EBML fix for non-static GOP sizes

src/input/input_ebml.cpp

@@ -40,6 +40,7 @@ namespace Mist{
     lastClusterBPos = 0;
     lastClusterTime = 0;
     bufferedPacks = 0;
+    wantBlocks = true;
   }
 
   std::string ASStoSRT(const char * ptr, uint32_t len){
@@ -349,6 +350,15 @@ namespace Mist{
         bool isVideo = (Trk.type == "video");
         bool isAudio = (Trk.type == "audio");
         bool isASS = (Trk.codec == "subtitle" && Trk.init.size());
+        //If this is a new video keyframe, flush the corresponding trackPredictor
+        if (isVideo && B.isKeyframe()){
+          while (TP.hasPackets(true)){
+            packetData &C = TP.getPacketData(true);
+            myMeta.update(C.time, C.offset, C.track, C.dsize, C.bpos, C.key);
+            TP.remove();
+          }
+          TP.flush();
+        }
         for (uint64_t frameNo = 0; frameNo < B.getFrameCount(); ++frameNo){
           if (frameNo){
             if (Trk.codec == "AAC"){
@@ -463,28 +473,32 @@ namespace Mist{
       }
     }
     EBML::Block B;
-    do{
+    if (wantBlocks){
-      if (!readElement()){
+      do{
-        // Make sure we empty our buffer first
+        if (!readElement()){
-        if (bufferedPacks && packBuf.size()){
+          // Make sure we empty our buffer first
-          for (std::map<uint64_t, trackPredictor>::iterator it = packBuf.begin();
+          if (bufferedPacks && packBuf.size()){
-               it != packBuf.end(); ++it){
+            for (std::map<uint64_t, trackPredictor>::iterator it = packBuf.begin();
-            trackPredictor &TP = it->second;
+                 it != packBuf.end(); ++it){
-            if (TP.hasPackets(true)){
+              trackPredictor &TP = it->second;
-              packetData &C = TP.getPacketData(myMeta.tracks[it->first].type == "video");
+              if (TP.hasPackets(true)){
-              fillPacket(C);
+                packetData &C = TP.getPacketData(myMeta.tracks[it->first].type == "video");
-              TP.remove();
+                fillPacket(C);
-              --bufferedPacks;
+                TP.remove();
-              return;
+                --bufferedPacks;
+                return;
+              }
             }
           }
+          // No more buffer? Set to empty
+          thisPacket.null();
+          return;
         }
-        // No more buffer? Set to empty
+        B = EBML::Block(ptr);
-        thisPacket.null();
+      }while (!B || B.getType() != EBML::ELEM_BLOCK || !selectedTracks.count(B.getTrackNum()));
-        return;
+    }else{
-      }
       B = EBML::Block(ptr);
-    }while (!B || B.getType() != EBML::ELEM_BLOCK || !selectedTracks.count(B.getTrackNum()));
+    }
 
     uint64_t tNum = B.getTrackNum();
     uint64_t newTime = lastClusterTime + B.getTimecode();
@@ -493,6 +507,24 @@ namespace Mist{
     bool isVideo = (Trk.type == "video");
     bool isAudio = (Trk.type == "audio");
     bool isASS = (Trk.codec == "subtitle" && Trk.init.size());
+
+    //If this is a new video keyframe, flush the corresponding trackPredictor
+    if (isVideo && B.isKeyframe() && bufferedPacks){
+      if (TP.hasPackets(true)){
+        wantBlocks = false;
+        packetData &C = TP.getPacketData(true);
+        fillPacket(C);
+        TP.remove();
+        --bufferedPacks;
+        return;
+      }
+    }
+    if (isVideo && B.isKeyframe()){
+      TP.flush();
+    }
+    wantBlocks = true;
+
+
     for (uint64_t frameNo = 0; frameNo < B.getFrameCount(); ++frameNo){
       if (frameNo){
         if (Trk.codec == "AAC"){
@@ -531,6 +563,7 @@ namespace Mist{
   }
 
