]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/chaturbate.py
d/p/remove-autoupdate-mechanism.patch: Update patch with metadata at the top.
[youtubedl] / youtube_dl / extractor / chaturbate.py
index 0b67ba67d436e24b85eca9d71d2749f46a114f01..e3eba4be94660759537e969701f9e043fff5b73d 100644 (file)
@@ -1,5 +1,7 @@
 from __future__ import unicode_literals
 
 from __future__ import unicode_literals
 
+import re
+
 from .common import InfoExtractor
 from ..utils import ExtractorError
 
 from .common import InfoExtractor
 from ..utils import ExtractorError
 
@@ -17,33 +19,57 @@ class ChaturbateIE(InfoExtractor):
         },
         'params': {
             'skip_download': True,
         },
         'params': {
             'skip_download': True,
-        }
+        },
+        'skip': 'Room is offline',
     }, {
         'url': 'https://en.chaturbate.com/siswet19/',
         'only_matching': True,
     }]
 
     }, {
         'url': 'https://en.chaturbate.com/siswet19/',
         'only_matching': True,
     }]
 
+    _ROOM_OFFLINE = 'Room is currently offline'
+
     def _real_extract(self, url):
         video_id = self._match_id(url)
 
         webpage = self._download_webpage(url, video_id)
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
 
         webpage = self._download_webpage(url, video_id)
 
-        m3u8_url = self._search_regex(
-            r'src=(["\'])(?P<url>http.+?\.m3u8.*?)\1', webpage,
-            'playlist', default=None, group='url')
+        m3u8_urls = []
+
+        for m in re.finditer(
+                r'(["\'])(?P<url>http.+?\.m3u8.*?)\1', webpage):
+            m3u8_fast_url, m3u8_no_fast_url = m.group('url'), m.group(
+                'url').replace('_fast', '')
+            for m3u8_url in (m3u8_fast_url, m3u8_no_fast_url):
+                if m3u8_url not in m3u8_urls:
+                    m3u8_urls.append(m3u8_url)
 
 
-        if not m3u8_url:
+        if not m3u8_urls:
             error = self._search_regex(
             error = self._search_regex(
-                r'<span[^>]+class=(["\'])desc_span\1[^>]*>(?P<error>[^<]+)</span>',
-                webpage, 'error', group='error')
-            raise ExtractorError(error, expected=True)
+                [r'<span[^>]+class=(["\'])desc_span\1[^>]*>(?P<error>[^<]+)</span>',
+                 r'<div[^>]+id=(["\'])defchat\1[^>]*>\s*<p><strong>(?P<error>[^<]+)<'],
+                webpage, 'error', group='error', default=None)
+            if not error:
+                if any(p in webpage for p in (
+                        self._ROOM_OFFLINE, 'offline_tipping', 'tip_offline')):
+                    error = self._ROOM_OFFLINE
+            if error:
+                raise ExtractorError(error, expected=True)
+            raise ExtractorError('Unable to find stream URL')
 
 
-        formats = self._extract_m3u8_formats(m3u8_url, video_id, ext='mp4')
+        formats = []
+        for m3u8_url in m3u8_urls:
+            m3u8_id = 'fast' if '_fast' in m3u8_url else 'slow'
+            formats.extend(self._extract_m3u8_formats(
+                m3u8_url, video_id, ext='mp4',
+                # ffmpeg skips segments for fast m3u8
+                preference=-10 if m3u8_id == 'fast' else None,
+                m3u8_id=m3u8_id, fatal=False, live=True))
+        self._sort_formats(formats)
 
         return {
             'id': video_id,
             'title': self._live_title(video_id),
 
         return {
             'id': video_id,
             'title': self._live_title(video_id),
-            'thumbnail': 'https://cdn-s.highwebmedia.com/uHK3McUtGCG3SMFcd4ZJsRv8/roomimage/%s.jpg' % video_id,
+            'thumbnail': 'https://roomimg.stream.highwebmedia.com/ri/%s.jpg' % video_id,
             'age_limit': self._rta_search(webpage),
             'is_live': True,
             'formats': formats,
             'age_limit': self._rta_search(webpage),
             'is_live': True,
             'formats': formats,