从char缓冲区读取C / C ++以填充结构

Question

I have a message which is sent over a UDP socket. The message is of type RPCMessage

typedef struct
{
    enum {Request, Reply} messageType;
    unsigned int RPCId;
    unsigned int procedure;
    int arg1;
    int arg2;
} RPCMessage;

when I send am preparing to send a message, I construct the message as follows:

RPCMessage toSend = {RPCMessage::Reply, htonl(rpcId), htonl(procedureId), htonl(int1), htonl(int2);

When I receive the message, it is received into a char[] I call buffer. It has been suggested that rather than just casting the entire buffer to an RPCMessage, it would be better to have something where I read each argument from the buffer one by one and then could have the first argument as the correct enum type and use ntohl on each other argument. Would the best way to do this just be something like:

RPCMessage::messageType type;
unsigned int id, procedure;
int int1, int2;
sscanf(buffer, %d%d%d%d%d, type, id, procedure, int1, int2);
RPCMessage received = {type, ntohl(id), ntohl(procedure), ntohl(int1), ntohl(int2));

or is there another/better way?

Answer 1

The best way would be to use a robust serialization framework like boost serialization.

http://www.boost.org/doc/libs/1_51_0/libs/serialization/doc/index.html

It depends how complex your messages can get.

Answer 2

I would recommend casting to structs. Primary reason - performance.

You can almost always achieve binary compatibility between the layout of your struct and the data in your buffer. There are pragmas that control padding and alignment, etc. In C++ there are special alignment specification for that.

I do not want to say just cast and problems will disappear . You still need to handle byte order, character repr can be different, etc.

Serialization should be used when you want to have different repesentaion of your structs or when your data structure is complex, like a tree of structures.