benchmarks/

JSON Benchmarks

This directory contains benchmarking tools to compare JSON against the standard library encoding/json when compiled to WebAssembly using TinyGo.

Build Script

The build.sh script compiles the WASM binaries with different JSON implementations.

Usage

# Build with JSON (default)
./build.sh

# Build with encoding/json (stdlib)
./build.sh stlib

Run the benchmark server

# Start the local server to serve the compiled WASM
go run ./web/server.go

You can then open http://localhost:8080 in a browser to see the benchmark results.

Output

The compiled WASM binary is output to web/public/main.wasm.

The script will display:

• Uncompressed binary size
• Gzipped size (for realistic deployment comparison)

Source Files

• clients/tinyjson/main.go - Implementation using JSON
• clients/stdlib/main.go - Implementation using encoding/json (stdlib)

Both files implement the same functionality to ensure fair comparison.

Performance Results

Last updated: 2026-07-11 (after fixing fmt.JSONEscape to bulk-copy safe byte runs instead of writing one byte at a time — see fmt/docs/PLAN.md; average of 3 runs, go test -bench=. -benchmem -count=3 ./tests/...)

Go Benchmark (go test -bench)

Run: go test -bench=. -benchmem ./tests/...

Before the JSONEscape fix, Encode was essentially tied with stdlib (753 ns/op vs 758 ns/op) — a CPU profile (go tool pprof) found fmt.JSONEscape writing one byte at a time accounted for 43.75% of Encode's CPU time, most of it spent re-escaping static, always-safe JSON field names on every call. Fixing it to bulk-copy safe runs (mirroring how encoding/json scans for the next special character) dropped JSONEscape out of the profile entirely and made Encode ~24% faster than stdlib instead of a wash.

WASM Binary Size

Compiled with tinygo build -target wasm -no-debug -opt=z.

Run: ./build.sh (tinywasm/json) and ./build.sh stlib (stdlib) from this directory.

Analysis

tinywasm/json is 83% smaller gzipped (20 KB vs 118 KB) making it ideal for web apps where bundle size matters. By eliminating the reflect package, it not only significantly reduces the final WASM binary size, but also makes encoding ~1.3x faster, decoding ~2.1x faster, and roundtripping ~1.7x faster than the standard library.

Use tinywasm/json when: Bundle size and raw decoding performance are critical, or running in restricted WASM environments. Use Stdlib when: You need to work dynamically with arbitrary unknown schemas (like map[string]any), as tinywasm/json is optimized strictly for predefined structs (fmt.Encodable/fmt.Decodable).

See the main README for detailed benchmark results.