Skip to content

Latest commit

 

History

History
409 lines (309 loc) · 11.6 KB

README.md

File metadata and controls

409 lines (309 loc) · 11.6 KB

@benchristel/taste

A fast JavaScript test framework, for browser apps, NPM packages, and Node.js scripts.

import {test, expect, is} from "@benchristel/taste"

test("greet", {
  "greets the world"() {
    expect(greet("world"), is, "Hello, world!")
  },

  "defaults to a generic greeting"() {
    expect(greet(null), is, "Hello, you person, you!")
  },
})

function greet(name) {
  return `Hello, ${name || "you person, you"}!`
}

Taste's design philosophy is that a test is just a function. Testing doesn't have to be more complicated than that! Some nice consequences of this idea:

  • Running a test suite is just calling a function.
  • A matcher is just a function that returns a boolean.
  • An assertion is just a function that throws an exception.

The result of all this is that your tests can run anywhere JS can. You can easily run them in the browser, or in NodeJS, Bun, or Deno. Heck, your smart TV could probably run Taste, though I haven't tried this (let me know if you do).

Taste achieves this simplicity without major sacrifices to ergonomics or aesthetics. Just look at the example above!

Fast?

Oh, yes. Here are the results of running Taste's own test suite 768 times in Firefox on a 2.4 GHz Intel CPU:

134400 tests ran, and found no issues. 521 ms elapsed.

134400 tests in about half a second! I'm going to bet your program doesn't have that many tests.

You should take this measurement with a grain of salt, though, because when the same tests run many times, the JIT has more time to optimize the code. Still, I think this is pretty cool.

Build Status

Taste's own tests are written using Taste. You can run the tests in your browser here.

Try it!

https://benchristel.github.io/try-taste/

There is also a set of downloadable koans/tutorials that walk you through Taste's features from basic to advanced, and serve as a reference for how to integrate Taste into a project.

Node.js Quick Start

// copy-paste this into test.mjs
import {test, expect, is} from "@benchristel/taste"
import {getAllTests, runTests, formatTestResultsAsText} from "@benchristel/taste"

// add a test to the global suite
test("my first Taste test", {
  "runs"() {
    expect(1 + 1, is, 2)
  },
})

// run the tests and print the results
runTests(getAllTests())
  .then(formatTestResultsAsText)
  .then(console.log)

Then run time node test.mjs. Replace node with bun to go even faster! You can also use bun --hot test.mjs to run your tests on every code change.

Project Templates

Installation

yarn add @benchristel/taste

Features

  • Tests can run in the browser. You can integrate test results into the dev UI of your app!
  • Tests can live in the same files as production code. Or you can use any other organization scheme. The tests are automatically stripped out of production builds of your app.
  • The tests are fast—up to 50x faster than Jest tests. Taste can run tens of thousands of tests per second.
  • Custom matchers couldn't be simpler. Any function that returns a boolean can be used in a test assertion. Test failures still get pretty-formatted as you'd expect.

Non-features

  • async tests are not run in parallel. Thus, if you await a 100ms timer in a test, your suite will take 100ms longer to run. It's up to you to design your code so promises can resolve quickly in tests. The ideal is to resolve all promises in microtasks, which will allow your async tests to run as fast as synchronous code.
  • By default, there is no timeout for async tests. If you accidentally await a promise that never resolves, your tests will hang with no indication of what's wrong. Reduce the pain of this eventuality by running your tests on every code change so you can quickly revert mistakes. You can add timeouts to your test suite with @benchristel/taste-timeout.
  • There are no equivalents of Jest's beforeEach etc., nor are there nested describe or context blocks. You can de-duplicate repeated setup by simply extracting functions and calling them, as you would for duplicated production code. In my experience, this makes for more readable tests in the long run. However, Taste is flexible enough that you could build your own Jest-like test-definition syntax on top of it, if you wanted to.
  • Taste has no facilities for mocking. It's clearer, and easy enough, to roll your own test doubles if you need to. Or use another mocking library; there are plenty to choose from.

Recommended Integrations

  • Add Vite for auto-refresh, and you can get near-instantaneous test feedback whenever you change a JavaScript file.
  • To run browser tests in CI or as a pre-push hook, you can set up Puppeteer, which runs the tests in a headless Chrome browser. That might sound like a lot of overhead, but it's still slightly faster than using Jest (1.2s vs. 1.3s startup time on my machine).

