A high-performance, memory-free JSON parser for MQL5, based on a flat tape architecture.
Single-pass iterative state machine (as is, without extra loops, "pure" o(n)): no recursion, no dynamic memory fragmentation, no overhead from function calls.

---

Main Features

• Tape-based zero-alloc model: the entire JSON is parsed into a single contiguous long[] array
• Single flat loop: one switch over a token enum, no helper function calls during parsing, strictly O(n)
• JIT perfect-hash key lookup: object access starts as linear probing (StrEquals); once a node passes TSN_JSON_JIT_MIN_REF_TO_HASING accesses, a minimal perfect hash table is built for it on the fly (k = n, no slack), promoting that node's lookups to O(1). No FNV hashing is computed upfront during parsing.
• Handle-based navigation: CJsonNode is a lightweight struct (pointer + two ints), copying is free
• In-place mutation: SetInt, SetDbl, SetBool modify the tape directly without re-parsing
• Iterator support: CJsonIteratorObj and CJsonIteratorArray for clean traversal
• Three input formats, one API: AssingFile auto-detects by extension — .json (plain text, needs Parse()), .jsonasm (assembly-like text dump of the tape, needs ParseAssembly()), .jsontc (precompiled binary tape, ready to use immediately, no parsing step)
• Compiled tape cache (.jsontc): SaveCompiled persists the tape, the source JSON (or a reference to its file), and every perfect-hash table already built — so a reload via AssingFile skips parsing entirely and keeps any JIT-hashed nodes already O(1)

Parse and navigate

#include "Src\\JsonNode.mqh"

TSN::CJsonParser parser;
parser.Assing("{\"symbol\":\"EURUSD\",\"bid\":1.2345,\"ask\":1.2347,\"active\":true}");
parser.CorrectPadingForSwar();
parser.Parse();

TSN::CJsonNode root = parser.GetRoot();
string symbol = root["symbol"].ToString();
double bid = root["bid"].ToDouble(0.0);
bool active = root["active"].ToBool(false);

Build json as json string or valid json with CJsonBuilder

  TSN::CJsonBuilderStr build;
  build.PutChar('"');
  build.Obj();
  build.Key("Valor").ValS("└");
  build.Key("asas").ValS("\r\n\v\t aaaaa");
  uchar data[4] = {'h', 'o', 'l', 'a'};
  build.KeyWV("valora").ValU(data);
  build.EndObj();
  build.PutChar('"');
  Print(build.Build()); 
/*
"{\"Valor\":\"\\u2514\",\"asas\":\"\\r\\n\\v\\t aaaaa\",\"valora\":\"hola\"}"
*/  

Iterate an object

TSN::CJsonIteratorObj it = root.BeginObj();
while(it.IsValid())
 {
  PrintFormat("%s : %s", it.Key(), it.Val().ToString());
  it.Next();
 }

Iterate an array

TSN::CJsonNode arr = root["prices"];
TSN::CJsonIteratorArray it = arr.BeginArr();
while(it.IsValid())
 {
  Print(it.Val().ToDouble(0.0));
  it.Next();
 }

Parse from file

TSN::CJsonParser parser;
parser.AssingFile("data.json", false);
parser.CorrectPadingForSwar();
parser.Parse();

Parse a .jsonasm file (text dump of the tape)

AssingFile detects the .jsonasm extension and loads the raw text — you still need to call ParseAssembly() (instead of Parse()) to turn it into a tape:

TSN::CJsonParser parser;
parser.AssingFile("data.jsonasm", false);
parser.CorrectPadingForSwar();
parser.ParseAssembly();

TSN::CJsonNode root = parser.GetRoot();

You can also go the other way and dump an already-parsed tape into .jsonasm text (useful for debugging/inspecting the tape):

uchar asm_text[];
parser.ToJsonAsm(asm_text);
Print(CharArrayToString(asm_text));

Compile and reload a .jsontc tape cache

SaveCompiled writes the parsed tape (plus the source JSON and every perfect-hash table already built) to a binary .jsontc file. Reloading it via AssingFile skips parsing entirely:

//--- Compile once
TSN::CJsonParser parser;
parser.Assing(my_json_string);
parser.CorrectPadingForSwar();
parser.Parse();
parser.GetRoot()["symbol"]; // (optional) warm up perfect-hash tables before saving
parser.SaveCompiled("data.jsontc", false);

