]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/chaturbate.py
Set upstream metadata fields: Bug-Database, Bug-Submit, Repository, Repository-Browse.
[youtubedl] / youtube_dl / extractor / chaturbate.py
index 8fbc91c1fbae17f8c46adfe1cb947ff64f5d11b4..a459dcb8d5fbedc00f068da023392f0898d687a2 100644 (file)
@@ -3,11 +3,15 @@ from __future__ import unicode_literals
 import re
 
 from .common import InfoExtractor
-from ..utils import ExtractorError
+from ..utils import (
+    ExtractorError,
+    lowercase_escape,
+    url_or_none,
+)
 
 
 class ChaturbateIE(InfoExtractor):
-    _VALID_URL = r'https?://(?:[^/]+\.)?chaturbate\.com/(?P<id>[^/?#]+)'
+    _VALID_URL = r'https?://(?:[^/]+\.)?chaturbate\.com/(?:fullvideo/?\?.*?\bb=)?(?P<id>[^/?&#]+)'
     _TESTS = [{
         'url': 'https://www.chaturbate.com/siswet19/',
         'info_dict': {
@@ -21,6 +25,9 @@ class ChaturbateIE(InfoExtractor):
             'skip_download': True,
         },
         'skip': 'Room is offline',
+    }, {
+        'url': 'https://chaturbate.com/fullvideo/?b=caylin',
+        'only_matching': True,
     }, {
         'url': 'https://en.chaturbate.com/siswet19/',
         'only_matching': True,
@@ -31,12 +38,40 @@ class ChaturbateIE(InfoExtractor):
     def _real_extract(self, url):
         video_id = self._match_id(url)
 
-        webpage = self._download_webpage(url, video_id)
+        webpage = self._download_webpage(
+            'https://chaturbate.com/%s/' % video_id, video_id,
+            headers=self.geo_verification_headers())
+
+        found_m3u8_urls = []
+
+        data = self._parse_json(
+            self._search_regex(
+                r'initialRoomDossier\s*=\s*(["\'])(?P<value>(?:(?!\1).)+)\1',
+                webpage, 'data', default='{}', group='value'),
+            video_id, transform_source=lowercase_escape, fatal=False)
+        if data:
+            m3u8_url = url_or_none(data.get('hls_source'))
+            if m3u8_url:
+                found_m3u8_urls.append(m3u8_url)
+
+        if not found_m3u8_urls:
+            for m in re.finditer(
+                    r'(\\u002[27])(?P<url>http.+?\.m3u8.*?)\1', webpage):
+                found_m3u8_urls.append(lowercase_escape(m.group('url')))
+
+        if not found_m3u8_urls:
+            for m in re.finditer(
+                    r'(["\'])(?P<url>http.+?\.m3u8.*?)\1', webpage):
+                found_m3u8_urls.append(m.group('url'))
 
-        m3u8_formats = [(m.group('id').lower(), m.group('url')) for m in re.finditer(
-            r'hlsSource(?P<id>.+?)\s*=\s*(?P<q>["\'])(?P<url>http.+?)(?P=q)', webpage)]
+        m3u8_urls = []
+        for found_m3u8_url in found_m3u8_urls:
+            m3u8_fast_url, m3u8_no_fast_url = found_m3u8_url, found_m3u8_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_formats:
+        if not m3u8_urls:
             error = self._search_regex(
                 [r'<span[^>]+class=(["\'])desc_span\1[^>]*>(?P<error>[^<]+)</span>',
                  r'<div[^>]+id=(["\'])defchat\1[^>]*>\s*<p><strong>(?P<error>[^<]+)<'],
@@ -50,7 +85,13 @@ class ChaturbateIE(InfoExtractor):
             raise ExtractorError('Unable to find stream URL')
 
         formats = []
-        for m3u8_id, m3u8_url in m3u8_formats:
+        for m3u8_url in m3u8_urls:
+            for known_id in ('fast', 'slow'):
+                if '_%s' % known_id in m3u8_url:
+                    m3u8_id = known_id
+                    break
+            else:
+                m3u8_id = None
             formats.extend(self._extract_m3u8_formats(
                 m3u8_url, video_id, ext='mp4',
                 # ffmpeg skips segments for fast m3u8