]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/acast.py
debian/changelog: Annotate with bug being closed.
[youtubedl] / youtube_dl / extractor / acast.py
index 6dace305141423ec35d25abe3c4283476a762b2f..b17c792d23a824bb3684d250071efdf42135a2b0 100644 (file)
@@ -7,72 +7,109 @@ import functools
 from .common import InfoExtractor
 from ..compat import compat_str
 from ..utils import (
+    clean_html,
+    float_or_none,
     int_or_none,
-    parse_iso8601,
+    try_get,
+    unified_timestamp,
     OnDemandPagedList,
 )
 
 
 class ACastIE(InfoExtractor):
     IE_NAME = 'acast'
-    _VALID_URL = r'https?://(?:www\.)?acast\.com/(?P<channel>[^/]+)/(?P<id>[^/#?]+)'
+    _VALID_URL = r'''(?x)
+                    https?://
+                        (?:
+                            (?:(?:embed|www)\.)?acast\.com/|
+                            play\.acast\.com/s/
+                        )
+                        (?P<channel>[^/]+)/(?P<id>[^/#?]+)
+                    '''
     _TESTS = [{
-        # test with one bling
-        'url': 'https://www.acast.com/condenasttraveler/-where-are-you-taipei-101-taiwan',
-        'md5': 'ada3de5a1e3a2a381327d749854788bb',
-        'info_dict': {
-            'id': '57de3baa-4bb0-487e-9418-2692c1277a34',
-            'ext': 'mp3',
-            'title': '"Where Are You?": Taipei 101, Taiwan',
-            'timestamp': 1196172000,
-            'upload_date': '20071127',
-            'description': 'md5:a0b4ef3634e63866b542e5b1199a1a0e',
-            'duration': 211,
-        }
-    }, {
-        # test with multiple blings
         'url': 'https://www.acast.com/sparpodcast/2.raggarmordet-rosterurdetforflutna',
-        'md5': '55c0097badd7095f494c99a172f86501',
+        'md5': '16d936099ec5ca2d5869e3a813ee8dc4',
         'info_dict': {
             'id': '2a92b283-1a75-4ad8-8396-499c641de0d9',
             'ext': 'mp3',
             'title': '2. Raggarmordet - Röster ur det förflutna',
+            'description': 'md5:4f81f6d8cf2e12ee21a321d8bca32db4',
             'timestamp': 1477346700,
             'upload_date': '20161024',
-            'description': 'md5:4f81f6d8cf2e12ee21a321d8bca32db4',
-            'duration': 2797,
+            'duration': 2766.602563,
+            'creator': 'Anton Berg & Martin Johnson',
+            'series': 'Spår',
+            'episode': '2. Raggarmordet - Röster ur det förflutna',
         }
+    }, {
+        'url': 'http://embed.acast.com/adambuxton/ep.12-adam-joeschristmaspodcast2015',
+        'only_matching': True,
+    }, {
+        'url': 'https://play.acast.com/s/rattegangspodden/s04e09-styckmordet-i-helenelund-del-22',
+        'only_matching': True,
+    }, {
+        'url': 'https://play.acast.com/s/sparpodcast/2a92b283-1a75-4ad8-8396-499c641de0d9',
+        'only_matching': True,
     }]
 
     def _real_extract(self, url):
         channel, display_id = re.match(self._VALID_URL, url).groups()
+        s = self._download_json(
+            'https://feeder.acast.com/api/v1/shows/%s/episodes/%s' % (channel, display_id),
+            display_id)
+        media_url = s['url']
+        if re.search(r'[0-9a-f]{8}-(?:[0-9a-f]{4}-){3}[0-9a-f]{12}', display_id):
+            episode_url = s.get('episodeUrl')
+            if episode_url:
+                display_id = episode_url
+            else:
+                channel, display_id = re.match(self._VALID_URL, s['link']).groups()
         cast_data = self._download_json(
-            'https://embed.acast.com/api/acasts/%s/%s' % (channel, display_id), display_id)
+            'https://play-api.acast.com/splash/%s/%s' % (channel, display_id),
+            display_id)['result']
+        e = cast_data['episode']
+        title = e.get('name') or s['title']
         return {
-            'id': compat_str(cast_data['id']),
+            'id': compat_str(e['id']),
             'display_id': display_id,
-            'url': [b['audio'] for b in cast_data['blings'] if b['type'] == 'BlingAudio'][0],
-            'title': cast_data['name'],
-            'description': cast_data.get('description'),
-            'thumbnail': cast_data.get('image'),
-            'timestamp': parse_iso8601(cast_data.get('publishingDate')),
-            'duration': int_or_none(cast_data.get('duration')),
+            'url': media_url,
+            'title': title,
+            'description': e.get('summary') or clean_html(e.get('description') or s.get('description')),
+            'thumbnail': e.get('image'),
+            'timestamp': unified_timestamp(e.get('publishingDate') or s.get('publishDate')),
+            'duration': float_or_none(e.get('duration') or s.get('duration')),
+            'filesize': int_or_none(e.get('contentLength')),
+            'creator': try_get(cast_data, lambda x: x['show']['author'], compat_str),
+            'series': try_get(cast_data, lambda x: x['show']['name'], compat_str),
+            'season_number': int_or_none(e.get('seasonNumber')),
+            'episode': title,
+            'episode_number': int_or_none(e.get('episodeNumber')),
         }
 
 
 class ACastChannelIE(InfoExtractor):
     IE_NAME = 'acast:channel'
-    _VALID_URL = r'https?://(?:www\.)?acast\.com/(?P<id>[^/#?]+)'
-    _TEST = {
-        'url': 'https://www.acast.com/condenasttraveler',
+    _VALID_URL = r'''(?x)
+                    https?://
+                        (?:
+                            (?:www\.)?acast\.com/|
+                            play\.acast\.com/s/
+                        )
+                        (?P<id>[^/#?]+)
+                    '''
+    _TESTS = [{
+        'url': 'https://www.acast.com/todayinfocus',
         'info_dict': {
-            'id': '50544219-29bb-499e-a083-6087f4cb7797',
-            'title': 'Condé Nast Traveler Podcast',
-            'description': 'md5:98646dee22a5b386626ae31866638fbd',
+            'id': '4efc5294-5385-4847-98bd-519799ce5786',
+            'title': 'Today in Focus',
+            'description': 'md5:9ba5564de5ce897faeb12963f4537a64',
         },
-        'playlist_mincount': 20,
-    }
-    _API_BASE_URL = 'https://www.acast.com/api/'
+        'playlist_mincount': 35,
+    }, {
+        'url': 'http://play.acast.com/s/ft-banking-weekly',
+        'only_matching': True,
+    }]
+    _API_BASE_URL = 'https://play.acast.com/api/'
     _PAGE_SIZE = 10
 
     @classmethod
@@ -85,7 +122,7 @@ class ACastChannelIE(InfoExtractor):
             channel_slug, note='Download page %d of channel data' % page)
         for cast in casts:
             yield self.url_result(
-                'https://www.acast.com/%s/%s' % (channel_slug, cast['url']),
+                'https://play.acast.com/s/%s/%s' % (channel_slug, cast['url']),
                 'ACast', cast['id'])
 
     def _real_extract(self, url):