]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/newgrounds.py
Update upstream source from tag 'upstream/2019.01.16'
[youtubedl] / youtube_dl / extractor / newgrounds.py
index 2e72e8915aab601b6916fd18e4f64090a613986e..82e7cf52216ac431a09d4503caaf7cdb787cd876 100644 (file)
 from __future__ import unicode_literals
 
-import json
 import re
 
 from .common import InfoExtractor
+from ..utils import (
+    extract_attributes,
+    int_or_none,
+    parse_duration,
+    parse_filesize,
+    unified_timestamp,
+)
 
 
 class NewgroundsIE(InfoExtractor):
-    _VALID_URL = r'https?://(?:www\.)?newgrounds\.com/audio/listen/(?P<id>[0-9]+)'
-    _TEST = {
-        'url': 'http://www.newgrounds.com/audio/listen/549479',
+    _VALID_URL = r'https?://(?:www\.)?newgrounds\.com/(?:audio/listen|portal/view)/(?P<id>[0-9]+)'
+    _TESTS = [{
+        'url': 'https://www.newgrounds.com/audio/listen/549479',
         'md5': 'fe6033d297591288fa1c1f780386f07a',
         'info_dict': {
             'id': '549479',
             'ext': 'mp3',
             'title': 'B7 - BusMode',
             'uploader': 'Burn7',
-        }
-    }
+            'timestamp': 1378878540,
+            'upload_date': '20130911',
+            'duration': 143,
+        },
+    }, {
+        'url': 'https://www.newgrounds.com/portal/view/673111',
+        'md5': '3394735822aab2478c31b1004fe5e5bc',
+        'info_dict': {
+            'id': '673111',
+            'ext': 'mp4',
+            'title': 'Dancin',
+            'uploader': 'Squirrelman82',
+            'timestamp': 1460256780,
+            'upload_date': '20160410',
+        },
+    }, {
+        # source format unavailable, additional mp4 formats
+        'url': 'http://www.newgrounds.com/portal/view/689400',
+        'info_dict': {
+            'id': '689400',
+            'ext': 'mp4',
+            'title': 'ZTV News Episode 8',
+            'uploader': 'BennettTheSage',
+            'timestamp': 1487965140,
+            'upload_date': '20170224',
+        },
+        'params': {
+            'skip_download': True,
+        },
+    }]
 
     def _real_extract(self, url):
-        mobj = re.match(self._VALID_URL, url)
-        music_id = mobj.group('id')
-        webpage = self._download_webpage(url, music_id)
-        
+        media_id = self._match_id(url)
+
+        webpage = self._download_webpage(url, media_id)
+
         title = self._html_search_regex(
-            r',"name":"([^"]+)",', webpage, 'music title')
+            r'<title>([^>]+)</title>', webpage, 'title')
+
+        media_url = self._parse_json(self._search_regex(
+            r'"url"\s*:\s*("[^"]+"),', webpage, ''), media_id)
+
+        formats = [{
+            'url': media_url,
+            'format_id': 'source',
+            'quality': 1,
+        }]
+
+        max_resolution = int_or_none(self._search_regex(
+            r'max_resolution["\']\s*:\s*(\d+)', webpage, 'max resolution',
+            default=None))
+        if max_resolution:
+            url_base = media_url.rpartition('.')[0]
+            for resolution in (360, 720, 1080):
+                if resolution > max_resolution:
+                    break
+                formats.append({
+                    'url': '%s.%dp.mp4' % (url_base, resolution),
+                    'format_id': '%dp' % resolution,
+                    'height': resolution,
+                })
+
+        self._check_formats(formats, media_id)
+        self._sort_formats(formats)
+
         uploader = self._html_search_regex(
-            r',"artist":"([^"]+)",', webpage, 'music uploader')
-        
-        music_url_json_string = self._html_search_regex(
-            r'({"url":"[^"]+"),', webpage, 'music url') + '}'
-        music_url_json = json.loads(music_url_json_string)
-        music_url = music_url_json['url']
+            (r'(?s)<h4[^>]*>(.+?)</h4>.*?<em>\s*Author\s*</em>',
+             r'(?:Author|Writer)\s*<a[^>]+>([^<]+)'), webpage, 'uploader',
+            fatal=False)
+
+        timestamp = unified_timestamp(self._html_search_regex(
+            (r'<dt>\s*Uploaded\s*</dt>\s*<dd>([^<]+</dd>\s*<dd>[^<]+)',
+             r'<dt>\s*Uploaded\s*</dt>\s*<dd>([^<]+)'), webpage, 'timestamp',
+            default=None))
+        duration = parse_duration(self._search_regex(
+            r'(?s)<dd>\s*Song\s*</dd>\s*<dd>.+?</dd>\s*<dd>([^<]+)', webpage,
+            'duration', default=None))
+
+        filesize_approx = parse_filesize(self._html_search_regex(
+            r'(?s)<dd>\s*Song\s*</dd>\s*<dd>(.+?)</dd>', webpage, 'filesize',
+            default=None))
+        if len(formats) == 1:
+            formats[0]['filesize_approx'] = filesize_approx
+
+        if '<dd>Song' in webpage:
+            formats[0]['vcodec'] = 'none'
 
         return {
-            'id': music_id,
+            'id': media_id,
             'title': title,
-            'url': music_url,
             'uploader': uploader,
+            'timestamp': timestamp,
+            'duration': duration,
+            'formats': formats,
         }
+
+
+class NewgroundsPlaylistIE(InfoExtractor):
+    _VALID_URL = r'https?://(?:www\.)?newgrounds\.com/(?:collection|[^/]+/search/[^/]+)/(?P<id>[^/?#&]+)'
+    _TESTS = [{
+        'url': 'https://www.newgrounds.com/collection/cats',
+        'info_dict': {
+            'id': 'cats',
+            'title': 'Cats',
+        },
+        'playlist_mincount': 46,
+    }, {
+        'url': 'http://www.newgrounds.com/portal/search/author/ZONE-SAMA',
+        'info_dict': {
+            'id': 'ZONE-SAMA',
+            'title': 'Portal Search: ZONE-SAMA',
+        },
+        'playlist_mincount': 47,
+    }, {
+        'url': 'http://www.newgrounds.com/audio/search/title/cats',
+        'only_matching': True,
+    }]
+
+    def _real_extract(self, url):
+        playlist_id = self._match_id(url)
+
+        webpage = self._download_webpage(url, playlist_id)
+
+        title = self._search_regex(
+            r'<title>([^>]+)</title>', webpage, 'title', default=None)
+
+        # cut left menu
+        webpage = self._search_regex(
+            r'(?s)<div[^>]+\bclass=["\']column wide(.+)',
+            webpage, 'wide column', default=webpage)
+
+        entries = []
+        for a, path, media_id in re.findall(
+                r'(<a[^>]+\bhref=["\']/?((?:portal/view|audio/listen)/(\d+))[^>]+>)',
+                webpage):
+            a_class = extract_attributes(a).get('class')
+            if a_class not in ('item-portalsubmission', 'item-audiosubmission'):
+                continue
+            entries.append(
+                self.url_result(
+                    'https://www.newgrounds.com/%s' % path,
+                    ie=NewgroundsIE.ie_key(), video_id=media_id))
+
+        return self.playlist_result(entries, playlist_id, title)