testing

name	testing
description	Instructions for writing and maintaining tests in GitHub Desktop. Covers unit tests, UI component tests, and ad-hoc E2E tests. Use this skill when implementing features or bugfixes to write relevant tests, update existing tests, run the full suite to check for regressions, and produce screenshots and videos for Pull Request documentation.

Testing in GitHub Desktop

This document describes the three tiers of tests in GitHub Desktop, how to run them, and the patterns you should follow when writing new tests or updating existing ones as part of a feature or bugfix.

Overview

Tier	Purpose	Location	Runner
Unit / integration (non-UI)	Pure logic, stores, models, git operations	app/test/unit/	node:test via yarn test
UI component	React components rendered in JSDOM	app/test/unit/ui/	node:test + React Testing Library via yarn test
E2E (ad-hoc)	Full app launched with Playwright + Electron	app/test/e2e/	Playwright via yarn test:e2e:*

When to use each tier

• Unit / integration: new or changed logic in app/src/lib/, app/src/models/, git operations, store behavior, utility functions, IPC contracts.
• UI component: new or changed React components, dialog behavior, banners, toolbar items, list rendering.
• Ad-hoc E2E: only for temporary validation of a feature or bugfix across the full app. E2E tests you write are meant to be run locally and to capture screenshots/video for the PR, not to be merged into the permanent smoke suite.

---

Running Tests

# All unit and UI tests
yarn test

# A specific test file
yarn test app/test/unit/my-feature-test.ts

# All tests in a directory (recursive)
yarn test app/test/unit/ui

# E2E — build unpackaged app + run (fast local iteration)
yarn test:e2e:unpackaged

# E2E — run against an already-built unpackaged app
DESKTOP_E2E_APP_MODE=unpackaged npx playwright test --config app/test/e2e/playwright.config.ts

# E2E — full packaged build + run (production-like)
yarn test:e2e:packaged

The test runner (script/test.mjs) discovers files matching -test.(ts|tsx|js|jsx|mts|mjs) recursively in app/test/unit/ by default, or in the paths you pass.

---

Test Verification Workflow

After implementing any change you must run the full unit test suite:

yarn test

If any tests fail:

1. Determine whether the failure is a regression (a bug you introduced) or an expected behavior change (your change intentionally altered the behavior).
2. If it is a regression, fix the code.
3. If it is an expected change, update the test assertions so they reflect the new correct behavior.
4. Re-run the suite until everything passes.

Then verify linting:

yarn lint

If lint errors are reported and you want to auto-fix them:

yarn lint:fix

Note: yarn lint:fix rewrites files across the repository (Prettier + ESLint --fix). Only run it when you intend to apply those edits — do not use it as a read-only check.

---

Bug-First Testing

When fixing a bug:

1. Write a failing test first that reproduces the bug.
2. Verify the test fails.
3. Apply the fix.
4. Verify the test now passes.

This proves the fix works and protects against regressions.

---

Unit / Integration Tests (Non-UI)

File conventions

• Location: app/test/unit/, mirroring the source tree (e.g. app/src/lib/git/clone.ts → app/test/unit/git/clone-test.ts).
• File name: *-test.ts.
• Extension: .ts (use .tsx only when the file contains JSX).

Imports

import { describe, it, beforeEach } from 'node:test'
import assert from 'node:assert'

Use node:assert for all assertions — never Jest or Chai matchers.

Test structure

Synchronous tests are fine for pure logic:

describe('MyFeature', () => {
  it('does something useful', () => {
    const result = myFunction('input')
    assert.equal(result, 'expected')
  })
})

Use async when the test or its helpers need it. Pass the test context t when using helpers that register cleanup via t.after():

it('creates a repo', async t => {
  const repo = await setupEmptyRepository(t)
  // repo's temp directory is cleaned up automatically after the test
})

