zqz/platform-external-webrtc
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove RED/RTX workaround from sender/receiver and VideoEngine2.

Browse Source 
In older Chrome versions, the associated payload type in the RTX header
of retransmitted packets was always set to be the original media payload type,
regardless of the actual payload type of the packet. This meant that packets
encapsulated with RED headers had incorrect payload type information in the
RTX header. Due to an assumption in the receiver, this incorrect payload type
information would effectively be undone, leading to a working system.

Albeit working, this behaviour was undesired, and thus removed. In the interim,
several workarounds were introduced to not destroy interop between old and
new Chrome versions:
  (1) https://codereview.webrtc.org/1649493004
      - If no payload type mapping existed for RED over RTX, the payload type
        of the underlying media would be used.
      - If RED had been negotiated, received RTX packets would always be
        assumed to contain RED.
  (2) https://codereview.webrtc.org/1964473002
      - If RED was removed from the remote description answer, it would be
        disabled in the local receiver as well.
  (3) https://codereview.webrtc.org/2033763002
      - If RED was negotiated in the SDP, it would always be used, regardless
        if ULPFEC was negotiated and used, or not.

Since the Chrome versions that exhibited the original bug now are very old,
this CL removes the workarounds from (1) and (2). In particular, after this
change, we will have the following behaviour:
  - We assume that a payload type mapping for RED over RTX always is set.
    If this is not the case, the RTX packet is not sent.
  - The associated payload type of received RTX packets will always be obeyed.
  - The (non)-existence of RED in the remote description does not affect the
    local receiver.
The workaround in (3) still needs to exist, in order to interop with receivers
that did not have the workarounds in (1) and (2) removed. The change in (3)
can be removed in a couple of Chrome versions.

TESTED=Using AppRTC between patched Chrome (connected to ethernet) and standard Chrome M54 (connected to lossy internal Google WiFi), with and without FEC turned off using AppRTC flag. Also using "Munge SDP" sample on patched Chrome over loopback interface, with 100ms delay and 5% packet loss simulated using tc.
BUG=webrtc:6650

Review-Url: https://codereview.webrtc.org/2469093003
Cr-Commit-Position: refs/heads/master@{#15038}
This commit is contained in:
brandtr
2016-11-11 03:28:30 -08:00
committed by Commit bot
parent ea51d081c8
commit e6f98c7a37
12 changed files with 79 additions and 165 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
webrtc/modules/rtp_rtcp/source/rtp_sender.cc
View File

@ -1158,16 +1158,12 @@ std::unique_ptr<RtpPacketToSend> RTPSender::BuildRtxPacket(
rtc::CritScope lock(&send_critsect_);
if (!sending_media_)
return nullptr;
// Replace payload type, if a specific type is set for RTX.
auto kv = rtx_payload_type_map_.find(packet.PayloadType());
// Use rtx mapping associated with media codec if we can't find one,
// assume it's red.
// TODO(holmer): Remove once old Chrome versions don't rely on this.
// Replace payload type.
auto kv = rtx_payload_type_map_.find(packet.PayloadType());
if (kv == rtx_payload_type_map_.end())
kv = rtx_payload_type_map_.find(payload_type_);
if (kv != rtx_payload_type_map_.end())
rtx_packet->SetPayloadType(kv->second);
return nullptr;
rtx_packet->SetPayloadType(kv->second);
// Replace sequence number.
rtx_packet->SetSequenceNumber(sequence_number_rtx_++);