High-Performance JSON

[Jonathan Pereira]

If you are integrating Large Language Models (LLMs), the standard  JAson.mqh   The library is a bottleneck. Its recursive architecture and excessive memory allocations (creating thousands of objects for a single payload) are unacceptable for production systems.

We rebuilt the JSON parser from scratch.

Introducing   fast_json

An iterative, production-ready, resource-free JSON library designed for speed and stability.

Key technical improvements:

1. Tape Architecture with Zero Allocation : Uses a single contiguous memory buffer ( long[] ) instead of thousands of fragmented objects. Drastically reduces the pressure on the garbage collector.
2. Iterative state machine : replaces recursion with an explicit stack, eliminating the risk of stack overflow even with deeply nested payloads.
3. SWAR Scanning : Implements SIMD (SWAR) text scanning to process 8 bytes per CPU cycle.
4. O(1) Introspection :   HasKey()  and   Size()  They are instantaneous, unlike the linear O(N) search found in legacy libraries.

Benchmark results (Intel i7-10750H)

Operation	fast_json	Legacy Library (Jason)	Accelerate
To analyze	137 ms	1540 ms	11.2x
Serialize	264 ms	568 ms	2.1x
Total round trips	421 ms	2129 ms	5.0x

Availability

The library is open source and available immediately.

Repository:   https://forge.mql5.io/14134597/fast_json.git

Codebase: https://www.mql5.com/pt/code/68596