Handling Messages

For first, you can "compress" that giant switch. For example you could instance a specific data decoder for each header. Switching on the value only and forwarding to the appropriate function. But you probably knew that already.

Register a set of decoders, each with a packet type and iterate on them. As to each decoder: is this your header? The first decoder matching gets the header+data.

This of course comes at quite some loss in perf compared to a raw switch but unless you have thousands of packet types I believe it's going to be a problem only in theory.

You can use a jump table, for example. You could define something similar to:

struct IPacketHandler
{
    virtual void ProcessMessage( Message& message ) = 0;
};
 
class MessageDispatcher
{
    IPacketHandler* m_packetHandlers[ MAXIMUM_HANDLERS ];
 
public:
    void RegisterHandler( IPacketHandler *packetHandler, int messageId )
    {
        assert( messageId >= 0 && messageId < MAXIMUM_HANDLERS );
        m_packetHandlers[ messageId ] = packetHandler;
    }
 
    void Dispatch( Message &message )
    {
        if ( NULL != m_packetHandlers[ message.id ] )
            m_packetHandlers[ message.id ]->ProcessMessage( message );
    }
}

This is not a complete solution, just to demonstrate what I mean. There would need to be more error checking/handling, initialization etc. and it could be expanded on a fair amount also. This is also C++ code, but the equivalent in almost any other language should be pretty similar.

n!

void Dispatch( Message &message )
{ 
      if ( NULL != m_packetHandlers[ message.id ] )
            m_packetHandlers[ message.id ]->ProcessMessage( message );
}

This is extraordinarily dangerous. The message is coming over UDP, it is completely untrusted, and you are not validating the message ID as being within the range of valid message handlers. As an attacker, this makes compromise of the machine almost trivial.

As I said, it's just a basic illustration and needs a lot of checking and error handling added, though no more than any other implementation. There is nothing inherently wrong with the approach suggested itself.

n!

The approach isn't wrong, but it is inherently more dangerous than a switch statement precisely because of the risk of an out of bounds index to a function pointer table.

If you get a switch statement wrong, it will do the wrong thing. If you get a function pointer wrong, it will do an arbitrary thing, and in this case, arbitrary is exactly what an attacker is looking for.

Yes, that's why you need to validate what you're using to index into the table. If you've already validated the input is fine, you can't do anything arbitrary no matter what they've sent you (because you've validated that you can't).

There is nothing inherently more dangerous in the given example over a switch statement, as long as your code has the correct validation there (which the original post already points out).

n!

I think switch statements are dumb, because they increase coupling significantly. Every time I add a new packet type, I have to go in and update every switch statement that uses packet type as input. That leads to a maintenance nightmare. In object oriented terms, switch statements break the open/closed principle. I'm not a huge fan of all things OO, but that principle is actually based on very sound learnings.

A linear array is fine, if array-out-of-bounds is tested first. You could use a class that automatically does this, to make safety a standard feature.

Another option is a hash table. Hash tables are great because the actual range of the command ID doesn't matter. You can use command IDs 1, 12, and 1337, without wasting any space in the heap. There is a small constant cost involved in following the hash-table links, though, which you avoid in the array case. If that small constant cost actually matters in profiling, you're probably building a router/firewall/switch, rather than a game server, though :-)

So here's what I ended up doing:

I used a Dictionary to look up my message types. I store the dictionary as static in the abstract Message class. I also have a static constructor in each child message class which adds it to the dictionary. The dictionary stores event handlers which call a static method in the child message to create an instance of the message. The instance of the message is then stored in the event args and returned through there, for the calling class to call Message.DoStuff().

I suppose I could simplify it by having the event call a static method to actually perform the actions as I am already passing in my header and the data through the event argument.

Once I get something more game like, I will also implement it as a switch statement and see which performs better.

By the way: Now that you have message dispatch working, the next thing you're going to want to do is to pack many messages into a single UDP datagram. Unless your game is very slow-paced and users only send a single message per second or less, that's likely to be needed.

C# reflection based example

using System;
using System.Collections.Generic;
using System.Linq;
using System.Reflection;

namespace Server
{
    //game message
    class Message
    {
        public byte[] Data { get; set; }

        public byte Opcode { get; set; }
    }


    interface IMessageHandler
    {
        void DoMessage(Message message);
    }

    internal class OpcodeAttribute : Attribute
    {
        //or enum
        public byte Opcode { get; set; }

        public OpcodeAttribute(byte opcode)
        {
            Opcode = opcode;
        }
    }

    [Opcode(5)]
    class WalkMessageHandler : IMessageHandler
    {
        public void DoMessage(Message message)
        {
            byte[] bytes = message.Data;
            //todo parse message 
        }
    }


    class MessageHandlerBucket
    {
        private readonly Dictionary<byte,IMessageHandler> _handlers = new Dictionary<byte, IMessageHandler>();

        /// <summary>
        /// 
        /// </summary>
        /// <param name="handlersAssembly">all hanlders in one assembly in this case</param>
        public MessageHandlerBucket(Assembly handlersAssembly)
        {
            IEnumerable<Type> types = handlersAssembly.GetTypes().Where(x => x.IsSubclassOf(typeof (IMessageHandler)));
            foreach (var type in types)
            {
                object opcode = type.GetCustomAttributes(typeof (OpcodeAttribute), true).FirstOrDefault();
                if (opcode == null)
                    continue;
                var opcodeAttribute = (OpcodeAttribute) opcode;
                byte code = opcodeAttribute.Opcode;
                var handler = (IMessageHandler) Activator.CreateInstance(type);
                _handlers.Add(code, handler);
            }
        }


        public void Handle(Message message)
        {
            byte opcode = message.Opcode;
            IMessageHandler handler = _handlers[opcode];
            if(handler== null) return;//or throw
            handler.DoMessage(message);
        }
    }
}