Assertion patterns

Pattern	Use
assert.equal(a, b)	Abstract equality (==) — use when coercion is intentional
assert.strictEqual(a, b)	Strict equality (===) — preferred; catches type mismatches
assert.deepEqual(a, b)	Deep structural equality
assert.notEqual(a, b)	Abstract inequality (!=)
assert.notStrictEqual(a, b)	Strict inequality (!==)
assert.ok(value)	Truthy check
assert.rejects(asyncFn, /pattern/)	Async rejection with message
assert.throws(fn, /pattern/)	Sync throw

assert.equal vs assert.strictEqual: assert.equal(a, b) uses the == operator (abstract equality), so assert.equal(42, '42') passes. assert.strictEqual(a, b) uses ===, so it also checks that types match. Prefer assert.strictEqual in most cases to avoid silent type-coercion surprises. Use assert.equal only when you explicitly want coercion semantics.

Existing helpers — reuse them

Helper file	Key exports	Purpose
app/test/helpers/repositories.ts	setupEmptyRepository(t), setupFixtureRepository(t, name), setupConflictedRepo(t)	Create temporary git repos with automatic cleanup
app/test/helpers/repository-scaffolding.ts	makeCommit(), createBranch(), switchTo(), cloneRepository()	Build git state (commits, branches)
app/test/helpers/temp.ts	createTempDirectory(t)	Temporary directory with auto-cleanup via t.after()
app/test/helpers/mock-api.ts	createMockAPI(overrides), createMockAPIRepository(), createMockAPIIdentity()	Proxy-based mock API — rejects unmocked methods to prevent real HTTP requests
app/test/helpers/mock-ipc.ts	MockIPC	Records send()/invoke() calls, simulates main→renderer messages via emit()
app/test/helpers/app-store-test-harness.ts	createTestStores(), createTestAccountsStore(), createTestRepositoriesStore()	Factory functions for wired-up test store instances backed by in-memory storage
app/test/helpers/test-stats-store.ts	TestStatsStore	In-memory stats store for verifying metric increments
app/test/helpers/stores/	InMemoryStore, AsyncInMemoryStore	Key-value stores for testing code that depends on persistent storage
app/test/helpers/databases/	TestRepositoriesDatabase, TestIssuesDatabase, etc.	Dexie database wrappers with reset() for cleanup
app/test/helpers/git.ts	getTipOrError(), getRefOrError(), getBranchOrError()	Safe git object accessors for tests
app/test/helpers/random-data.ts	generateString()	Random hex strings using crypto

Patterns to follow

Factory functions for dependencies — create stores, databases, and API instances through dedicated factory functions, not raw constructors:

const stores = createTestStores()
const api = createMockAPI({
  fetchRepository: async () => createMockAPIRepository(),
})

Promise wrappers with timeouts for callback-based async APIs:

async function waitForResult(store, ...args): Promise<Result> {
  return new Promise((resolve, reject) => {
    const timeout = setTimeout(
      () => reject(new Error('Timed out')),
      5_000
    )
    store.getResult(...args, result => {
      clearTimeout(timeout)
      resolve(result)
    })
  })
}

State machine testing — verify store transitions by calling methods and asserting intermediate states:

signInStore.beginDotComSignIn()
const state = signInStore.getState()
assert.equal(state?.kind, SignInStep.Authentication)

Compile-time contract verification — use TypeScript's type system to catch missing cases at compile time (see ipc-contract-test.ts for example):

type AssertExactUnion<TExpected, TActual> = [
  Exclude<TExpected, TActual>,
  Exclude<TActual, TExpected>,
] extends [never, never]
  ? true
  : never

---

UI Component Tests

File conventions

• Location: app/test/unit/ui/.
• File name: *-test.tsx (must be .tsx for JSX).

Critical import rule

Always import render utilities from the project's wrapper module:

import { render, fireEvent, screen, waitFor, within } from '../../helpers/ui/render'

Never import directly from @testing-library/react. The wrapper module (app/test/helpers/ui/render.tsx) imports app/test/helpers/ui/setup.ts as a side-effect, which:

1. Polyfills ResizeObserver (not available in JSDOM).
2. Aligns globalThis.Event/CustomEvent with the jsdom window versions.
3. Registers an afterEach(cleanup) hook so the DOM is cleaned between tests.