API Documentation

Writing Tests

import {
  test,
  expect,
  is,
  equals,
  not,
  which,
} from "@benchristel/taste"

test(subject: string, TestDefinitions): void

The test method is the usual way of writing tests in Taste. The TestDefinitions type is a map of strings (test scenario names) to test functions:

type TestDefinitions = {
  [scenario: string]: () => void | Promise<void>,
}

Typical usage looks like this:

import {test, expect, is} from "@benchristel/taste"

test("a Counter", {
  "starts at zero"() {
    const c = new Counter()
    expect(c.value(), is, 0)
  },

  "increments once"() {
    const c = new Counter()
    c.increment()
    expect(c.value(), is, 1)
  },
})

When the test results are formatted using Taste's built-in formatter (formatTestResultsAsText), the subject name and scenario name will be concatenated for each test. So in the above example, you'd see failure messages like:

a Counter increments once
  expect(
    0,
    is,
    1
  )

expect(actual, predicate: () => boolean, ...expected): void

Throws an error if the given predicate returns a falsey value when passed the expected and actual values.

Note that the subject is passed as the last argument to the predicate function. This is to accommodate predicates that are designed to be curried.

Here's an example of how you might define a custom isGreaterThan predicate:

import {test, expect, is} from "@benchristel/taste"

function isGreaterThan(reference, n) {
  return n > reference
}

test("rolling a die", {
  "produces a positive number"() {
    expect(rollDie(), isGreaterThan, 0)
  },
})

is(expected, actual): boolean

Returns whether its arguments are ===.

is is curried, so the following are equivalent:

expect(1 + 1, is, 2)
expect(1 + 1, is(2))

equals(expected, actual): boolean

Returns whether two objects are deeply equal. The rules for comparison are complicated, but work similarly to the ones in Jest, Jasmine, and other test frameworks you might be familiar with. For examples of behavior, see the tests.

equals({a: 1}, {a: 1}) // true
equals({a: 1}, {a: 2}) // false

equals is curried, so the following are equivalent:

expect(1 + 1, equals, 2)
expect(1 + 1, equals(2))

not(p: () => boolean): () => boolean

Returns a negated version of the given predicate p; i.e. not(p) returns true for some arguments iff p returns falsey for those arguments.

The following are equivalent:

expect(2 + 2, not(is), 5)
expect(2 + 2, not(is(5)))

curry(func, name)

Returns a curried version of the given function. Partially-applied functions generated from the curried function will be pretty-printed by formatTestResultsAsText in a format that includes the given name.

which(predicate): magic

Given a predicate, returns a magical object that equals any value for which the predicate returns truthy. You can use which to customize how objects should be compared when using equals—usually to relax some constraint.

const isBetween = curry((min, max, n) => {
  return n >= min && n <= max
}, "isBetween")

test("randomDndCharacter", {
  "generates reasonable ability scores"() {
    expect(randomDndCharacter(), equals, {
      str: which(isBetween(7, 18)),
      dex: which(isBetween(7, 18)),
      con: which(isBetween(7, 18)),
      int: which(isBetween(7, 18)),
      wis: which(isBetween(7, 18)),
      cha: which(isBetween(7, 18)),
    })
  },
})

Running Tests

import {
  getAllTests,
  runTests,
  formatTestResultsAsText,
} from "@benchristel/taste"

getAllTests(): Array<Test>

Returns all tests that have been registered via test(). Nilpotent, but may return different results on subsequent calls if there are intervening calls to test().

runTests(Array<Test>): Promise<SuiteResults>

Runs tests asynchronously and returns their results as a machine-readable object.

formatTestResultsAsText(SuiteResults): string

Formats test results from runTests as a human-readable string.

Integrating with build tools

Snowpack

Taste works with Snowpack with no special configuration.

I recommend bundling your code for production with @snowpack/plugin-webpack. This will automatically remove any taste tests from your production build.

Non-Taste-specific note: if you use snowpack+webpack, do not put any JavaScript directly in your index.html; as of this writing, the webpack plugin will not build it correctly and it will not be run in the bundled version of your site. Instead, do something like this to include your root index.js module:

<script type="module" src="./index.js"></script>

Vite + React

Taste works great with Vite! I recommend it.

To ensure your tests run automatically and you do not see duplicate test results when components get hot-reloaded, I recommend turning hot module replacement (HMR) off for files that contain tests. You can do that like this:

import { defineConfig } from 'vite'
import react from '@vitejs/plugin-react'

// https://vitejs.dev/config/
export default defineConfig({
  plugins: [react({include: ["**/*.impl.js", "**/*.impl.jsx"]})],
  build: {
    target: "chrome91",
  },
})

Using this config, you'd place all of your non-test code in *.impl.js or *.impl.jsx files, and only those files would get hot-reloaded. An alternative would be to use something like exclude: ["**/*.test.js"] if you use that naming convention.

Removing tests from production builds

If you're using Webpack, Rollup, or most other module bundlers, you shouldn't have to do anything special to remove Taste tests from the production build of your app. If you find that tests are showing up in bundled code, ensure that process.env.NODE_ENV === "production" and your optimizer is configured for tree-shaking / dead code elimination.

Development

This section contains instructions for working on Taste itself.

Fast test runner:

yarn dev
open http://localhost:8080

Robot-friendly test runner:

yarn test