[asterisk-dev] [Code Review] Force directmedia sessions to use a common codec

Terry Wilson twilson at digium.com
Wed Oct 6 00:04:42 CDT 2010

This is an automatically generated e-mail. To reply, visit:

Review request for Asterisk Developers.


>From the issue:

It is "well known" that setting up directmedia paths between SIP devices with different preferred-codec lists can be problematic. The solution has previously been to either disable directmedia, or use the codec-negotiation patch which is available for Asterisk 1.2 and 1.4.

PROBLEM: In its simplest form, the problem is (I believe) as follows:

When RTP tries to set up a direct media path, it does so by setting each party separately, and giving each party free reign over their codec choice. Thus under some circumstances the 2 remote devices can make 2 choices which are different. If the 2 parties disagree on the negotiated codec, you generally get silence in both directions.

SOLUTION: Make rtp.c just a bit more clever about what codecs each party is offered - In fact, do not allow a choice, just tell them what is acceptable.

EXAMPLE: Device A (g722|alaw) calls via asterisk to device B (alaw) - Initially a transcoded path is setup between the 2 parties. directmedia is enabled, and they have a common codec (alaw).

At present, during negotiation, RTP tells A to use (alaw) and B to use (g722|alaw) - chan_sip.c then notes those codec requests, ignores them and tells A to use (g722|alaw) and B to use (alaw)

Of course this results in A using g722, and B using alaw - Resulting in silence.

The patch (to be attached) changes:

main/rtp.c to make it choose a single audio and video codec where possible before passing the directmedia request on to the channels.

channels/chan_sip.c to make it use the passed in codec offering from RTP as long as it is doing a directmedia reinvite - At any other time, old behaviour remains, allowing better codec choices where possible.

This addresses bug 17403.


  /trunk/channels/chan_sip.c 290542 

Diff: https://reviewboard.asterisk.org/r/967/diff


I verified the issue by recreating the aforementioned scenario. I ported the patch to trunk and verified that audio was negotiated properly for that scenario.



More information about the asterisk-dev mailing list