开发者

In protobuf-net is it possible to partially deserialize a message based on a base type

开发者 https://www.devze.com 2023-02-10 07:48 出处:网络
in protobuf-net is it possible to partially deserialize a message based on a base type? In my system I have an inheritance hierarchy where every message inherits from a MessageBase. The MessageBase

in protobuf-net is it possible to partially deserialize a message based on a base type?

In my system I have an inheritance hierarchy where every message inherits from a MessageBase. The MessageBase has a uint MessageType. Ideally I only want to deserialize the MessageBase and check if it’s a MessageType I’m interested in then I can either throw away the message or make the decision to deserialize the actual message. This is to save the cost of deserialize ( I have a cpu cycle budget and lots of messages to process).

Example usage shown below.

Thanks a lot.

 MessageBase msgBase = ..deserialize; 

  if(msgBase.MessageType = 1)//1 is the Tick msg type
  {
     Tick tick = ..deserialize actual msg;
     //do something with tick
   }
  //throw away msgBase 

[ProtoContract,Prot开发者_高级运维oInclude(1, typeof(Tick))]
public class MessageBase
{
    protected uint _messageType;

    [ProtoMember(1)]
    public uint MessageType
    {
        get { return _messageType; }
        set{ _messageType = value;}
    }
}

[ProtoContract]
public public class Tick : MessageBase
{
    private int _tickId;
    private double _value;

    public Tick()
    {
        _messageType = 1;
    }

    [ProtoMember(1)]
    public int TickID
    {
        get { return _tickId; }
        set { _tickId = value; }
    }
    [ProtoMember(2)]
    public double Value
    {
        get { return _value; }
        set { _value = value; }
    }
}


If it is part of the message, then at the moment: no. Since it is field 1, I could potentially pre-screen them, but even that is a hack (there is no guarantee that field 1 is first - the spec makes it clear that you must allow for any ordering).

However!

there may be an option, if you are open to a little refactoring. If this is a linear heterogenous sequence of messages, then another way to encode it is to use the SerializeWithLengthPrefix implementation, passing a different tag for each message type - then you have a sequence like (being a bit liberal with the representation)

1:[tick-body] 2:[some-other-body] 1:[tick body] etc

of course it depends a bit on the other end matching, but unless I am mistaken this ties in pretty well with the SAX-like processing discussed here (as a proposal), which incidentally is also entirely compatible with how the NonGeneric deserialize works. Here's an example, that only deserializes the Bar objects, showing "2" and "4" on the console:

using System;
using System.IO;
using ProtoBuf;
[ProtoContract]
class Foo
{
    [ProtoMember(1)]
    public int A { get; set; }
}
[ProtoContract]
class Bar
{
    [ProtoMember(1)]
    public int B { get; set; }
}
static class Program
{
    static void Main()
    {
        using (var ms = new MemoryStream())
        {
            Serializer.SerializeWithLengthPrefix(ms, new Foo { A = 1 }, PrefixStyle.Base128, 1);
            Serializer.SerializeWithLengthPrefix(ms, new Bar { B = 2 }, PrefixStyle.Base128, 2);
            Serializer.SerializeWithLengthPrefix(ms, new Foo { A = 3 }, PrefixStyle.Base128, 1);
            Serializer.SerializeWithLengthPrefix(ms, new Bar { B = 4 }, PrefixStyle.Base128, 2);
            ms.Position = 0;

            // we want all the Bar - so we'll use a callback that says "Bar" for 2, else null (skip)
            object obj;
            while (Serializer.NonGeneric.TryDeserializeWithLengthPrefix(ms, PrefixStyle.Base128,
                tag => tag == 2 ? typeof(Bar) : null, out obj))
            {
                Console.WriteLine(((Bar)obj).B);
            }
        }
    }
}

On the wire, this is actually compatible with a parent object:

repeated foo foo = 1;
repeated bar bar = 2;

If the proposed option generate_visitors gets implemented you should be able to consume the same type of heterogenous data stream from any client. The obvious mapping would be something like an optional property on [ProtoContract] to help with this - but I don't want to add that until the new protobuf feature is clear, as so far it looks like an exact match to my implementation. Which is nice.

0

精彩评论

暂无评论...
验证码 换一张
取 消