Skip to content

Latest commit

 

History

History

benchmark-a-function

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

README.md

Benchmark a function

bench() is the library's in-process microbenchmarking path - mitata-shaped time budget + min-samples, batched when a call is sub-microsecond, no subprocess or wall-clock command involved. Register tasks with group()/task() in a suite file, then run the suite through the library:

// suite.ts
import { group, task } from "ostia"

group("dedupe", () => {
  task("naive (indexOf scan, O(n²))", () => dedupeNaive(input))
  task("Set-based (O(n))", () => dedupeSet(input))
})
// demo.ts
import { bench, renderers } from "ostia"

const doc = await bench({ suites: ["suite.ts"] })
const { text } = await renderers.table.render(doc, {})
console.log(text)
Task                            Median     Spread             Range              Relative
-----------------------------------------------------------------------------------------
dedupe:
  naive (indexOf scan, O(n²))   217.8 µs   208.3 µs…227.2 µs  197.3 µs…825.5 µs  9.08× slower
  Set-based (O(n))              24.0 µs    23.2 µs…24.9 µs    21.6 µs…258.8 µs   1.00×

Each suite file runs in its own spawned child, isolated from the caller's state - the same isolation ostia bench suite.ts... gives you from the CLI. The result is a regular ProfileDocument: same renderers as ostia time/ostia compare, and it can be saved and diffed against a baseline the same way - see gate-a-regression for the CLI-config version of that, or bench/README.md at the repo root for how this project benchmarks its own hot paths.

Run it

bun run demo