Skipping this import will cause test failures or leaks.

Rendering and querying

import assert from 'node:assert'
import { describe, it } from 'node:test'
import * as React from 'react'
import { render, screen, fireEvent } from '../../helpers/ui/render'

describe('MyComponent', () => {
  it('renders a button and responds to clicks', () => {
    let clicked = 0
    render(<MyComponent onClick={() => clicked++} />)

    const button = screen.getByRole('button', { name: 'Submit' })
    assert.ok(button)

    fireEvent.click(button)
    assert.equal(clicked, 1)
  })
})

Querying elements:

Method	Use
screen.getByRole('button', { name: 'X' })	Accessible role + name (preferred)
screen.getByText('Hello')	Visible text content
screen.getByTestId('my-id')	data-testid attribute
view.container.querySelector('.css-class')	CSS selector on the render container
screen.queryByRole(...)	Returns null instead of throwing (for absence checks)

Assertions use node:assert, not Jest matchers:

assert.notEqual(view.container.querySelector('.my-class'), null)
assert.equal(screen.queryByRole('button', { name: 'Gone' }), null)

Re-rendering

const view = render(<MyComponent visible={true} />)
// ... assert initial state ...
view.rerender(<MyComponent visible={false} />)
// ... assert updated state ...

Callback verification

Capture callbacks in local variables and assert after interaction:

let dismissed = 0
render(<Banner onDismissed={() => dismissed++} />)
fireEvent.click(screen.getByRole('button', { name: 'Dismiss this message' }))
assert.equal(dismissed, 1)

Timer mocking

For components with timeouts (banners, auto-dismiss, debounce):

import { afterEach, beforeEach, describe, it } from 'node:test'
import {
  advanceTimersBy,
  enableTestTimers,
  resetTestTimers,
} from '../../helpers/ui/timers'

describe('auto-dismissing banner', () => {
  beforeEach(() => enableTestTimers(['setTimeout']))
  afterEach(() => resetTestTimers())

  it('dismisses after timeout', () => {
    let dismissed = 0
    render(<Banner timeout={500} onDismissed={() => dismissed++} />)

    advanceTimersBy(500)
    assert.equal(dismissed, 1)
  })
})

Clipboard testing

Register restore() in afterEach so the mock is always torn down even when an assertion throws:

import { afterEach, it } from 'node:test'
import { captureClipboardWrites } from '../../helpers/ui/electron'

describe('CopyButton', () => {
  let restore: () => void
  let writes: string[]

  afterEach(() => restore?.())

  it('copies text to clipboard', () => {
    ;({ writes, restore } = captureClipboardWrites())
    render(<CopyButton text="hello" />)
    fireEvent.click(screen.getByRole('button'))
    assert.deepEqual(writes, ['hello'])
  })
})

Calling restore() inline at the end of the test body is not safe — if any assertion before it throws, the global clipboard.writeText mock stays patched and will silently contaminate subsequent tests.

ESLint note

The react/jsx-no-bind rule is disabled for test files, so inline arrow functions in JSX are fine in tests.

---

Ad-hoc E2E Tests

E2E tests launch the real Desktop app via Playwright's Electron support. Use them only for temporary validation of your work — to capture screenshots and video for the Pull Request. Do not add tests to the permanent smoke suite (app-launch.e2e.ts) unless explicitly asked.

File conventions

• Location: app/test/e2e/.
• File name: *.e2e.ts (Playwright config matches this pattern).
• Do not modify app-launch.e2e.ts unless explicitly asked.

⚠️ Delete ad-hoc specs before opening your PR. Playwright's config matches every *.e2e.ts file in app/test/e2e/, so any file you create there will run in CI. Ad-hoc specs are for local validation only — stage and run them locally, then git rm them before committing.

