From: Reinhard Nissl.

The fix stopps dropping properly received data.
I realized the problem when switching channels in VDR. From time to time, it happend that switching took more than 50 ms (the timeout value used in _x_read_abort()), which caused the select() to timeout.
In the case where 'stream->demux_action_pending' was set, the already received data was dropped due to returning 0. This finally resulted in your demuxer to fail with DEMUX_END, as it couldn't read 6 bytes from the PES header.
So for the case, that there is some data available, it now takes twice the timeout (i. e. 100 ms), before a return 0 happens.
Some more information: as in rare cases switching channels could take even longer than 100 ms, I've also added a loop around _x_read_about() in my VDR input plugin, which returns 0 just after this state lasts for 5 seconds.

CVS patchset: 6557
CVS date: 2004/05/16 21:45:24

1 files changed, 2 insertions, 2 deletions

diff --git a/src/xine-engine/demux.c b/src/xine-engine/demux.c
index 0708d3e51..6399c0776 100644
--- a/src/xine-engine/demux.c
+++ b/src/xine-engine/demux.c
@@ -20,7 +20,7 @@
  * Demuxer helper functions
  * hide some xine engine details from demuxers and reduce code duplication
  *
- * $Id: demux.c,v 1.48 2004/04/07 18:10:20 valtri Exp $ 
+ * $Id: demux.c,v 1.49 2004/05/16 21:45:24 jcdutton Exp $ 
  */
 
 
@@ -431,7 +431,7 @@ off_t _x_read_abort (xine_stream_t *stream, int fd, char *buf, off_t todo) {
          * cannot be stopped when no more data is available.
          */
         if( stream->demux_action_pending )
-          return 0;
+          return total;
       } else {
         break;
       }