Protocol buffers issue, multiple serializations into a binary file

Question

I am getting some weird behaviour from protobuf binary file io. I am pre-processing a text corpus into a protobuf intermediary file. My serialization class looks as follows:

  class pb_session_printer
  {
  public:
    pb_session_printer(std::string & filename)
      : out(filename.c_str(), std::fstream::out | std::fstream::trunc | 
                              std::fstream|binary)
      {}

    void print_batch(std::vector<session> & pb_sv)
    {
      boost::lock_guard<boost::mutex> lock(m);

      BOOST_FOREACH(session & s, pb_sv)
      {
        std::cout << out.tellg() << ":";
        s.SerializeToOstream(&out);
        out.flush();
        std::cout << s.session_id() << ":" << s.action_size() << std::endl;
      }
      exit(0);
    }

    std::fstream out;
    boost::mutex m;
  };

A snippet of output looks like :

0:0:8
132:1:8
227:2:6
303:3:6
381:4:19
849:5:9
1028:6:2
1048:7:18
1333:8:28
2473:9:24

The first field shows that serialization is proceeding as normal.

When I run my loading program :

int main()
{
  std::fstream in_file("out_file", std::fstream::in | std::ios::binary);
  session s;

  std::cout << in_file.tellg() << std::endl;
  s.ParseFromIstream(&in_file);
  std::cout << in_file.tellg() << std::endl;
  std::cout << s.session_id() << std::endl;

  s.ParseFromIstream(&in_file);
}

I get:

0
-1
111
libprotobuf ERROR google/protobuf/message_lite.cc:123] Can't parse message of type 
"session" because it is missing required fields: session_id

session_id : 111 is an entry towards the end of the stream, I clearly don't understand the semantics of binary-io facilities of the library. Please help.

Answer 1

If you write multiple protobuffers in a single file you will need to write the size of the protobuf + protobuffer and read them in seperately (so without ParseFromIstream as Cat Plus Plus mentioned). When you have read in the protobuffer you can parse it with ParseFromArray .

Your file would look size this (the spaces are just for readability):

size protobuf size protobuf size protobuf etc.

Answer 2

Message::ParseFromIstream is documented to consume the entire input. Since you're serialising a sequence of messages of the same type, you can just create a new message with repeated field of that type, and work with that.