nark-serialization

Now the newest version is in topling-ark

An easy, fast, small, portable C++ object serialization framework

Features

1. Declarative syntax, Easy to use, Easy to integrate
2. Fast at runtime, Fast at compile time
3. Serialized data is small, Compiled code is small
4. Support STL containers & types
5. Can serialize to BigEndian/LittleEndian in any platform
6. Support variant int/uint encoding
7. Optional serialization version control
8. More and more...

Bench Mark

I have written a benchmark vs boost-serialization, but have not written benchmarks with other serilization libraries. Fortunately, there is a benchmark of many serialization library, including boost-serialization.

You can run benchmark of nark-serialization vs boost-serialization yourself.

Quick Start

nark-serialization is powerful, but using it is very simple:

#include <nark/io/DataIO.hpp>
#include <nark/io/MemStream.hpp>
#include <nark/io/FileStream.hpp>

struct MyData1 {
    int  a, b, c;
    var_int32_t d;  // d.t is int32 value
    var_uint64_t e; // e.t is uint64 value
    std::string f;
    std::map<std::string, int> g;
    std::set<int> h;
 
	// The simplest use case(without version control):
    DATA_IO_LOAD_SAVE(MyData1, &a&b&c&d&e&f&g&h)
};
 
struct MyData2 {
    int  a, b, c;
    var_int32_t d;
    var_uint64_t e;
    std::string f;
    std::map<std::string, int> g;
    std::set<int> h;
 
    DATA_IO_LOAD_SAVE_V(MyData2, // With version control
        1, // 1 is current version
        &a&b&c&d&e&f&g&h)
};
 
struct MyData3 {
    int  a, b, c;
    boost::int32_t d;
    boost::uint64_t e;
    std::string f;
    std::map<std::string, int> g;
    std::set<int> h;
    std::multiset<int> i;
    unsigned version;
 
    DATA_IO_LOAD_SAVE_V(MyData3, // With version control
        2, // current version
        &a &b &c
        &as_var_int(d) // 'd' defined as  int32_t, serialized as var_int32_t
        &as_var_int(e) // 'e' defined as uint64_t, serialized as var_uint64_t
        &f &g &h
        &vmg.since(2, i) // 'i' is added in version=2
        &vmg.get_version(version) // optional, you can keep version number for latter use
        )
};
 
// in system or third-party's header, can not be changed
struct SysData1 {
    int a;
    unsigned b;
    string c;
};
DATA_IO_LOAD_SAVE_E(SysData1, &a &b &c)

struct SysData2 {
    int a;
    unsigned b;
    string c;
};
DATA_IO_LOAD_SAVE_EV(SysData2, &a &b &vmg.since(2, c))

int main(int argc, char* argv[]) {
    PortableDataOutput<AutoGrownMemIO> output; // 'Portable' is BigEndian
    PortableDataInput<MemIO> input; // 'Portable' is BigEndian

    MyData1 d1;
    // set d1 values
    // ...
    MyData2 d2;
    // set d2 values
    // ...
    MyData3 d3;
    // set d3 values
    // ...
	SysData1 sd1;
	SysData2 sd2;
    output << d1 << d2 << d3 << sd1 << sd2; // saving

    input = output.head(); // shallow copy, input is the written part of output
    input >> d1 >> d2 >> d3 >> sd1 >> sd2; // loading

	output.rewind(); // rewind the write pointer
    output & d1 & d2 & d3 & sd1 & sd2; // saving, '&' is an alias of '<<'

    input = output.head(); // shallow copy, input is the written part of output 
    input & d1 & d2 & d3 & sd1 & sd2; // loading, '&' is an alias of '>>'
}

Is it a magic? I wrote the first version of nark-serialization at July 2006, three months later, I wrote nark-rpc on top of nark-serialization.

Compile

$ cd path/to/nark-base
$ make # compile nark-base
$ cd test-serialization
$ make # compile test and benchmark
$ rls/bench.exe # run benchmark

More

To be done