instead of assuming it should occur right at given position.
This helps finding human-readable error descriptions in RTMP server replies.

Originally committed as revision 20575 to svn://svn.ffmpeg.org/ffmpeg/trunk