Imports

import {
  test,
  expect,
  controlMockServer,
  getMockRequests,
  dismissMoveToApplicationsDialog,
} from './e2e-fixtures'
import type { Page } from '@playwright/test'

Test structure

test.describe.configure({ mode: 'serial' })

test.describe('My Feature E2E', () => {
  test('launches app and shows feature', async ({ mainWindow: page }) => {
    // Wait for the React app to mount
    await page.waitForFunction(
      () =>
        (document.getElementById('desktop-app-container')?.innerHTML.length ??
          0) > 100,
      null,
      { timeout: 30000 }
    )

    // ... interact with the app ...
  })
})

All tests run serially in the same Electron session (one app launch per test file).

Locating elements

// CSS selector
const button = page.locator('button:has-text("Finish")')

// XPath
const item = page.locator('//div[contains(@class, "list-item")]')

// Waiting for visibility
await button.waitFor({ state: 'visible', timeout: 15000 })

Setting React controlled inputs

For most inputs, Playwright's .fill() works fine. However, some React controlled inputs ignore .fill() because they rely on React's synthetic event system rather than native DOM events. If .fill() doesn't update the React state (i.e., the value appears empty after filling), use this workaround that fires both input and change events through React's internal value setter:

await input.evaluate((el, value) => {
  const inp = el as HTMLInputElement
  Object.getOwnPropertyDescriptor(HTMLInputElement.prototype, 'value')
    ?.set?.call(inp, value)
  inp.dispatchEvent(new Event('input', { bubbles: true }))
  inp.dispatchEvent(new Event('change', { bubbles: true }))
}, 'my-value')

Use .fill() first — only fall back to the workaround when .fill() does not produce the expected state change in React.

Assertions

Direct assertions on locators:

await expect(locator).toContainText('expected text', { timeout: 15000 })
await expect(locator).toBeVisible()
await expect(locator).not.toBeVisible()

Polling assertions for async conditions (git state, server requests):

await expect
  .poll(() => getSmokeRepoCurrentBranch(), {
    timeout: 15000,
    intervals: [1000],
  })
  .toBe('my-branch')

IPC events

Trigger menu events or app actions from the renderer:

await page.evaluate(() => {
  require('electron').ipcRenderer.emit('menu-event', {}, 'show-about')
})

Taking screenshots

Take screenshots at key UI moments during the test. Save them under playwright-videos/ so they are collected alongside videos:

await page.screenshot({
  path: 'playwright-videos/01-feature-dialog-open.png',
})

Name screenshots with a numeric prefix so they appear in order. Be descriptive:

await page.screenshot({ path: 'playwright-videos/02-branch-created.png' })
await page.screenshot({ path: 'playwright-videos/03-diff-view.png' })

Video recording

Videos are recorded automatically by the fixture configuration at 1280×800 resolution. They are saved in the playwright-videos/ directory. You do not need to configure recording — just run the tests.

Running ad-hoc E2E tests

For local iteration, use the unpackaged mode to avoid a full packaging step:

# Build unpackaged + run all E2E tests
yarn test:e2e:unpackaged

# Run only your specific test file (after building)
DESKTOP_E2E_APP_MODE=unpackaged npx playwright test \
  --config app/test/e2e/playwright.config.ts \
  app/test/e2e/my-feature.e2e.ts

⚠️ Do NOT use yarn build:dev for E2E tests. The development build produces an index.html that loads the renderer bundle from http://localhost:3000/build/renderer.js (the webpack dev server). Without the dev server running, the React app never mounts and the Playwright waitForFunction on desktop-app-container will time out silently.

Always use yarn test:e2e:build:unpackaged (or the combined yarn test:e2e:unpackaged) which runs a production build with DESKTOP_SKIP_PACKAGE=1. This bundles renderer.js directly into out/ so the app is self-contained.

