保存文件

Node::SerializeState —— MapBuilder::SerializeState —— MapBuilder::SerializeState —— cartographer\io\internal\mapping_state_serialization.cc文件中的WritePbStream函数

实际是MapBuilderBridge::SerializeState，Node::HandleWriteState调用的也是这个函数。在最新版本里是bool MapBuilderBridge::SerializeState(const std::string& filename, const bool include_unfinished_submaps);

cartographer::io::ProtoStreamWriter writer(filename);
map_builder_->SerializeState(&writer);
return writer.Close();

然后是

void MapBuilder::SerializeState(bool include_unfinished_submaps,
                                io::ProtoStreamWriterInterface* const writer)
{
  io::WritePbStream(*pose_graph_, all_trajectory_builder_options_, writer,
                    include_unfinished_submaps);
}

proto_stream.cc包含ProtoStreamWriter和ProtoStreamReader两个类，这里用的是类ProtoStreamWriter，在它的构造函数里以二进制方式，写入8个字节的魔术数以确定proto stream格式，

WritePbStream如下，这里的writer就是类ProtoStreamWriter

void WritePbStream(
    const mapping::PoseGraph& pose_graph,
    const std::vector<mapping::proto::TrajectoryBuilderOptionsWithSensorIds>&
        trajectory_builder_options,
    ProtoStreamWriterInterface* const writer, bool include_unfinished_submaps)
{
  writer->WriteProto(CreateHeader());
  writer->WriteProto(
      SerializePoseGraph(pose_graph, include_unfinished_submaps));
  writer->WriteProto(SerializeTrajectoryBuilderOptions(
      trajectory_builder_options,
      GetValidTrajectoryIds(pose_graph.GetTrajectoryStates())));

  SerializeSubmaps(pose_graph.GetAllSubmapData(), include_unfinished_submaps,
                   writer);
  SerializeTrajectoryNodes(pose_graph.GetTrajectoryNodes(), writer);
  SerializeTrajectoryData(pose_graph.GetTrajectoryData(), writer);
  SerializeImuData(pose_graph.GetImuData(), writer);
  SerializeOdometryData(pose_graph.GetOdometryData(), writer);
  SerializeFixedFramePoseData(pose_graph.GetFixedFramePoseData(), writer);
  SerializeLandmarkNodes(pose_graph.GetLandmarkNodes(), writer);
}

保存的是子图、轨迹节点、轨迹数据、IMU数据、里程计、Landmark等等

ProtoStreamWriter保存数据时用到了common::FastGzipString函数，实质是boost::iostreams::gzip_compressor

读pbstream文件

这个就参考纯定位流程，Node::LoadState —— MapBuilderBridge::LoadState —— MapBuilder::LoadState

MapBuilderBridge::LoadState(const std::string& state_filename, bool load_frozen_state)使用ProtoStreamReader

cartographer::io::ProtoStreamReader stream(state_filename);
map_builder_->LoadState(&stream, load_frozen_state);

到了MapBuilder::LoadState，可以看出如果只是读pbstream文件，不需要实例化map_builder，除去加载数据到后端，把函数里的大部分内容提取出来即可，主要是类型io::ProtoStreamDeserializer和proto::PoseGraph等等。

另外可以参考pbstream_main.cc