des-fuenften-spiels-video-1309159.html',
-        'md5': 'f2cdf638d7aa47654e251e1aee360af1',
+        'md5': '5b6c2f4add9d62912ed5fc78a1faed80',
         'info_dict': {
-            'id': '1309159',
+            'id': '580988',
             'ext': 'mp4',
             'title': 'Schach-WM in der Videoanalyse: Carlsen nutzt die Fehlgriffe des Titelverteidigers',
             'description': 'md5:c2322b65e58f385a820c10fa03b2d088',
             'duration': 983,
             'upload_date': '20131115',
+            'timestamp': 1384546642,
         },
     }, {
         'url': 'http://www.spiegel.de/video/astronaut-alexander-gerst-von-der-iss-station-beantwortet-fragen-video-1519126-embed.html',
-        'md5': 'd8eeca6bfc8f1cd6f490eb1f44695d51',
+        'md5': '97b91083a672d72976faa8433430afb9',
         'info_dict': {
-            'id': '1519126',
+            'id': '601883',
             'ext': 'mp4',
             'description': 'SPIEGEL ONLINE-Nutzer durften den deutschen Astronauten Alexander Gerst über sein Leben auf der ISS-Station befragen. Hier kommen seine Antworten auf die besten sechs Fragen.',
             'title': 'Fragen an Astronaut Alexander Gerst: "Bekommen Sie die Tageszeiten mit?"',
             'upload_date': '20140904',
+            'timestamp': 1409834160,
         }
     }, {
         'url': 'http://www.spiegel.de/video/astronaut-alexander-gerst-von-der-iss-station-beantwortet-fragen-video-1519126-iframe.html',
         'only_matching': True,
+    }, {
+        # nexx video
+        'url': 'http://www.spiegel.de/video/spiegel-tv-magazin-ueber-guellekrise-in-schleswig-holstein-video-99012776.html',
+        'only_matching': True,
     }]
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
-        webpage, handle = self._download_webpage_handle(url, video_id)
+        metadata_url = 'http://www.spiegel.de/video/metadata/video-%s.json' % video_id
+        handle = self._request_webpage(metadata_url, video_id)
 
         # 302 to spiegel.tv, like http://www.spiegel.de/video/der-film-zum-wochenende-die-wahrheit-ueber-maenner-video-99003272.html
         if SpiegeltvIE.suitable(handle.geturl()):
             return self.url_result(handle.geturl(), 'Spiegeltv')
 
-        video_data = extract_attributes(self._search_regex(r'(<div[^>]+id="spVideoElements"[^>]+>)', webpage, 'video element', default=''))
-
-        title = video_data.get('data-video-title') or get_element_by_attribute('class', 'module-title', webpage)
-        description = video_data.get('data-video-teaser') or self._html_search_meta('description', webpage, 'description')
-
-        base_url = self._search_regex(
-            [r'server\s*:\s*(["\'])(?P<url>.+?)\1', r'var\s+server\s*=\s*"(?P<url>[^"]+)\"'],
-            webpage, 'server URL', group='url')
-
-        xml_url = base_url + video_id + '.xml'
-        idoc = self._download_xml(xml_url, video_id)
-
-        formats = []
-        for n in list(idoc):
-            if n.tag.startswith('type') and n.tag != 'type6':
-                format_id = n.tag.rpartition('type')[2]
-                video_url = base_url + n.find('./filename').text
-                formats.append({
-                    'format_id': format_id,
-                    'url': video_url,
-                    'width': int(n.find('./width').text),
-                    'height': int(n.find('./height').text),
-                    'abr': int(n.find('./audiobitrate').text),
-                    'vbr': int(n.find('./videobitrate').text),
-                    'vcodec': n.find('./codec').text,
-                    'acodec': 'MP4A',
-                })
-        duration = float(idoc[0].findall('./duration')[0].text)
-
-        self._check_formats(formats, video_id)
-        self._sort_formats(formats)
+        video_data = self._parse_json(self._webpage_read_content(
+            handle, metadata_url, video_id), video_id)
+        title = video_data['title']
+        nexx_id = video_data['nexxOmniaId']
+        domain_id = video_data.get('nexxOmniaDomain') or '748'
 
         return {
+            '_type': 'url_transparent',
             'id': video_id,
+            'url': 'nexx:%s:%s' % (domain_id, nexx_id),
             'title': title,
-            'description': description.strip() if description else None,
-            'duration': duration,
-            'upload_date': unified_strdate(video_data.get('data-video-date')),
-            'formats': formats,
+            'description': strip_or_none(video_data.get('teaser')),
+            'duration': parse_duration(video_data.get('duration')),
+            'timestamp': unified_timestamp(video_data.get('datum')),
+            'ie_key': NexxIE.ie_key(),
         }
 
 
@@ -121,6 +109,26 @@ class SpiegelArticleIE(InfoExtractor):
 
         },
         'playlist_count': 6,
+    }, {
+        # Nexx iFrame embed
+        'url': 'http://www.spiegel.de/sptv/spiegeltv/spiegel-tv-ueber-schnellste-katapult-achterbahn-der-welt-taron-a-1137884.html',
+        'info_dict': {
+            'id': '161464',
+            'ext': 'mp4',
+            'title': 'Nervenkitzel Achterbahn',
+            'alt_title': 'Karussellbauer in Deutschland',
+            'description': 'md5:ffe7b1cc59a01f585e0569949aef73cc',
+            'release_year': 2005,
+            'creator': 'SPIEGEL TV',
+            'thumbnail': r're:^https?://.*\.jpg$',
+            'duration': 2761,
+            'timestamp': 1394021479,
+            'upload_date': '20140305',
+        },
+        'params': {
+            'format': 'bestvideo',
+            'skip_download': True,
+        },
     }]
 
     def _real_extract(self, url):
@@ -143,6 +151,9 @@ class SpiegelArticleIE(InfoExtractor):
         entries = [
             self.url_result(compat_urlparse.urljoin(
                 self.http_scheme() + '//spiegel.de/', embed_path))
-            for embed_path in embeds
-        ]
-        return self.playlist_result(entries)
+            for embed_path in embeds]
+        if embeds:
+            return self.playlist_result(entries)
+
+        return self.playlist_from_matches(
+            NexxEmbedIE._extract_urls(webpage), ie=NexxEmbedIE.ie_key())