]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/myspace.py
debian/patches: Refresh patches to apply cleanly.
[youtubedl] / youtube_dl / extractor / myspace.py
index 050f54a5a1653e3a71662eac566819525d2d590a..e164d5940044fc57bd9d760927b8f83c638fa198 100644 (file)
+# coding: utf-8
+from __future__ import unicode_literals
+
 import re
-import json
 
 from .common import InfoExtractor
 from ..utils import (
-    compat_str,
+    ExtractorError,
+    int_or_none,
+    parse_iso8601,
 )
 
 
 class MySpaceIE(InfoExtractor):
-    _VALID_URL = r'https?://myspace\.com/([^/]+)/video/[^/]+/(?P<id>\d+)'
-
-    _TEST = {
-        u'url': u'https://myspace.com/coldplay/video/viva-la-vida/100008689',
-        u'info_dict': {
-            u'id': u'100008689',
-            u'ext': u'flv',
-            u'title': u'Viva La Vida',
-            u'description': u'The official Viva La Vida video, directed by Hype Williams',
-            u'uploader': u'Coldplay',
-            u'uploader_id': u'coldplay',
+    _VALID_URL = r'''(?x)
+                    https?://
+                        myspace\.com/[^/]+/
+                        (?P<mediatype>
+                            video/[^/]+/(?P<video_id>\d+)|
+                            music/song/[^/?#&]+-(?P<song_id>\d+)-\d+(?:[/?#&]|$)
+                        )
+                    '''
+
+    _TESTS = [{
+        'url': 'https://myspace.com/fiveminutestothestage/video/little-big-town/109594919',
+        'md5': '9c1483c106f4a695c47d2911feed50a7',
+        'info_dict': {
+            'id': '109594919',
+            'ext': 'mp4',
+            'title': 'Little Big Town',
+            'description': 'This country quartet was all smiles while playing a sold out show at the Pacific Amphitheatre in Orange County, California.',
+            'uploader': 'Five Minutes to the Stage',
+            'uploader_id': 'fiveminutestothestage',
+            'timestamp': 1414108751,
+            'upload_date': '20141023',
         },
-        u'params': {
-            # rtmp download
-            u'skip_download': True,
+    }, {
+        # songs
+        'url': 'https://myspace.com/killsorrow/music/song/of-weakened-soul...-93388656-103880681',
+        'md5': '1d7ee4604a3da226dd69a123f748b262',
+        'info_dict': {
+            'id': '93388656',
+            'ext': 'm4a',
+            'title': 'Of weakened soul...',
+            'uploader': 'Killsorrow',
+            'uploader_id': 'killsorrow',
         },
-    }
+    }, {
+        'add_ie': ['Youtube'],
+        'url': 'https://myspace.com/threedaysgrace/music/song/animal-i-have-become-28400208-28218041',
+        'info_dict': {
+            'id': 'xqds0B_meys',
+            'ext': 'webm',
+            'title': 'Three Days Grace - Animal I Have Become',
+            'description': 'md5:8bd86b3693e72a077cf863a8530c54bb',
+            'uploader': 'ThreeDaysGraceVEVO',
+            'uploader_id': 'ThreeDaysGraceVEVO',
+            'upload_date': '20091002',
+        },
+    }, {
+        'url': 'https://myspace.com/starset2/music/song/first-light-95799905-106964426',
+        'only_matching': True,
+    }, {
+        'url': 'https://myspace.com/thelargemouthbassband/music/song/02-pure-eyes.mp3-94422330-105113388',
+        'only_matching': True,
+    }]
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
-        video_id = mobj.group('id')
+        video_id = mobj.group('video_id') or mobj.group('song_id')
+        is_song = mobj.group('mediatype').startswith('music/song')
         webpage = self._download_webpage(url, video_id)
-        context = json.loads(self._search_regex(r'context = ({.*?});', webpage,
-            u'context'))
-        video = context['video']
-        rtmp_url, play_path = video['streamUrl'].split(';', 1)
+        player_url = self._search_regex(
+            r'videoSwf":"([^"?]*)', webpage, 'player URL', fatal=False)
+
+        def formats_from_stream_urls(stream_url, hls_stream_url, http_stream_url, width=None, height=None):
+            formats = []
+            vcodec = 'none' if is_song else None
+            if hls_stream_url:
+                formats.append({
+                    'format_id': 'hls',
+                    'url': hls_stream_url,
+                    'protocol': 'm3u8_native',
+                    'ext': 'm4a' if is_song else 'mp4',
+                    'vcodec': vcodec,
+                })
+            if stream_url and player_url:
+                rtmp_url, play_path = stream_url.split(';', 1)
+                formats.append({
+                    'format_id': 'rtmp',
+                    'url': rtmp_url,
+                    'play_path': play_path,
+                    'player_url': player_url,
+                    'protocol': 'rtmp',
+                    'ext': 'flv',
+                    'width': width,
+                    'height': height,
+                    'vcodec': vcodec,
+                })
+            if http_stream_url:
+                formats.append({
+                    'format_id': 'http',
+                    'url': http_stream_url,
+                    'width': width,
+                    'height': height,
+                    'vcodec': vcodec,
+                })
+            return formats
+
+        if is_song:
+            # songs don't store any useful info in the 'context' variable
+            song_data = self._search_regex(
+                r'''<button.*data-song-id=(["\'])%s\1.*''' % video_id,
+                webpage, 'song_data', default=None, group=0)
+            if song_data is None:
+                # some songs in an album are not playable
+                self.report_warning(
+                    '%s: No downloadable song on this page' % video_id)
+                return
+
+            def search_data(name):
+                return self._search_regex(
+                    r'''data-%s=([\'"])(?P<data>.*?)\1''' % name,
+                    song_data, name, default='', group='data')
+            formats = formats_from_stream_urls(
+                search_data('stream-url'), search_data('hls-stream-url'),
+                search_data('http-stream-url'))
+            if not formats:
+                vevo_id = search_data('vevo-id')
+                youtube_id = search_data('youtube-id')
+                if vevo_id:
+                    self.to_screen('Vevo video detected: %s' % vevo_id)
+                    return self.url_result('vevo:%s' % vevo_id, ie='Vevo')
+                elif youtube_id:
+                    self.to_screen('Youtube video detected: %s' % youtube_id)
+                    return self.url_result(youtube_id, ie='Youtube')
+                else:
+                    raise ExtractorError(
+                        'Found song but don\'t know how to download it')
+            self._sort_formats(formats)
+            return {
+                'id': video_id,
+                'title': self._og_search_title(webpage),
+                'uploader': search_data('artist-name'),
+                'uploader_id': search_data('artist-username'),
+                'thumbnail': self._og_search_thumbnail(webpage),
+                'duration': int_or_none(search_data('duration')),
+                'formats': formats,
+            }
+        else:
+            video = self._parse_json(self._search_regex(
+                r'context = ({.*?});', webpage, 'context'),
+                video_id)['video']
+            formats = formats_from_stream_urls(
+                video.get('streamUrl'), video.get('hlsStreamUrl'),
+                video.get('mp4StreamUrl'), int_or_none(video.get('width')),
+                int_or_none(video.get('height')))
+            self._sort_formats(formats)
+            return {
+                'id': video_id,
+                'title': video['title'],
+                'description': video.get('description'),
+                'thumbnail': video.get('imageUrl'),
+                'uploader': video.get('artistName'),
+                'uploader_id': video.get('artistUsername'),
+                'duration': int_or_none(video.get('duration')),
+                'timestamp': parse_iso8601(video.get('dateAdded')),
+                'formats': formats,
+            }
+
 
