]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/go.py
Update upstream source from tag 'upstream/2020.03.24'
[youtubedl] / youtube_dl / extractor / go.py
index 4c9be47b4faa0e67baa305d22aadbf43387650bc..03cfba91f00ab6134150090c3088fb7f5d6666d7 100644 (file)
@@ -25,54 +25,143 @@ class GoIE(AdobePassIE):
         },
         'watchdisneychannel': {
             'brand': '004',
-            'requestor_id': 'Disney',
+            'resource_id': 'Disney',
         },
         'watchdisneyjunior': {
             'brand': '008',
-            'requestor_id': 'DisneyJunior',
+            'resource_id': 'DisneyJunior',
         },
         'watchdisneyxd': {
             'brand': '009',
-            'requestor_id': 'DisneyXD',
+            'resource_id': 'DisneyXD',
+        },
+        'disneynow': {
+            'brand': '011',
+            'resource_id': 'Disney',
         }
     }
-    _VALID_URL = r'https?://(?:(?P<sub_domain>%s)\.)?go\.com/(?:[^/]+/)*(?:vdka(?P<id>\w+)|(?:[^/]+/)*(?P<display_id>[^/?#]+))' % '|'.join(_SITE_INFO.keys())
+    _VALID_URL = r'''(?x)
+                    https?://
+                        (?:
+                            (?:(?P<sub_domain>%s)\.)?go|
+                            (?P<sub_domain_2>abc|freeform|disneynow)
+                        )\.com/
+                        (?:
+                            (?:[^/]+/)*(?P<id>[Vv][Dd][Kk][Aa]\w+)|
+                            (?:[^/]+/)*(?P<display_id>[^/?\#]+)
+                        )
+                    ''' % '|'.join(list(_SITE_INFO.keys()))
     _TESTS = [{
-        'url': 'http://abc.go.com/shows/castle/video/most-recent/vdka0_g86w5onx',
+        'url': 'http://abc.go.com/shows/designated-survivor/video/most-recent/VDKA3807643',
         'info_dict': {
-            'id': '0_g86w5onx',
+            'id': 'VDKA3807643',
             'ext': 'mp4',
-            'title': 'Sneak Peek: Language Arts',
-            'description': 'md5:7dcdab3b2d17e5217c953256af964e9c',
+            'title': 'The Traitor in the White House',
+            'description': 'md5:05b009d2d145a1e85d25111bd37222e8',
         },
         'params': {
             # m3u8 download
             'skip_download': True,
         },
+        'skip': 'This content is no longer available.',
     }, {
-        'url': 'http://abc.go.com/shows/after-paradise/video/most-recent/vdka3335601',
-        'only_matching': True,
+        'url': 'http://watchdisneyxd.go.com/doraemon',
+        'info_dict': {
+            'title': 'Doraemon',
+            'id': 'SH55574025',
+        },
+        'playlist_mincount': 51,
+    }, {
+        'url': 'http://freeform.go.com/shows/shadowhunters/episodes/season-2/1-this-guilty-blood',
+        'info_dict': {
+            'id': 'VDKA3609139',
+            'ext': 'mp4',
+            'title': 'This Guilty Blood',
+            'description': 'md5:f18e79ad1c613798d95fdabfe96cd292',
+            'age_limit': 14,
+        },
+        'params': {
+            'geo_bypass_ip_block': '3.244.239.0/24',
+            # m3u8 download
+            'skip_download': True,
+        },
+    }, {
+        'url': 'https://abc.com/shows/the-rookie/episode-guide/season-02/03-the-bet',
+        'info_dict': {
+            'id': 'VDKA13435179',
+            'ext': 'mp4',
+            'title': 'The Bet',
+            'description': 'md5:c66de8ba2e92c6c5c113c3ade84ab404',
+            'age_limit': 14,
+        },
+        'params': {
+            'geo_bypass_ip_block': '3.244.239.0/24',
+            # m3u8 download
+            'skip_download': True,
+        },
     }, {
         'url': 'http://abc.go.com/shows/the-catch/episode-guide/season-01/10-the-wedding',
         'only_matching': True,
     }, {
         'url': 'http://abc.go.com/shows/world-news-tonight/episode-guide/2017-02/17-021717-intense-stand-off-between-man-with-rifle-and-police-in-oakland',
         'only_matching': True,
+    }, {
+        # brand 004
+        'url': 'http://disneynow.go.com/shows/big-hero-6-the-series/season-01/episode-10-mr-sparkles-loses-his-sparkle/vdka4637915',
+        'only_matching': True,
+    }, {
+        # brand 008
+        'url': 'http://disneynow.go.com/shows/minnies-bow-toons/video/happy-campers/vdka4872013',
+        'only_matching': True,
+    }, {
+        'url': 'https://disneynow.com/shows/minnies-bow-toons/video/happy-campers/vdka4872013',
+        'only_matching': True,
     }]
 
+    def _extract_videos(self, brand, video_id='-1', show_id='-1'):
+        display_id = video_id if video_id != '-1' else show_id
+        return self._download_json(
