code_size

Embedded Binary Size Benchmark

Measures JsonFusion's code footprint for embedded ARM Cortex-M7 targets.

Overview

This benchmark compiles a realistic embedded configuration parser and measures the resulting binary size. The goal is to understand JsonFusion's flash memory requirements for resource-constrained embedded systems.

Model

The test uses EmbeddedConfig - a realistic embedded configuration with:

• Fixed-size arrays (static allocation, no heap)
• Validation constraints (range<>, min_items<>)
• Nested structures (Network, Controller, Motor, Sensor, Logging)
• Optional fields (std::optional)
• ~2.5 KB data size (16 motors, 16 sensors, various strings)

This represents a typical embedded device configuration (motor controllers, IoT devices, industrial systems).

Target Platform

ARM Cortex-M7 (common in STM32H7, SAMV7, etc.):

• 32-bit ARM architecture
• Thumb-2 instruction set
• Hardware floating-point unit (FPv5-D16)
• Representative of modern embedded systems

Quick Start

Recommended: Use the Python build script (easier to maintain and extend):

./build.py

This builds 2 configurations:

1. size_optimized (-Os -flto) - Minimize code size
2. aggressive_opt (-O3 -flto) - Maximum speed

Results

Each build outputs:

• Code size (.text section) - Flash memory usage
• Const data (.rodata section) - Validation tables, literals
• Initialized data (.data section) - Should be minimal
• Uninitialized data (.bss section) - Global config (~2.5 KB)
• Top symbols - Largest functions by size

Comparisons

To compare with alternatives:

Hand-Written Parser

Create minimal hand-written parser (no validation, error-prone).

Other Embedded Libraries

• ArduinoJson
• jsmn (minimal tokenizer)
• cJSON
• yajl (event-driven / SAX parser)