Handling the welcome flow and macOS dialogs

If your test launches from a fresh state, you will encounter the welcome flow. Handle it like the smoke test does:

// Skip the welcome sign-in
const skipButton = page.locator('a.skip-button')
await skipButton.waitFor({ state: 'visible', timeout: 30000 })
await skipButton.click()

// Fill name/email and finish
const nameInput = page.locator('input[placeholder="Your Name"]')
await nameInput.waitFor({ state: 'visible', timeout: 15000 })
if ((await nameInput.inputValue()) === '') {
  await nameInput.fill('GitHub Desktop E2E')
}
const emailInput = page.locator('input[placeholder="[e-mail]"]')
if ((await emailInput.inputValue()) === '') {
  await emailInput.fill('[e-mail]')
}
await page.locator('button:has-text("Finish")').click()
await page.waitForSelector('#welcome', { state: 'hidden', timeout: 15000 })

// Dismiss macOS "Move to Applications" dialog if it appears
await dismissMoveToApplicationsDialog(page)

Attaching artifacts to Pull Requests

After running E2E tests, collect artifacts from playwright-videos/:

• Screenshots: the .png files you captured with page.screenshot().
• Videos: the .webm files recorded automatically by Playwright.
• Traces: the trace-*.zip files saved by the fixture teardown.

Attach the screenshots and video to the Pull Request description or as comments to show the new UI additions and prove the feature works end-to-end.

Faking data or state for ad-hoc E2E screenshots

Some UI features are only visible under specific conditions — for example, a signed-in GitHub.com account, a populated model list, or an active Copilot subscription. In ad-hoc E2E tests whose sole purpose is capturing screenshots for a Pull Request, you can temporarily modify source code to bypass those conditions and show the full UI.

Workflow:

1. Make the minimum temporary changes needed (e.g. hardcode a store property, inject fake data, force a boolean flag).
2. Rebuild with yarn test:e2e:build:unpackaged.
3. Run your ad-hoc E2E spec and capture screenshots/video.
4. Revert every temporary change before committing. Verify with git diff <file> that no fake data leaks into the branch.
5. Delete the ad-hoc E2E spec.

Tips:

• Prefer overriding in getState() (in AppStore) rather than deep in a store or API layer. This keeps the blast radius small — a single file, a few lines — and easy to revert cleanly.
• Use realistic but obviously fake data. If you inject model names, use real-looking IDs and display names so the screenshots read naturally.
• Guard with __DEV__ only if you plan to commit the fake data (not recommended). For ad-hoc screenshots the code is reverted immediately, so a plain unconditional override is simpler and avoids issues with __DEV__ being false in production E2E builds (RELEASE_CHANNEL=production).
• Watch out for tree-shaking. The E2E build uses NODE_ENV=production. Constants like __DEV__ are false in that mode, so any code guarded by if (__DEV__) will be dead-code-eliminated by webpack. If you need the fake data to survive the production build, don't guard it.
• Verify the revert is complete. After capturing artifacts, run git diff -- <files you touched> and confirm zero output before moving on.

Example — forcing a populated Copilot model list:

// In AppStore.getState(), temporarily replace:
copilotModels: this.copilotModels,
copilotAvailable: this.copilotStore.isAvailable,

// With:
copilotModels:
  this.copilotModels !== null && this.copilotModels.length > 0
    ? this.copilotModels
    : fakeModels,       // defined as a const above the class
copilotAvailable: true,

After screenshots are captured, revert these two lines back to their originals.

---

Test Helpers Reference

Global test environment (app/test/globals.mts)

This file is loaded automatically by the test runner. It:

• Imports fake-indexeddb/auto and global-jsdom/register for browser API simulation.
• Defines Webpack build-time constants (__DEV__, __APP_NAME__, etc.).
• Mocks the electron module (clipboard, shell, ipcRenderer).
• Removes MessageChannel/MessagePort/BroadcastChannel to prevent test hangs (React 16 + Dexie cleanup issue).

You do not need to set up any of this manually — it runs before every test file.

Environment variables (.test.env)

Loaded automatically by the test runner. Sets GIT_AUTHOR_NAME, GIT_COMMITTER_NAME, etc. so git operations produce deterministic results.

---

Checklist

When you are done implementing a feature or bugfix, verify:

• Wrote unit tests for new or changed logic.
• Wrote UI component tests for new or changed React components.
• Existing tests updated if behavior intentionally changed.
• yarn test passes with no failures.
• yarn lint passes (run yarn lint:fix to auto-fix if needed).
• (If applicable) Ran ad-hoc E2E test and captured screenshots/video.
• (If applicable) Attached screenshots and video to the PR.