benchmark

name: benchmark description: The @typia/benchmark runner, fixtures, and per-CPU result archive. Read before running, modifying, or publishing benchmark results.

Read before running, modifying, or publishing benchmark results. The benchmark is the private @typia/benchmark workspace under benchmark/.

What it measures

Throughput (MB/sec) of typia against competing libraries, across the canonical structure fixtures (ObjectSimple, ObjectHierarchical, ObjectRecursive, the union and array variants, UltimateUnion). src/index.ts drives ten categories:

Category	Operation	Compared against
`is`	`typia.is` type guard (truthy path)	typebox, ajv, io-ts, zod, class-validator
`assert`	`typia.assert` (throwing path)	same
`validate`	`typia.validate` (error-collecting path)	same
`assert-error` / `validate-error`	the same two on invalid input	same
`optimizer`	optimized `is` codegen	typebox, class-validator
`stringify`	`typia.json.*Stringify`	fast-json-stringify, `JSON.stringify`, class-transformer
`server-assert` / `server-stringify` / `server-performance`	request/response cost in a real server	Express and Fastify, typia vs pure vs class-transformer

Each program wraps an operation in a benchmark Suite; BenchmarkServer runs programs in tgrid worker processes; BenchmarkReporter writes the markdown report and HorizontalBarChart renders the D3 SVGs.

Running

pnpm template   # regenerate src/programs/ from src/template/*.ts, then format + build
pnpm build      # ttsc build → bin/   (tsconfig applies the typia transform plugin)
pnpm start      # node bin/ → writes results for this machine
pnpm start:bun  # bun bin/bun.js → Bun-runtime results

pnpm start keys the output directory off os.cpus()[0].model, so a run writes to benchmark/results/<your CPU model>/ — README.md (the report: host metadata, one section per category with a table and an SVG) plus images/*.svg. A Bun run lands under benchmark/results/<CPU>/bun/.

Conventions

src/programs/ is generated. It is produced by pnpm template from src/template/*.ts (one template module per category). Don't hand-edit src/programs/; to add a library or category, edit the matching template module, then regenerate. The benchmark carries its own fixtures under src/structures/ (it does not import @typia/template), so add or change shapes there.
Results are archived per CPU. Each results/<CPU model>/ directory is one machine's committed report. A new machine adds a new directory; don't overwrite another machine's results with yours. The markdown tables and SVGs are the published artifact (no JSON is kept in results/).
Add a competitor by adding its programs and schemas. Library-specific schemas live under src/structures/<library>/; the template that generates that category's programs must reference them. Keep the structure set identical across libraries so a row compares like for like.