+class MySpaceAlbumIE(InfoExtractor):
+    IE_NAME = 'MySpace:album'
+    _VALID_URL = r'https?://myspace\.com/([^/]+)/music/album/(?P<title>.*-)(?P<id>\d+)'
+
+    _TESTS = [{
+        'url': 'https://myspace.com/starset2/music/album/transmissions-19455773',
+        'info_dict': {
+            'title': 'Transmissions',
+            'id': '19455773',
+        },
+        'playlist_count': 14,
+        'skip': 'this album is only available in some countries',
+    }, {
+        'url': 'https://myspace.com/killsorrow/music/album/the-demo-18596029',
+        'info_dict': {
+            'title': 'The Demo',
+            'id': '18596029',
+        },
+        'playlist_count': 5,
+    }]
+
+    def _real_extract(self, url):
+        mobj = re.match(self._VALID_URL, url)
+        playlist_id = mobj.group('id')
+        display_id = mobj.group('title') + playlist_id
+        webpage = self._download_webpage(url, display_id)
+        tracks_paths = re.findall(r'"music:song" content="(.*?)"', webpage)
+        if not tracks_paths:
+            raise ExtractorError(
+                '%s: No songs found, try using proxy' % display_id,
+                expected=True)
+        entries = [
+            self.url_result(t_path, ie=MySpaceIE.ie_key())
+            for t_path in tracks_paths]
         return {
-            'id': compat_str(video['mediaId']),
-            'title': video['title'],
-            'url': rtmp_url,
-            'play_path': play_path,
-            'ext': 'flv',
-            'description': video['description'],
-            'thumbnail': video['imageUrl'],
-            'uploader': video['artistName'],
-            'uploader_id': video['artistUsername'],
+            '_type': 'playlist',
+            'id': playlist_id,
+            'display_id': display_id,
+            'title': self._og_search_title(webpage),
+            'entries': entries,
         }