Testing practices

Testing practices

Fires when writing or modifying tests, picking a test pattern, or reviewing test code in this monorepo.

The skill covers the module-test harness, the describe/it API, test file conventions, coverage, parameterized tests, async patterns, and sinon integration.

Test framework

• Unit tests use @monochromatic-dev/module-test/ts, a custom harness built on chai, sinon, and expect-type
• Unit tests import package behavior from built dist, never from sibling source files
• Browser and e2e tests use Playwright, executed inside a podman container
• All tests run through mise run; never invoke bun test or playwright directly
• Each test file is self-contained with top-level await describe(...); no external test runner needed

Running tests

# Build all packages, then run all tests (unit + browser + e2e)
mise run buildAndTest  # alias: bt

# Build, then run a specific unit test file
mise run buildAndTest -- package/module/async-time/src/wait.unit.test.ts

# Run tests only when dist is already fresh
mise run test          # alias: t

# Watch mode
mise run watch:test    # alias: tW

Required for unit tests: unit tests import package code from built dist. Run mise run buildAndTest after source changes, including when narrowing to one file, because mise run test alone can exercise stale artifacts. Use mise run test only when the same verification sequence already rebuilt dist, or when the file tests the test harness itself instead of package output.

Browser and e2e tests

Browser and e2e tests run inside a podman container image (monochromatic-playwright). Build the image first with mise run prepare:playwright.

# Run browser tests (all browsers)
mise run test:browser

# Run browser tests for a specific browser
mise run test:browser:chromium
mise run test:browser:firefox

# Run e2e tests
mise run test:e2e

Playwright configuration lives in playwright.browser.config.ts and playwright.e2e.config.ts. The browser test server (playwright/serve.ts) uses h3.

Test file naming

• Unit tests: {name}.unit.test.ts -- co-located alongside source
• Browser tests: {name}.browser.test.ts
• E2E tests: {name}.e2e.test.ts

Test discovery uses rg --files --glob '**/*.unit.test.*' -- no configuration file needed.

Test file setup

import {
  describe,
  expect,
  it,
} from '@monochromatic-dev/module-test/ts';

import { wait, } from '../dist/final/neutral/index.mjs';

await describe({
  name: wait.name,
  children: [
    it({
      name: 'resolves to undefined',
      fn: async () => {
        const result = await wait(1,);
        expect(result,).toBeUndefined();
      },
    },),
  ],
},);

Key differences from bun: test / Jest:

• Import describe, it, expect from @monochromatic-dev/module-test/ts (not bun:test)
• Import package code under test from built dist, not from sibling src files
• Use it (not test)
• describe and it take an options object { name, children/fn }, not positional (name, callback) arguments
• describe returns a promise -- use await describe(...) at the top level
• Children are concurrent by default via Promise.allSettled

Importing code under test

Treat direct imports from sibling source files as a test bug. Tests for exported package behavior must import from the built dist entry point, for example ../dist/final/neutral/index.mjs from a src/*.unit.test.ts file. This verifies the artifact users consume and catches export-map, build, and bundling errors.

Relative source imports are allowed only for fixtures, mock data, or test-only helpers that are not part of package behavior.

Describe name conventions

When a describe block names a specific imported function, use functionName.name instead of a string literal. This keeps names in sync with refactors.

// Imported function -- use .name
describe({ name: wait.name, ... })

// Constant or descriptive category -- use string literal
describe({ name: 'SEVERITY_MAP', ... })
describe({ name: 'error code constants', ... })
describe({ name: 'valid fixtures', ... })

For the top-level describe in a file, use an empty string name: '' when the file tests a single module. The empty-name suite is invisible in output -- the filename already identifies what is being tested. Nest the real function-named describes as children.

await describe({
  name: '',
  children: [
    describe({
      name: stripAnsi.name,
      children: [ ... ],
    }),
    describe({
      name: extractRuleName.name,
      children: [ ... ],
    }),
  ],
});

Coverage requirements

Target 100% test coverage. When certain lines or branches cannot be tested (e.g. error handling for impossible states), use V8 ignore comments:

/* v8 ignore next -- @preserve */
if (impossibleCondition)
  throw new Error('This should never happen',);

// For multiple lines:
/* v8 ignore next 3 -- @preserve */
if (untestableCondition) {
  console.error('Untestable path',);
  return fallbackValue;
}

Test structure

• Use descriptive test names that explain expected behavior
• Group related tests using nested describe blocks as children
• Keep describe names unique at the same scope within a file -- duplicate names cause misattributed results (see TROUBLESHOOTING.testing.md)
• Use .map() to generate parameterized it entries in the children array
• Test both happy path and error scenarios

Parameterized tests with .map()

Since there is no test.each, generate it entries with .map() and spread into children:

await describe({
  name: shouldIgnoreFile.name,
  children: [
    ...IGNORED_EXTENSIONS.map(function mapExt(ext,) {
      return it({
        name: `returns true for ${String(ext,)} extension`,
        fn: async () => {
          expect(shouldIgnoreFile(`/some/path/file${String(ext,)}`,),).toBe(
            true,
          );
        },
      },);
    },),
    it({
      name: 'returns false for plain .ts files',
      fn: async () => {
        expect(shouldIgnoreFile('/some/path/file.ts',),).toBe(false,);
      },
    },),
  ],
},);

Test timeouts

Pass timeout in the options object (milliseconds):

it({
  name: 'fetches embeddings from external API',
  fn: async () => {
    const result = await embed({ input: 'test', },);
    expect(result,).toBeDefined();
  },
  timeout: 30_000,
},);

Suites also accept timeout which applies to all children collectively.

