]> gitweb @ CieloNegro.org - youtube-dl.git/commitdiff
[npr] fix extraction(closes #10793)(closes #13440)
authorRemita Amine <remitamine@gmail.com>
Tue, 12 Mar 2019 15:52:28 +0000 (16:52 +0100)
committerRemita Amine <remitamine@gmail.com>
Tue, 12 Mar 2019 15:52:28 +0000 (16:52 +0100)
youtube_dl/extractor/npr.py

index 1777aa10b537a8359b8acb4a5b19f5d169cb00d6..a5e8baa7e2542f4e8d6a8c83dea7bddecf82413d 100644 (file)
@@ -1,7 +1,6 @@
 from __future__ import unicode_literals
 
 from .common import InfoExtractor
-from ..compat import compat_urllib_parse_urlencode
 from ..utils import (
     int_or_none,
     qualities,
@@ -9,16 +8,16 @@ from ..utils import (
 
 
 class NprIE(InfoExtractor):
-    _VALID_URL = r'https?://(?:www\.)?npr\.org/player/v2/mediaPlayer\.html\?.*\bid=(?P<id>\d+)'
+    _VALID_URL = r'https?://(?:www\.)?npr\.org/(?:sections/[^/]+/)?\d{4}/\d{2}/\d{2}/(?P<id>\d+)'
     _TESTS = [{
-        'url': 'http://www.npr.org/player/v2/mediaPlayer.html?id=449974205',
+        'url': 'https://www.npr.org/sections/allsongs/2015/10/21/449974205/new-music-from-beach-house-chairlift-cmj-discoveries-and-more',
         'info_dict': {
             'id': '449974205',
             'title': 'New Music From Beach House, Chairlift, CMJ Discoveries And More'
         },
         'playlist_count': 7,
     }, {
-        'url': 'http://www.npr.org/player/v2/mediaPlayer.html?action=1&t=1&islist=false&id=446928052&m=446929930&live=1',
+        'url': 'https://www.npr.org/sections/deceptivecadence/2015/10/09/446928052/music-from-the-shadows-ancient-armenian-hymns-and-piano-jazz',
         'info_dict': {
             'id': '446928052',
             'title': "Songs We Love: Tigran Hamasyan, 'Your Mercy is Boundless'"
@@ -32,30 +31,46 @@ class NprIE(InfoExtractor):
                 'duration': 402,
             },
         }],
+    }, {
+        # mutlimedia, not media title
+        'url': 'https://www.npr.org/2017/06/19/533198237/tigers-jaw-tiny-desk-concert',
+        'info_dict': {
+            'id': '533198237',
+            'title': 'Tigers Jaw: Tiny Desk Concert',
+        },
+        'playlist': [{
+            'md5': '12fa60cb2d3ed932f53609d4aeceabf1',
+            'info_dict': {
+                'id': '533201718',
+                'ext': 'mp4',
+                'title': 'Tigers Jaw: Tiny Desk Concert',
+                'duration': 402,
+            },
+        }],
+        'expected_warnings': ['Failed to download m3u8 information'],
     }]
 
     def _real_extract(self, url):
         playlist_id = self._match_id(url)
 
-        config = self._download_json(
-            'http://api.npr.org/query?%s' % compat_urllib_parse_urlencode({
+        story = self._download_json(
+            'http://api.npr.org/query', playlist_id, query={
                 'id': playlist_id,
-                'fields': 'titles,audio,show',
+                'fields': 'audio,multimedia,title',
                 'format': 'json',
                 'apiKey': 'MDAzMzQ2MjAyMDEyMzk4MTU1MDg3ZmM3MQ010',
-            }), playlist_id)
-
-        story = config['list']['story'][0]
+            })['list']['story'][0]
+        playlist_title = story.get('title', {}).get('$text')
 
-        KNOWN_FORMATS = ('threegp', 'mp4', 'mp3')
+        KNOWN_FORMATS = ('threegp', 'm3u8', 'smil', 'mp4', 'mp3')
         quality = qualities(KNOWN_FORMATS)
 
         entries = []
-        for audio in story.get('audio', []):
-            title = audio.get('title', {}).get('$text')
-            duration = int_or_none(audio.get('duration', {}).get('$text'))
+        for media in story.get('audio', []) + story.get('multimedia', []):
+            media_id = media['id']
+
             formats = []
-            for format_id, formats_entry in audio.get('format', {}).items():
+            for format_id, formats_entry in media.get('format', {}).items():
                 if not formats_entry:
                     continue
                 if isinstance(formats_entry, list):
@@ -64,19 +79,30 @@ class NprIE(InfoExtractor):
                 if not format_url:
                     continue
                 if format_id in KNOWN_FORMATS:
-                    formats.append({
-                        'url': format_url,
-                        'format_id': format_id,
-                        'ext': formats_entry.get('type'),
-                        'quality': quality(format_id),
-                    })
+                    if format_id == 'm3u8':
+                        formats.extend(self._extract_m3u8_formats(
+                            format_url, media_id, 'mp4', 'm3u8_native',
+                            m3u8_id='hls', fatal=False))
+                    elif format_id == 'smil':
+                        smil_formats = self._extract_smil_formats(
+                            format_url, media_id, transform_source=lambda s: s.replace(
+                                'rtmp://flash.npr.org/ondemand/', 'https://ondemand.npr.org/'))
+                        self._check_formats(smil_formats, media_id)
+                        formats.extend(smil_formats)
+                    else:
+                        formats.append({
+                            'url': format_url,
+                            'format_id': format_id,
+                            'quality': quality(format_id),
+                        })
             self._sort_formats(formats)
+
             entries.append({
-                'id': audio['id'],
-                'title': title,
-                'duration': duration,
+                'id': media_id,
+                'title': media.get('title', {}).get('$text') or playlist_title,
+                'thumbnail': media.get('altImageUrl', {}).get('$text'),
+                'duration': int_or_none(media.get('duration', {}).get('$text')),
                 'formats': formats,
             })
 
-        playlist_title = story.get('title', {}).get('$text')
         return self.playlist_result(entries, playlist_id, playlist_title)