//--- Reload later — no Parse()/ParseAssembly() needed, tape and hash tables are restored as-is
TSN::CJsonParser parser2;
parser2.AssingFile("data.jsontc", false);
TSN::CJsonNode root = parser2.GetRoot();

There's also an overload that stores a reference to an external JSON file instead of embedding it:

parser.SaveCompiled("data.jsontc", false, "data.json", false);

---

Performance

Benchmark: twitter.json (616.7 KB), 1000 iterations. Same file, Iterations, and Machine (Laptop)

Note: Here we measure how quickly the JSON is parsed.
All benchmarks are located in: Test\Ben\

Parser	Language	Time (ms, total / 1000 iter)
simdjson::dom::parser (reused parser)	C++	381
simdjson.Parser (reused parser)	Python	475
JsonParserByLeo ASM (One copy of array - JSONASM File R460kb~)	MQL5	652-653 Stable
JsonParserByLeo (One copy of array)	MQL5	662 - 667
fast_json v3.4 (One copy of array)	MQL5	1229 - 1300
YamlParserByLeo (One copy of array)	MQL5	1300-1304
Claude Opus 4.8 (Effort=Max) (Best Ai now) (Extended reasoning activated) Generating code (Fast json lib) (One copy of array)	MQL5	1312-1313 Stable
orjson	Python	1863
ding9736/MQL5-JsonLib (One copy of string) (no dom building only tape parsing)	MQL5	2145-2163
ryml (Pure parse time)	C++	2846.55
MQL5 Lite (goose) Metaeditor AI Model	MQL5	3242-3250
simdjson.Parser + as_dict	Python	4309
simdjson.loads	Python	4955
ujson	Python	5322
json (stdlib)	Python	5860
JAson (One copy of array)	MQL5	~(19070-19220)
ToyJson3 (One copy of string) Only tokenization	MQL5	~(24600) Stable
CJsonNode (One copy of array)	MQL5	~(76720-80470)

Performance in acceses

Benchmark: test.json, 1000 iterations. Same file, Iterations, and Machine (Laptop)

Note: Here we measure how fast access to a node: root["key..."].
All benchmarks are located in: Test\All\

Parser	Language	Time (microseconds, total / 1000 iter)
JsonParserByLeo	MQL5	161-163
fast_json v3.4	MQL5	223 Stable
Claude Opus 4.8 (Effort=Max) (Best Ai now) (Extended reasoning activated) Generating code (Fast json lib) (One copy of array)	MQL5	595-605

Performance notes

• Machine: (Laptop Lenovo ~2017, Windows 10 build 19045, Intel Core i5-8250U @ 1.60GHz, AVX2, GMT-5).
• MQL5 run: MetaTrader 5 x64 build 5836
• Python/C++ runs: Same machine (py = 3.10.9) and C++ (CXX23~)

---

Repository Structure

JsonParserByLeo/
├── Src/ # Full code (Defines, Node, Parser)
└── Test/ # Test and Benchmarks (vs)

---

License

Read Full License
By downloading or using this repository, you accept the license terms.

---

Requirements

See dependencies.json for the full list.

---

Installation

cd "C:\Users\YOUR_USER\AppData\Roaming\MetaQuotes\Terminal\YOUR_ID\MQL5\Shared Projects"
tsndep install "https://forge.mql5.io/nique_372/JsonParserByLeo.git"

Requires the tsndep package, available on PyPI. It automatically downloads and installs all declared dependencies.

---

Quick Start

1. Include the library:

#include "..\\JsonParserByLeo\\Src\\JsonNode.mqh"

2. Parse and access:

TSN::CJsonParser parser;
parser.Assing(my_json_string);
parser.CorrectPadingForSwar();
if(parser.Parse())
 {
  TSN::CJsonNode root = parser.GetRoot();
  double price = root["price"].ToDouble(0.0);
 }

3. Re-parse without re-copying (benchmark pattern):

parser.Assing(raw_string);   // copy once
parser.CorrectPadingForSwar();
for(int i = 0; i < 1000; i++)
  parser.Parse();             // re-parse in-place

---

Contact

• Platform: MQL5 Community
• Profile: https://www.mql5.com/es/users/nique_372
• Articles: https://www.mql5.com/es/users/nique_372/publications