Fix selectTrack order (#128)

Author: natario1

lib/src/main/java/com/otaliastudios/transcoder/internal/codec/Encoder.kt

@@ -100,6 +100,9 @@ internal class Encoder(
                     val flags = info.flags and BUFFER_FLAG_END_OF_STREAM.inv()
                     val buffer = buffers.getOutputBuffer(result)
                     val timeUs = info.presentationTimeUs
+                    buffer.clear()
+                    buffer.limit(info.offset + info.size)
+                    buffer.position(info.offset)
                     val data = WriterData(buffer, timeUs, flags) {
                         codec.releaseOutputBuffer(result, false)
                     }

lib/src/main/java/com/otaliastudios/transcoder/internal/transcode/DefaultTranscodeEngine.kt

@@ -102,9 +102,12 @@ internal class DefaultTranscodeEngine(
                 "videoUs=${timer.durationUs.videoOrNull()}"
         )
         while (true) {
-            val advanced =
-                    (segments.next(TrackType.AUDIO)?.advance() ?: false) or
-                    (segments.next(TrackType.VIDEO)?.advance() ?: false)
+            // Create both segments before reading. Creating the segment calls source.selectTrack,
+            // and if source is the same, it's important that both tracks are selected before
+            // reading (or even worse, seeking. DataSource.seek is broken if you add a track later on).
+            val audio = segments.next(TrackType.AUDIO)
+            val video = segments.next(TrackType.VIDEO)
+            val advanced = (audio?.advance() ?: false) or (video?.advance() ?: false)
             val completed = !advanced && !segments.hasNext() // avoid calling hasNext if we advanced.
 
             log.v("transcode(): executed step=$loop advanced=$advanced completed=$completed")

lib/src/main/java/com/otaliastudios/transcoder/internal/utils/eos.kt

@@ -13,17 +13,17 @@ internal fun DataSink.ignoringEos(ignore: () -> Boolean): DataSink
         = EosIgnoringDataSink(this, ignore)
 
 private class EosIgnoringDataSink(
-        private val source: DataSink,
+        private val sink: DataSink,
         private val ignore: () -> Boolean,
-) : DataSink by source {
+) : DataSink by sink {
     private val info = MediaCodec.BufferInfo()
     override fun writeTrack(type: TrackType, byteBuffer: ByteBuffer, bufferInfo: MediaCodec.BufferInfo) {
         if (ignore()) {
             val flags = bufferInfo.flags and MediaCodec.BUFFER_FLAG_END_OF_STREAM.inv()
             info.set(bufferInfo.offset, bufferInfo.size, bufferInfo.presentationTimeUs, flags)
-            source.writeTrack(type, byteBuffer, info)
+            sink.writeTrack(type, byteBuffer, info)
         } else {
-            source.writeTrack(type, byteBuffer, bufferInfo)
+            sink.writeTrack(type, byteBuffer, bufferInfo)
         }
     }
 }

lib/src/main/java/com/otaliastudios/transcoder/sink/DefaultDataSink.java

@@ -52,8 +52,9 @@ private QueuedSample(@NonNull TrackType type,
 
     private final static Logger LOG = new Logger("DefaultDataSink");
 
-    // I have no idea whether this value is appropriate or not...
-    private final static int BUFFER_SIZE = 64 * 1024;
+    // We must be able to handle potentially big buffers (e.g. first keyframe) in the queue.
+    // Got crashes with 152kb - let's use 256kb. TODO use a dynamic queue instead
+    private final static int BUFFER_SIZE = 256 * 1024;
 
     private boolean mMuxerStarted = false;
     private final MediaMuxer mMuxer;
@@ -117,10 +118,10 @@ public void setTrackFormat(@NonNull TrackType type, @NonNull MediaFormat format)
             mMuxerChecks.checkOutputFormat(type, format);
         }
         mLastFormat.set(type, format);
-        startIfNeeded();
+        maybeStart();
     }
 
-    private void startIfNeeded() {
+    private void maybeStart() {
         if (mMuxerStarted) return;
         boolean isTranscodingVideo = mStatus.get(TrackType.VIDEO).isTranscoding();
         boolean isTranscodingAudio = mStatus.get(TrackType.AUDIO).isTranscoding();
@@ -170,6 +171,12 @@ private void enqueue(@NonNull TrackType type,
         if (mQueueBuffer == null) {
             mQueueBuffer = ByteBuffer.allocateDirect(BUFFER_SIZE).order(ByteOrder.nativeOrder());
         }
+        LOG.v("enqueue(" + type + "): offset=" + bufferInfo.offset
+                + "\trealOffset=" + buffer.position()
+                + "\tsize=" + bufferInfo.size
+                + "\trealSize=" + buffer.remaining()
+                + "\tavailable=" + mQueueBuffer.remaining()
+                + "\ttotal=" + BUFFER_SIZE);
         buffer.limit(bufferInfo.offset + bufferInfo.size);
         buffer.position(bufferInfo.offset);
         mQueueBuffer.put(buffer);

lib/src/main/java/com/otaliastudios/transcoder/source/DefaultDataSource.java

@@ -148,9 +148,13 @@ public long seekTo(long desiredPositionUs) {
             // sync frames. MediaExtractor is not smart enough to sync the two tracks at the
             // video sync frame, so we must take care of this with the following trick.
             mExtractor.unselectTrack(mIndex.getAudio());
+            LOG.v("seekTo(): unselected AUDIO, seeking to " + (mOriginUs + desiredPositionUs) + " (extractorUs=" + mExtractor.getSampleTime() + ")");
             mExtractor.seekTo(mOriginUs + desiredPositionUs, MediaExtractor.SEEK_TO_PREVIOUS_SYNC);
+            LOG.v("seekTo(): unselected AUDIO and sought (extractorUs=" + mExtractor.getSampleTime() + ")");
             mExtractor.selectTrack(mIndex.getAudio()); // second seek might not be needed, but should not hurt.
+            LOG.v("seekTo(): reselected AUDIO, seeking to extractorUs (extractorUs=" + mExtractor.getSampleTime() + ")");
             mExtractor.seekTo(mExtractor.getSampleTime(), MediaExtractor.SEEK_TO_CLOSEST_SYNC);
+            LOG.v("seekTo(): seek workaround completed. (extractorUs=" + mExtractor.getSampleTime() + ")");
         } else {
             mExtractor.seekTo(mOriginUs + desiredPositionUs, MediaExtractor.SEEK_TO_PREVIOUS_SYNC);
         }