Tiny, quick JSON encoding/decoding in pure Lua.

Features

• Pure lua, should work on every version (5.1-5.4, JIT)
• Quick, focused on performance. (See benchmarks below)
• Very small, ~180 sloc.
• Decent error handling: Expected : to follow key for object at char 39

Usage

local json = require "qjson"
print(json.encode {
	hello = "world!",
	qjson = { "fast", "simple", "tiny" }
})

--[[
	{"qjson":["fast","simple","tiny"],"hello":"world!"}
]]

print(json.decode([[
	{ "foo": "bar" }
]]))

Notes

• null is output as a special table instance, retrieved from qjson.NULL
• This does not guarantee 100% compatibility with the more niche parts of the JSON spec (like unicode escapes)

Benchmarks

Using benchmarks/bench.lua (which tests the simdjson twitter example) through WSL:

LuaJIT 2.1.0-beta3 (Windows 11)
11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz

Name (Decode)	Min	Max	Avg	Avg / Best
rxi/json	0.005	0.009	0.00672	x1
actboy168/json	0.011	0.016	0.01295	x1.92708
luadist/dkjson	0.012	0.017	0.01426	x2.12202
vurv78/qjson	0.008	0.015	0.00939	x1.39732
grafi-tt/lunajson	0.005	0.012	0.007095	x1.0558

Name (Encode)	Min	Max	Avg	Avg / Best
rxi/json	0.01	0.02	0.011745	x7.93581
actboy168/json	0.009	0.014	0.01077	x7.27703
luadist/dkjson	0.015	0.018	0.01559	x10.5338
vurv78/qjson	0.001	0.003	0.00148	x1
grafi-tt/lunajson	0.007	0.01	0.008645	x5.84122

Lua 5.3 (WSL : Windows 11)
11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz

Name (Decode)	Min	Max	Avg	Avg / Best
actboy168/json	0.019036	0.024134	0.0204356	x1.63325
rxi/json	0.045677	0.059829	0.0487656	x3.89742
vurv78/qjson	0.01478	0.02029	0.0160733	x1.28461
luadist/dkjson	0.028986	0.038415	0.0314367	x2.51247
grafi-tt/lunajson	0.01162	0.015419	0.0125123	x1

Name (Encode)	Min	Max	Avg	Avg / Best
actboy168/json	0.016462	0.019902	0.0176806	x3.0031
rxi/json	0.016091	0.020716	0.0182224	x3.09512
vurv78/qjson	0.005352	0.008078	0.00588746	x1
luadist/dkjson	0.022643	0.04247	0.0249368	x4.23559
grafi-tt/lunajson	0.011779	0.014569	0.0128647	x2.1851

From here, you can see this library is fastest at encoding, running 8-10x faster than the most commonly used libraries.

Decoding is getting there. Currently balancing performance between PUC-Lua and Luajit.