Save width/height of SPS nalus and restore them on the first packet of an IDR.

webrtc/modules/video_coding/h264_sps_pps_tracker.cc

Index: webrtc/modules/video_coding/h264_sps_pps_tracker.cc
diff --git a/webrtc/modules/video_coding/h264_sps_pps_tracker.cc b/webrtc/modules/video_coding/h264_sps_pps_tracker.cc
index 5dfdb49146a5f6874feef864074a8ffa72854166..72ffad6824262f0bb86e761e47f867daa0e1c0dc 100644
--- a/webrtc/modules/video_coding/h264_sps_pps_tracker.cc
+++ b/webrtc/modules/video_coding/h264_sps_pps_tracker.cc
@@ -45,6 +45,7 @@ H264SpsPpsTracker::PacketAction H264SpsPpsTracker::CopyAndFixBitstream(
   bool insert_packet = codec_header.nalus_length == 0 ? true : false;
 
   int pps_id = -1;
+  int sps_id = -1;
   size_t required_size = 0;
   for (size_t i = 0; i < codec_header.nalus_length; ++i) {
     const NaluInfo& nalu = codec_header.nalus[i];
@@ -55,6 +56,8 @@ H264SpsPpsTracker::PacketAction H264SpsPpsTracker::CopyAndFixBitstream(
         sps_data_[nalu.sps_id].data.reset(new uint8_t[nalu.size]);
         memcpy(sps_data_[nalu.sps_id].data.get(), data + nalu.offset,
                nalu.size);
+        sps_data_[nalu.sps_id].width = packet->width;
+        sps_data_[nalu.sps_id].height = packet->height;
         break;
       }
       case H264::NaluType::kPps: {
@@ -90,6 +93,7 @@ H264SpsPpsTracker::PacketAction H264SpsPpsTracker::CopyAndFixBitstream(
             return kRequestKeyframe;
           }
 
+          sps_id = pps->second.sps_id;
           pps_id = nalu.pps_id;
           required_size += pps->second.size + sizeof(start_code_h264);
           required_size += sps->second.size + sizeof(start_code_h264);
@@ -177,6 +181,11 @@ H264SpsPpsTracker::PacketAction H264SpsPpsTracker::CopyAndFixBitstream(
     memcpy(insert_at, data, data_size);
   }
 
+  if (sps_data_[sps_id].width > 0 && sps_data_[sps_id].height) {

[stefan-webrtc, 2017/03/14 10:32:20]

height > 0

Or, if you anyway initialize width/height to -1 you might as well always do this
assignment?

[philipel, 2017/03/14 11:51:12]

The problem is that I can't be certain that the SPS/PPS packet will have the
width/height set (the width, height is always in the SPS, but I save info from
the RTP header).

My experience with H264 so far is that no two implementations works the same,
and the way to get it to work is to make our implementation as forgiving as
possible, and this is the more general solution.

[stefan-webrtc, 2017/03/14 12:06:36]

I'm fairly sure width/height is required in the sps nalu, but if you want to be
on the safe side...

I think my main point was that sps_data_[sps_id].width is initialized to -1, and
if there was no width/height in the sps then I'd expect that we also want
packet->width to be -1? That would still be what happens if we remove the if
statement.

[philipel, 2017/03/14 13:52:41]

I looked at the code and we only know the width/height if we receive an SPS, so
yeah, this if check is redundant.

But it also made me realize that we need to save this info in InsertSpsPpsNalus.

+    packet->width = sps_data_[sps_id].width;
+    packet->height = sps_data_[sps_id].height;
+  }
+
   packet->dataPtr = buffer;
   packet->sizeBytes = required_size;
   return kInsert;