I have the following flatbuffer schema:
table Image {
...
}
table Pose {
r:[double] (required);
q:[double] (required);
}
table StampedImage {
pose: Pose (required);
image: Image (required);
}
I understand how to encode a flatbuffer object. For instance, to encode a Pose
object as a byte array:
double[] r = ...
double[] q = ...
FlatBufferBuilder fbb = new FlatBufferBuilder();
int rvec = Pose.createRVector(fbb, r);
int qvec = Pose.createQVector(fbb, q);
Pose.startPose(fbb);
Pose.addR(fbb, rvec);
Pose.addQ(fbb, qvec);
Pose.finishPoseBuffer(fbb, Pose.endPose(fbb));
byte[] encodedPose = fbb.sizedByteArray();
Furthermore, I understand how to "add" a pose to the parent class:
FlatBufferBuilder fbb = new FlatBufferBuilder();
int rOff = Pose.createRVector(fbb, r);
int qOff = Pose.createQVector(fbb, q);
Pose.startPose(fbb);
Pose.addR(fbb, rOff);
Pose.addQ(fbb, qOff);
int poseOff = Pose.endPose(fbb);
// Create the image
int imageOff = ...
StampedImage.startStampedImage(fbb);
StampedImage.addPose(fbb, poseOff);
StampedImage.addImage(fbb, imageOff);
fbb.finish(StampedImage.endStampedImage(fbb));
But what I would like to do is to add the encodedPose
data to the parent StampedImage
directly. My use case is that I am setting up a conglomerater... I am receiving pose data from one source as encoded byte arrays, and image data from another source (also as encoded byte arrays). So I want to be able to "combine" these objects into a single encoded object. For instance, the function signature would look like:
public byte[] encodeStampedImage( byte[] encodedPose, byte[] encodedImage ){
.... ?????
}
One approach would be to decode the objects, and then to reencode them, but that would have a huge overhead:
public byte[] encodeStampedImage( byte[] encodedPose, byte[] encodedImage ){
Pose pose = Pose.getRootAsPose( ByteBuffer.wrap(encodedPose ));
Image image = Image.getRootAsImage( ByteBuffer.wrap(encodedImage));
// Get r and q vectors from pose, then add them
FlatBufferBuilder fbb = new FlatBufferBuilder();
int rOff = Pose.createRVector(fbb, pose.rAsByteBuffer().array());
int qOff = Pose.createRVector(fbb, pose.qAsByteBuffer().array());
// Proceed as before...
}
I feel like there must be a better way to avoid these extra allocations. If the children are already encoded, isn't there some way to directly create a parent object from them?
There is currently no direct way to do this in the API. It is certainly possible, and in theory simple: you'd just add the bytes of the child-buffer to the parent, and then pass an offset to it to the parent constructor. That would require a new function in FlatBufferBuilder
though.