+            'http://api.contents.watchabc.go.com/vp2/ws/contents/3000/videos/%s/001/-1/%s/-1/%s/-1/-1.json' % (brand, show_id, video_id),
+            display_id)['video']
+
     def _real_extract(self, url):
-        sub_domain, video_id, display_id = re.match(self._VALID_URL, url).groups()
-        if not video_id:
-            webpage = self._download_webpage(url, display_id)
+        mobj = re.match(self._VALID_URL, url)
+        sub_domain = mobj.group('sub_domain') or mobj.group('sub_domain_2')
+        video_id, display_id = mobj.group('id', 'display_id')
+        site_info = self._SITE_INFO.get(sub_domain, {})
+        brand = site_info.get('brand')
+        if not video_id or not site_info:
+            webpage = self._download_webpage(url, display_id or video_id)
             video_id = self._search_regex(
-                # There may be inner quotes, e.g. data-video-id="'VDKA3609139'"
-                # from http://freeform.go.com/shows/shadowhunters/episodes/season-2/1-this-guilty-blood
-                r'data-video-id=["\']*VDKA(\w+)', webpage, 'video id')
-        site_info = self._SITE_INFO[sub_domain]
-        brand = site_info['brand']
-        video_data = self._download_json(
-            'http://api.contents.watchabc.go.com/vp2/ws/contents/3000/videos/%s/001/-1/-1/-1/%s/-1/-1.json' % (brand, video_id),
-            video_id)['video'][0]
+                (
+                    # There may be inner quotes, e.g. data-video-id="'VDKA3609139'"
+                    # from http://freeform.go.com/shows/shadowhunters/episodes/season-2/1-this-guilty-blood
+                    r'data-video-id=["\']*(VDKA\w+)',
+                    # https://abc.com/shows/the-rookie/episode-guide/season-02/03-the-bet
+                    r'\b(?:video)?id["\']\s*:\s*["\'](VDKA\w+)'
+                ), webpage, 'video id', default=video_id)
+            if not site_info:
+                brand = self._search_regex(
+                    (r'data-brand=\s*["\']\s*(\d+)',
+                     r'data-page-brand=\s*["\']\s*(\d+)'), webpage, 'brand',
+                    default='004')
+                site_info = next(
+                    si for _, si in self._SITE_INFO.items()
+                    if si.get('brand') == brand)
+            if not video_id:
+                # show extraction works for Disney, DisneyJunior and DisneyXD
+                # ABC and Freeform has different layout
+                show_id = self._search_regex(r'data-show-id=["\']*(SH\d+)', webpage, 'show id')
+                videos = self._extract_videos(brand, show_id=show_id)
+                show_title = self._search_regex(r'data-show-title="([^"]+)"', webpage, 'show title', fatal=False)
+                entries = []
+                for video in videos:
+                    entries.append(self.url_result(
+                        video['url'], 'Go', video.get('id'), video.get('title')))
+                entries.reverse()
+                return self.playlist_result(entries, show_id, show_title)
+        video_data = self._extract_videos(brand, video_id)[0]
+        video_id = video_data['id']
         title = video_data['title']
 
         formats = []
@@ -91,8 +180,8 @@ class GoIE(AdobePassIE):
                     'device': '001',
                 }
                 if video_data.get('accesslevel') == '1':
-                    requestor_id = site_info['requestor_id']
-                    resource = self._get_mvpd_resource(
+                    requestor_id = site_info.get('requestor_id', 'DisneyChannels')
+                    resource = site_info.get('resource_id') or self._get_mvpd_resource(
                         requestor_id, title, video_id, None)
                     auth = self._extract_mvpd_auth(
                         url, video_id, requestor_id, resource)
@@ -102,10 +191,10 @@ class GoIE(AdobePassIE):
                         'adobe_requestor_id': requestor_id,
                     })
                 else:
-                    self._initialize_geo_bypass(['US'])
+                    self._initialize_geo_bypass({'countries': ['US']})
                 entitlement = self._download_json(
                     'https://api.entitlement.watchabc.go.com/vp2/ws-secure/entitlement/2020/authorize.json',
-                    video_id, data=urlencode_postdata(data), headers=self.geo_verification_headers())
+                    video_id, data=urlencode_postdata(data))
                 errors = entitlement.get('errors', {}).get('errors', [])
                 if errors:
                     for error in errors: