Do not use B frames but B frame like references for AVC as discussed on the matroska-devel mailing list. A new hack "avc_use_bframes" can mark those frames as B frames.

src/common/hacks.cpp

@@ -36,6 +36,7 @@ static const char *mosu_hacks[] = {
   ENGAGE_NO_VARIABLE_DATA,
   ENGAGE_NO_DEFAULT_HEADER_VALUES,
   ENGAGE_FORCE_PASSTHROUGH_PACKETIZER,
+  ENGAGE_AVC_USE_BFRAMES,
   NULL
 };
 static vector<string> engaged_hacks;

src/common/hacks.h

@@ -33,6 +33,7 @@ using namespace std;
 #define ENGAGE_NO_VARIABLE_DATA "no_variable_data"
 #define ENGAGE_NO_DEFAULT_HEADER_VALUES "no_default_header_values"
 #define ENGAGE_FORCE_PASSTHROUGH_PACKETIZER "force_passthrough_packetizer"
+#define ENGAGE_AVC_USE_BFRAMES "avc_use_bframes"
 
 void MTX_DLL_API engage_hacks(const string &hacks);
 bool MTX_DLL_API hack_engaged(const string &hack);

src/input/r_qtmp4.cpp

@@ -29,6 +29,7 @@
 #include "aac_common.h"
 #include "avilib.h"
 #include "common.h"
+#include "hacks.h"
 #include "matroska.h"
 #include "p_aac.h"
 #include "p_mp3.h"
@@ -1037,7 +1038,8 @@ qtmp4_reader_c::handle_video_with_bframes(qtmp4_demuxer_ptr &dmx,
         mxerror(FMT_FN "The video track does not start with a key "
                 "frame and a P frame but contains B frames. This is not "
                 "supported.\n", ti->fname.c_str());
-      fref = dmx->references[1];
+      if (dmx->avc_use_bframes)
+        fref = dmx->references[1];
 
     } else {
       // This is a P frame. At the moment it references the second stored frame
@@ -1452,6 +1454,10 @@ qtmp4_reader_c::create_packetizer(int64_t tid) {
                                                 false, ti));
         ti->private_data = NULL;
 
+        if (hack_engaged(ENGAGE_AVC_USE_BFRAMES))
+          dmx->avc_use_bframes = true;
+        else
+          PTZR(dmx->ptzr)->relaxed_timecode_checking = true;
 
       } else {
         ti->private_size = dmx->v_stsd_size;

src/input/r_qtmp4.h

@@ -118,6 +118,7 @@ struct qtmp4_demuxer_t {
   uint32_t v_width, v_height, v_bitdepth;
   int64_t v_dts_offset;
   deque<int64_t> references;
+  bool avc_use_bframes;
   int64_t max_timecode;
   uint32_t a_channels, a_bitdepth;
   float a_samplerate;
@@ -137,6 +138,7 @@ struct qtmp4_demuxer_t {
     esds_parsed(false),
     v_stsd(NULL), v_stsd_size(0),
     v_width(0), v_height(0), v_bitdepth(0), v_dts_offset(0),
+    avc_use_bframes(false),
     max_timecode(0),
     a_channels(0), a_bitdepth(0), a_samplerate(0.0),
     priv(NULL), priv_size(0),

src/merge/pr_generic.cpp

@@ -55,6 +55,7 @@ generic_packetizer_c::generic_packetizer_c(generic_reader_c *nreader,
   enqueued_bytes = 0;
   safety_last_timecode = 0;
   safety_last_duration = 0;
+  relaxed_timecode_checking = false;
   last_cue_timecode = -1;
   correction_timecode_offset = 0;
   append_timecode_offset = 0;
@@ -943,7 +944,8 @@ generic_packetizer_c::add_packet2(packet_t *pack) {
   // 'timecode < safety_last_timecode' may only occur for B frames. In this
   // case we have the coding order, e.g. IPB1B2 and the timecodes
   // I: 0, P: 120, B1: 40, B2: 80.
-  if ((pack->timecode < safety_last_timecode) && (pack->fref < 0)) {
+  if (!relaxed_timecode_checking &&
+      (pack->timecode < safety_last_timecode) && (pack->fref < 0)) {
     if (htrack_type == track_audio) {
       int64_t needed_timecode_offset;
 

src/merge/pr_generic.h

@@ -446,6 +446,7 @@ public:
   int connected_to;
   int64_t correction_timecode_offset;
   int64_t append_timecode_offset, max_timecode_seen;
+  bool relaxed_timecode_checking;
 
 public:
   generic_packetizer_c(generic_reader_c *nreader, track_info_c *nti)