[Iaxclient-devel] Re: Bugfix in libiax2

[Mark S. <mar...@di...>]

Thanks, it's fixed.  It was correct in asterisk already, fortunately.

Mark

On Fri, 20 Feb 2004, Steve Kann wrote:

>
> Iaxclient-developers,
>
>     We just fixed a bug we found in libiax2, which caused it to never
> send full voice frames.  The basic problem is that the logic to check to
> see if we should send a full frame basically looks to see if the top
> half of the timestamp has changed, but it wasn't using the timestamp of
> the previous frame for the comparison, it was using the timestamp of the
> current frame.  Therefore, the top half will always be equal, and full
> frames will never be sent.
>
>     I'm not sure how this affects some of the problems that people were
> having with transferred calls, but it causes havoc to the jitterbuffer
> mechanism on asterisk servers.  This will likely screw up any calls to
> an asterisk server which has the jitterbuffer enabled (which is no
> longer the default, but seems pretty important for most internet calls).
>
>     The patch is below, and has been committed to sourceforge CVS.
>
>     Mark:  Please also apply this to digium libiax2 CVS if it looks
> correct to you.
>
> -SteveK
>
>
> Log Message:
> calc_timestamp modifies pvt->lastsent.  Therefore, the next test is always true, (since ts will ==
> lastsent), and we'll never send full audio frames.
>
> If we don't send full audio frames, we quickly screw up the jitterbuffer mechanism on the receiving
> asterisk server, because the timestamp wraps, and nothing ever corrects it.
>
> This fix follows the way this section of code works in chan_iax2.c
>
> -SK/JO
>
>
>
> Index: iax.c
> ===================================================================
> RCS file: /cvsroot/iaxclient/iaxclient/lib/libiax2/src/iax.c,v
> retrieving revision 1.13
> retrieving revision 1.14
> diff -u -d -r1.13 -r1.14
> --- iax.c	13 Feb 2004 22:20:01 -0000	1.13
> +++ iax.c	20 Feb 2004 22:26:51 -0000	1.14
> @@ -631,9 +631,12 @@
>  		return -1;
>  	}
>
> +	/* this must come before the next call to calc_timestamp() since
> +	 calc_timestamp() will change lastsent to the returned value */
> +	lastsent = pvt->lastsent;
> +
>  	/* Calculate actual timestamp */
>  	fts = calc_timestamp(pvt, ts);
> -	lastsent = pvt->lastsent;
>
>  	if (((fts & 0xFFFF0000L) == (lastsent & 0xFFFF0000L))
>  		/* High two bits are the same on timestamp, or sending on a trunk */ &&
>
>
>
>