Only download 1 sec. with rtmpdump in test mode
authorrzhxeo <rzhxeot7z81b4700@mailcatch.com>
Sun, 6 Oct 2013 18:55:24 +0000 (20:55 +0200)
committerrzhxeo <rzhxeot7z81b4700@mailcatch.com>
Sun, 6 Oct 2013 18:55:24 +0000 (20:55 +0200)
youtube_dl/FileDownloader.py

index d6673fd3ab88d543086417f903cf2ff7d4019944..f1ff0b5202f9ecb8074307e42111cee99b10df2c 100644 (file)
@@ -267,7 +267,7 @@ class FileDownloader(object):
             self.to_screen(u'\r%s[download] 100%% of %s in %s' %
                 (clear_line, data_len_str, self.format_seconds(tot_time)))
 
-    def _download_with_rtmpdump(self, filename, url, player_url, page_url, play_path, tc_url):
+    def _download_with_rtmpdump(self, filename, url, player_url, page_url, play_path, tc_url, test):
         self.report_destination(filename)
         tmpfilename = self.temp_name(filename)
 
@@ -291,6 +291,8 @@ class FileDownloader(object):
             basic_args += ['--playpath', play_path]
         if tc_url is not None:
             basic_args += ['--tcUrl', url]
+        if test:
+            basic_args += ['--stop', '1']
         args = basic_args + [[], ['--resume', '--skip', '1']][self.params.get('continuedl', False)]
         if self.params.get('verbose', False):
             try:
@@ -408,7 +410,8 @@ class FileDownloader(object):
                                                 info_dict.get('player_url', None),
                                                 info_dict.get('page_url', None),
                                                 info_dict.get('play_path', None),
-                                                info_dict.get('tc_url', None))
+                                                info_dict.get('tc_url', None),
+                                                self.params.get('test', False))
 
         # Attempt to download using mplayer
         if url.startswith('mms') or url.startswith('rtsp'):