ZenJSON

zenjson is a JSON parser/generator for C++.

zenjson is very high-performance. It is much faster than RapidJSON both in parsing/serializing, and only lose to simdjson a little in parsing normal json files.

zenjson also has easy-to-use, intuitive syntax, benifits from modern C++.

zenjson is highly inspired from Richard Geldreich's pjson, vivkin's gason and RapidJSON.

Features

No dependencies.
Single header file. Simply drop zenjson.h into your project.
Small code size (<2k loc) and memory-friendly (16-24bytes per value).
In-stu parsing. That means it modifies and depends the input string.

Usage

Initialize

// create a json with value "null"
zjson::Json json1 = nullptr;
// create a json string
zjson::Json json2 = "hello world";
// create a empty json object. empty object is also treated as empty array in zenjson
zjson::Json json3 = {};
// create a complex json, using an initializer list
zjson::Json json = {
	{ "pi", 3.141 },
	{ "nothing", nullptr },
	{ "variable", json2 },
	{ "list", { 1, 0, 2 } },
	{ "object", {
		{"currency", "USD"},
		{"value", 42.99}
	    }
	}
};

Modify

// create a empty json object (or array)
zjson::Json json = {};
// you can modify or add members using operator []
json["pi"] = 3.141;
json["nothing"] = nullptr;
json["data"] = {};
// you can modify or push back elements too
json["data"][0] = 22;
json["data"][1] = "33";
// add an object inside the object
json["answer"]["everything"] = 42;
// modify or add a complex json
json["pi"] = { { "list", { 1, 0, 2 } } };

Obtain

double pi = json["pi"];
std::string hello = json["variable"];
bool enabled = json["enabled"];
int64_t twentytwo = json["data"][0];
// or uses getInt/Double/String/... interface, they support passing default values
// if node "backend" does not exist, got "opengl"
std::string backend = json["config"]["backend"].getString("opengl");

Serialization / Deserialization

zjson::Document doc;
// parsing from buffer
int error = doc.parse(jsonstr);
// dump to std::string
std::string jsonstr = doc.dump();
// dump to a buffer, outSize return the buffer size actully needed
bool ok = doc.dump(buffer, 1024 * 1024 * 4, &outSize, true);

Performance

All tested JSON files are same with nativejson-benchmark.

The following results run on my Intel Core i7-7700K, compiled by Visual Studio 2017, 64-bit build.

Parsing

Library	twitter.json /time	twitter.json /throughput	canada.json /time	canada.json /throughput	citm_catalog .json/time	citm_catalog .json/throughput
zenjson	0.296 ms	2034.7 MB/s	3.048 ms	704.3 MB/s	0.715 ms	2303.8 MB/s
RapidJSON	1.056 ms	570.3 MB/s	5.509 ms	389.7 MB/s	1.869 ms	881.3 MB/s
simdjson	0.292 ms	2062.5 MB/s	2.093 ms	1025.7 MB/s	0.692 ms	2380.3 MB/s
nlohmann/json	7.910 ms	76.1 MB/s	41.988 ms	51.1 MB/s	14.094 ms	116.9 MB/s

Notes:

RapidJSON uses RAPIDJSON_SSE2/ParseInsitu.
simdjson uses allocate_capacity for pre-allocation. It is compiled with /std:c++17, others are default.

Stringifying

Library	twitter.json	canada.json	citm_catalog.json
zenjson	0.712 ms	8.723 ms	0.777 ms
RapidJSON	1.877 ms	24.125 ms	3.299 ms
nlohmann/json	5.584 ms	83.046 ms	11.850 ms

luochen7452/zenjson