[funk] fix extraction(closes #17915)
authorRemita Amine <remitamine@gmail.com>
Wed, 10 Jul 2019 12:54:49 +0000 (13:54 +0100)
committerRemita Amine <remitamine@gmail.com>
Wed, 10 Jul 2019 12:54:49 +0000 (13:54 +0100)
youtube_dl/extractor/funk.py

index 7e1af95e0fcd86b72115ef60b4b6dabe52f94d77..81d1949fd22fac93e78460e47bd79a6a50ff0105 100644 (file)
@@ -1,89 +1,21 @@
 # coding: utf-8
 from __future__ import unicode_literals
 
-import itertools
 import re
 
 from .common import InfoExtractor
 from .nexx import NexxIE
-from ..compat import compat_str
 from ..utils import (
     int_or_none,
-    try_get,
+    str_or_none,
 )
 
 
-class FunkBaseIE(InfoExtractor):
-    _HEADERS = {
-        'Accept': '*/*',
-        'Accept-Language': 'en-US,en;q=0.9,ru;q=0.8',
-        'authorization': 'eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJjbGllbnROYW1lIjoid2ViYXBwLXYzMSIsInNjb3BlIjoic3RhdGljLWNvbnRlbnQtYXBpLGN1cmF0aW9uLWFwaSxuZXh4LWNvbnRlbnQtYXBpLXYzMSx3ZWJhcHAtYXBpIn0.mbuG9wS9Yf5q6PqgR4fiaRFIagiHk9JhwoKES7ksVX4',
-    }
-    _AUTH = 'eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJjbGllbnROYW1lIjoid2ViYXBwLXYzMSIsInNjb3BlIjoic3RhdGljLWNvbnRlbnQtYXBpLGN1cmF0aW9uLWFwaSxuZXh4LWNvbnRlbnQtYXBpLXYzMSx3ZWJhcHAtYXBpIn0.mbuG9wS9Yf5q6PqgR4fiaRFIagiHk9JhwoKES7ksVX4'
-
-    @staticmethod
-    def _make_headers(referer):
-        headers = FunkBaseIE._HEADERS.copy()
-        headers['Referer'] = referer
-        return headers
-
-    def _make_url_result(self, video):
-        return {
-            '_type': 'url_transparent',
-            'url': 'nexx:741:%s' % video['sourceId'],
-            'ie_key': NexxIE.ie_key(),
-            'id': video['sourceId'],
-            'title': video.get('title'),
-            'description': video.get('description'),
-            'duration': int_or_none(video.get('duration')),
-            'season_number': int_or_none(video.get('seasonNr')),
-            'episode_number': int_or_none(video.get('episodeNr')),
-        }
-
-
-class FunkMixIE(FunkBaseIE):
-    _VALID_URL = r'https?://(?:www\.)?funk\.net/mix/(?P<id>[^/]+)/(?P<alias>[^/?#&]+)'
+class FunkIE(InfoExtractor):
+    _VALID_URL = r'https?://(?:www\.)?funk\.net/(?:channel|playlist)/[^/]+/(?P<display_id>[0-9a-z-]+)-(?P<id>\d+)'
     _TESTS = [{
-        'url': 'https://www.funk.net/mix/59d65d935f8b160001828b5b/die-realste-kifferdoku-aller-zeiten',
-        'md5': '8edf617c2f2b7c9847dfda313f199009',
-        'info_dict': {
-            'id': '123748',
-            'ext': 'mp4',
-            'title': '"Die realste Kifferdoku aller Zeiten"',
-            'description': 'md5:c97160f5bafa8d47ec8e2e461012aa9d',
-            'timestamp': 1490274721,
-            'upload_date': '20170323',
-        },
-    }]
-
-    def _real_extract(self, url):
-        mobj = re.match(self._VALID_URL, url)
-        mix_id = mobj.group('id')
-        alias = mobj.group('alias')
-
-        lists = self._download_json(
-            'https://www.funk.net/api/v3.1/curation/curatedLists/',
-            mix_id, headers=self._make_headers(url), query={
-                'size': 100,
-            })['_embedded']['curatedListList']
-
-        metas = next(
-            l for l in lists
-            if mix_id in (l.get('entityId'), l.get('alias')))['videoMetas']
-        video = next(
-            meta['videoDataDelegate']
-            for meta in metas
-            if try_get(
-                meta, lambda x: x['videoDataDelegate']['alias'],
-                compat_str) == alias)
-
-        return self._make_url_result(video)
-
-
-class FunkChannelIE(FunkBaseIE):
-    _VALID_URL = r'https?://(?:www\.)?funk\.net/channel/(?P<id>[^/]+)/(?P<alias>[^/?#&]+)'
-    _TESTS = [{
-        'url': 'https://www.funk.net/channel/ba/die-lustigsten-instrumente-aus-dem-internet-teil-2',
+        'url': 'https://www.funk.net/channel/ba-793/die-lustigsten-instrumente-aus-dem-internet-teil-2-1155821',
+        'md5': '8dd9d9ab59b4aa4173b3197f2ea48e81',
         'info_dict': {
             'id': '1155821',
             'ext': 'mp4',
@@ -92,83 +24,26 @@ class FunkChannelIE(FunkBaseIE):
             'timestamp': 1514507395,
             'upload_date': '20171229',
         },
-        'params': {
-            'skip_download': True,
-        },
-    }, {
-        # only available via byIdList API
-        'url': 'https://www.funk.net/channel/informr/martin-sonneborn-erklaert-die-eu',
-        'info_dict': {
-            'id': '205067',
-            'ext': 'mp4',
-            'title': 'Martin Sonneborn erklärt die EU',
-            'description': 'md5:050f74626e4ed87edf4626d2024210c0',
-            'timestamp': 1494424042,
-            'upload_date': '20170510',
-        },
-        'params': {
-            'skip_download': True,
-        },
+
     }, {
-        'url': 'https://www.funk.net/channel/59d5149841dca100012511e3/mein-erster-job-lovemilla-folge-1/lovemilla/',
+        'url': 'https://www.funk.net/playlist/neuesteVideos/kameras-auf-dem-fusion-festival-1618699',
         'only_matching': True,
     }]
 
