Header-only, cross-platform and compact json serialization for C++ with no memory allocations for parsing.
Just drop the header (glw_json.h) into your C++ project and include it into a source file
#include "glw_json.h"
Licensed under BSD to simplify the usage.
Install cmake
git clone https://github.com/yak32/glw_json.git
cd glw_json
git submodule update --init
mkdir build
cd build
cmake ..
cmake --build .
./../tests/tests
(use Git Bash on Windows)
#include "glw_json.h"
using namespace json;
struct vec3{
float x,y,z;
};
template<typename T>
bool serialize(T& t, vec3& v){
SERIALIZE(x);
SERIALIZE(y);
SERIALIZE(z);
return true;
}
int main(){
vec3 v;
if (load_object_from_file("test.json", v) != JSON_OK)
return 0;
if (save_object_to_file("test2.json", v) != JSON_OK)
return 0;
return 1;
}
To avoid hash maps with allocations, values should be serialized in alphabetical order. BTW, initial implementation with hash map was 2x times slower (for average sized json object with around 50 properties) To sort properties inside of serialize function - With Sublime Text 2/3: select everything inside serialize() and press F9 - it will sort properties automatically.
STL, c-runtime.
JSON loading - allocation-free parsing, N Log(N) performance where N is number of properties in a JSON object. Parsing speed is comparable to RapidJSON, saving is slower.
Saving to JSON - linear performance, std::ofstream is used to save data ( some allocations )
benchmark.cpp contains some logic to compare performance with RapidJson (currently parsing and loading only). Current results:
Run on 10 year old PC (8 X 3342 MHz CPU s) - glw_json is 25% slower than Rapid JSON for parsing:
Benchmark | Time | CPU | Iterations |
---|---|---|---|
BM_glw_json_load | 51367 ns | 51618 ns | 11200 |
BM_glw_json_save | 306253 ns | 299944 ns | 2240 |
BM_rapid_json_load | 40850 ns | 40981 ns | 17920 |
Run cmake on CMakeLists.txt in tests folder. Tests require C++11 support to compile (glw_json.h doesn't).
Tested with Visual Studio 2013, 2015, GCC 4.3.6 to GCC 7.0 (HEAD), Clang 3.0 to 3.9.0 (trunc), compartible with C++98, C++11 and C++14.
#include "glw_json.h"
#include <sstream>
using namespace json;
using namespace std;
struct array_int_var {
std::vector<int> val;
};
template <typename T> bool serialize(T& t, array_int_var& v) {
return SERIALIZE(val);
}
struct obj_array_var1 {
int val1;
float val2;
};
struct obj_array_var2 {
int val1;
float val2;
vector<obj_array_var1> val3;
};
struct object_pointer_in_object {
int val1;
float val2;
obj_array_var1* val3;
};
template <typename T> bool serialize(T& t, obj_array_var1& v) {
bool b = true;
b &= SERIALIZE(val1);
b &= SERIALIZE(val2);
return b;
}
template <typename T> bool serialize(T& t, obj_array_var2& v) {
bool b = true;
b &= SERIALIZE(val1);
b &= SERIALIZE(val2);
b &= SERIALIZE(val3);
return b;
}
template <typename T> bool serialize(T& t, object_pointer_in_object& v) {
bool b = true;
b &= SERIALIZE(val1);
b &= SERIALIZE(val2);
b &= SERIALIZE(val3);
return b;
}
int main(){
vec3 v;
// array
stringstream ss;
array_int_var av;
av.val = {1, 2, 3};
if (save_object_to_stream(av, ss) != JSON_OK )
printf("failed");
// array of objects
stringstream sss;
obj_array_var2 oav;
oav.val1 = 1;
oav.val2 = 1.1f;
oav.val3.resize(1);
oav.val3[0].val1 = 1;
oav.val3[0].val2 = 1.1f;
if (JSON_OK != save_object_to_stream(oav, sss))
printf("failed");
object_pointer_in_object obj_in_obj;
if (JSON_OK != load_object_from_file("object in object test", obj_in_obj))
printf("failed");
return 1;
}