   void InputEBML::seek(int seekTime){
+    wantBlocks = true;
     packBuf.clear();
     bufferedPacks = 0;
     uint64_t mainTrack = getMainSelectedTrack();
@@ -541,17 +574,6 @@ namespace Mist{
     uint64_t partCount = 0;
     for (unsigned int i = 0; i < Trk.keys.size(); i++){
       if (Trk.keys[i].getTime() > seekTime){
-        if (i > 1){
-          partCount -= Trk.keys[i-1].getParts() + Trk.keys[i-2].getParts();
-          uint64_t partEnd = partCount + Trk.keys[i-2].getParts();
-          uint64_t partTime = Trk.keys[i-2].getTime();
-          for (uint64_t prt = partCount; prt < partEnd; ++prt){
-            INSANE_MSG("Replay part %llu, timestamp: %llu+%llu", prt, partTime, Trk.parts[prt].getOffset());
-            packBuf[mainTrack].add(partTime, Trk.parts[prt].getOffset(), mainTrack, 0, 0, false, isVideo, (void *)0);
-            packBuf[mainTrack].remove();
-            partTime += Trk.parts[prt].getDuration();
-          }
-        }
         break;
       }
       partCount += Trk.keys[i].getParts();

src/input/input_ebml.h

@@ -40,15 +40,13 @@ namespace Mist{
   class trackPredictor{
     public:
       packetData pkts[PKT_COUNT];
+      uint64_t frameOffset;
       uint16_t smallestFrame;
       uint64_t lastTime;
       uint64_t ctr;
       uint64_t rem;
       trackPredictor(){
-        smallestFrame = 0;
+        flush();
-        lastTime = 0;
-        ctr = 0;
-        rem = 0;
       }
       bool hasPackets(bool finished = false){
         if (finished){
@@ -57,6 +55,14 @@ namespace Mist{
           return (ctr - rem > 12);
         }
       }
+      /// Clears all internal values, for reuse as-new.
+      void flush(){
+        frameOffset = 0;
+        smallestFrame = 0;
+        lastTime = 0;
+        ctr = 0;
+        rem = 0;
+      }
       packetData & getPacketData(bool mustCalcOffsets){
         frameOffsetKnown = true;
         //grab the next packet to output
@@ -74,13 +80,20 @@ namespace Mist{
             //we assume it's just time stamp drift due to lack of precision.
             p.time = (lastTime + smallestFrame);
           }else{
-            p.offset = maxEBMLFrameOffset;
+            p.time -= frameOffset?frameOffset + smallestFrame:0;
+            p.offset = maxEBMLFrameOffset + frameOffset + smallestFrame;
           }
         }
         lastTime = p.time;
+        DONTEVEN_MSG("Outputting%s %llu + %llu, offset %llu (%llu -> %llu), display at %llu", (p.key?" (KEY)":""), p.time, frameOffset, p.offset, rem, rem % PKT_COUNT, p.time+p.offset);
         return p;
       }
       void add(uint64_t packTime, uint64_t packOffset, uint64_t packTrack, uint64_t packDataSize, uint64_t packBytePos, bool isKeyframe, bool isVideo, void * dataPtr = 0){
+        //If no packets have been removed yet and there is more than one packet, calculate frameOffset
+        if (!rem && ctr && packTime < pkts[0].time){
+          frameOffset = pkts[0].time - packTime;
+          INSANE_MSG("Setting frameOffset to %llu", frameOffset);
+        }
         if (isVideo && ctr && ctr >= rem){
           int32_t currOffset = packTime - pkts[(ctr-1)%PKT_COUNT].time;
           if (currOffset < 0){currOffset *= -1;}
@@ -131,6 +144,7 @@ namespace Mist{
     bool openStreamSource(){return true;}
     bool needHeader(){return needsLock() && !readExistingHeader();}
     double timeScale;
+    bool wantBlocks;
   };
 }