     def _real_extract(self, url):
-        mobj = re.match(self._VALID_URL, url)
-        channel_id = mobj.group('id')
-        alias = mobj.group('alias')
-
-        headers = self._make_headers(url)
-
-        video = None
-
-        # Id-based channels are currently broken on their side: webplayer
-        # tries to process them via byChannelAlias endpoint and fails
-        # predictably.
-        for page_num in itertools.count():
-            by_channel_alias = self._download_json(
-                'https://www.funk.net/api/v3.1/webapp/videos/byChannelAlias/%s'
-                % channel_id,
-                'Downloading byChannelAlias JSON page %d' % (page_num + 1),
-                headers=headers, query={
-                    'filterFsk': 'false',
-                    'sort': 'creationDate,desc',
-                    'size': 100,
-                    'page': page_num,
-                }, fatal=False)
-            if not by_channel_alias:
-                break
-            video_list = try_get(
-                by_channel_alias, lambda x: x['_embedded']['videoList'], list)
-            if not video_list:
-                break
-            try:
-                video = next(r for r in video_list if r.get('alias') == alias)
-                break
-            except StopIteration:
-                pass
-            if not try_get(
-                    by_channel_alias, lambda x: x['_links']['next']):
-                break
-
-        if not video:
-            by_id_list = self._download_json(
-                'https://www.funk.net/api/v3.0/content/videos/byIdList',
-                channel_id, 'Downloading byIdList JSON', headers=headers,
-                query={
-                    'ids': alias,
-                }, fatal=False)
-            if by_id_list:
-                video = try_get(by_id_list, lambda x: x['result'][0], dict)
-
-        if not video:
-            results = self._download_json(
-                'https://www.funk.net/api/v3.0/content/videos/filter',
-                channel_id, 'Downloading filter JSON', headers=headers, query={
-                    'channelId': channel_id,
-                    'size': 100,
-                })['result']
-            video = next(r for r in results if r.get('alias') == alias)
-
-        return self._make_url_result(video)
+        display_id, nexx_id = re.match(self._VALID_URL, url).groups()
+        video = self._download_json(
+            'https://www.funk.net/api/v4.0/videos/' + nexx_id, nexx_id)
+        return {
+            '_type': 'url_transparent',
+            'url': 'nexx:741:' + nexx_id,
+            'ie_key': NexxIE.ie_key(),
+            'id': nexx_id,
+            'title': video.get('title'),
+            'description': video.get('description'),
+            'duration': int_or_none(video.get('duration')),
+            'channel_id': str_or_none(video.get('channelId')),
+            'display_id': display_id,
+            'tags': video.get('tags'),
+            'thumbnail': video.get('imageUrlLandscape'),
+        }