Benchmarks

All numbers are median of 3 runs on macOS (Apple Silicon), Node.js v22.17.0. TypeDoc 0.28.18 with --skipErrorChecking flag.

Real-World: es-toolkit (603 files, 1322 symbols)

Measured against es-toolkit.

Metric	oxdoc	TypeDoc 0.28	Speedup
JSON Generation	0.24s	1.70s	7x faster
HTML Generation	0.25s	2.53s	10x faster
Peak Memory	131MB	470MB	3.6x less

Real-World: radashi (162 files, 437 symbols)

Measured against radashi.

Metric	oxdoc	TypeDoc 0.28	Speedup
JSON Generation	0.13s	1.12s	8.6x faster
Peak Memory	84MB	272MB	3.2x less

Synthetic Scale Test

Generated fixtures with increasing file counts, each file containing 3 documented symbols.

Files	Symbols	Time	Memory	Throughput
100	300	0.03s	6MB	~3,300/s
500	1,500	0.10s	10MB	~5,100/s
1,000	3,000	0.18s	6MB	~5,700/s
5,000	15,000	0.81s	33MB	~6,200/s

Why the Performance Difference?

TypeDoc loads the full TypeScript compiler (tsc) to resolve types, build a program graph, and check types — even with --skipErrorChecking, it still initializes the full compiler pipeline. This has a fixed overhead of ~1s regardless of project size.

oxdoc uses OXC parser (Rust NAPI) which parses syntax only — no type checking, no program graph. The parser returns AST + comments synchronously via parseSync(), and all heavy work runs in native Rust.

	TypeDoc	oxdoc
Parser	tsc (JavaScript)	OXC (Rust NAPI)
Type resolution	Full	None (signature as-is)
Fixed overhead	~1s	~0.05s
Per-file cost	~2ms	~0.15ms

Reproduce

git clone https://github.com/jiji-hoon96/oxdoc.git
cd oxdoc
pnpm install && pnpm build

# Synthetic benchmark
pnpm bench

# Real-world comparison (requires TypeDoc installed globally)
time node dist/cli/index.js generate --format json --output /tmp/oxdoc-out ./path/to/project/src
time npx typedoc --json /tmp/typedoc.json --entryPoints ./path/to/project/src/index.ts --skipErrorChecking