Skipping tests

Pass skip: true or skip: 'reason string':

it({
  name: 'unix-only behavior',
  skip: process.platform === 'win32',
  fn: async () => { ... },
})

Expected failures

Pass fails: true to mark a test that is expected to throw. A throwing test is PASS; a passing test is FAIL:

it({
  name: 'rejects invalid input',
  fails: true,
  fn: async () => {
    throw new Error('expected',);
  },
},);

Repeating tests (flake detection)

Pass repeats: N to run the test N additional times after the first:

it({
  name: 'not flaky',
  repeats: 2,  // runs 3 times total
  fn: async () => { ... },
})

Sequential execution

Children run concurrently by default. Set concurrency: 1 on describe to run children one at a time in array order:

describe({
  name: 'ordered operations',
  concurrency: 1,
  children: [
    it({ name: 'step 1', fn: async () => { ... } }),
    it({ name: 'step 2', fn: async () => { ... } }),
  ],
})

Children are lazy descriptors and do not run until the parent dispatches them, so concurrency: 1 sequences execution without any wrapping. The parent's effective concurrency is inherited by nested describes that don't set their own, so a single concurrency: 1 at the top sequences all descendants.

Region markers

Use //region and //endregion markers to organize test groups within a describe block:

await describe({
  name: wait.name,
  children: [
    //region Delay behavior

    it({
      name: 'resolves after the specified delay',
      fn: async () => {
        const DELAY = 20;
        const TOLERANCE = 5;

        const start = performance.now();
        await wait(DELAY,);
        const elapsed = performance.now() - start;

        expect(elapsed,).toBeGreaterThanOrEqual(DELAY - TOLERANCE,);
      },
    },),

    //endregion Delay behavior

    //region Return value

    it({
      name: 'returns undefined',
      fn: async () => {
        const result = await wait(1,);
        expect(result,).toBeUndefined();
      },
    },),
    //endregion Return value
  ],
},);

Region markers provide IDE folding and navigability in long test files.

Assertion count verification

The it function provides a scoped expect via the test context parameter. Use ctx.expect.assertions(n) to verify exactly N assertions run:

it({
  name: 'all assertions execute',
  fn: async ctx => {
    ctx.expect.assertions(2,);
    ctx.expect(await readFile(path,),).toBe('content',);
    ctx.expect(exists,).toBe(true,);
  },
},);

The global expect also works but does not support assertion counting.

Sinon sandbox for spies/stubs

The test context provides a sinon sandbox that auto-restores after the test:

it({
  name: 'logs a message',
  fn: async ctx => {
    const spy = ctx.sinon.spy(console, 'log',);
    doSomething();
    expect(spy,).toHaveBeenCalledWith('hello',);
  },
},);

Async error assertions

Await the async operation first, then assert on the caught error. .rejects and .resolves are legacy APIs -- avoid them in new test code.

.rejects.toThrow() passes the rejected value to chai's .throw(), which expects a function to call. The harness patches around this mismatch, but the indirection obscures stack traces and is inherently fragile. .resolves.toBe(x) adds nothing over const v = await p; expect(v).toBe(x).

// Preferred: await first, assert on the error
it({
  name: 'rejects on missing file',
  fn: async () => {
    let caught: unknown;
    try {
      await readConfig('/nonexistent',);
    }
    catch (error) {
      caught = error;
    }
    expect(caught,).toBeInstanceOf(ConfigError,);
    expect((caught as Error).message,).toContain('not found',);
  },
},);

// Preferred: await first, assert on the resolved value
it({
  name: 'returns parsed config',
  fn: async () => {
    const result = await readConfig('/valid',);
    expect(result,).toHaveProperty('version', 2,);
  },
},);

Type-level testing

Use expectTypeOf from expect-type, re-exported by @monochromatic-dev/module-test/ts:

import {
  describe,
  expectTypeOf,
  it,
} from '@monochromatic-dev/module-test/ts';

import type { ChatRole, } from '../dist/final/neutral/index.mjs';

await describe({
  name: 'ChatRole',
  children: [
    it({
      name: 'is exactly the role union',
      fn: async () => {
        expectTypeOf<ChatRole>().toEqualTypeOf<'system' | 'user' | 'assistant'>();
      },
    },),
  ],
},);

Available matchers

The expect function provides Jest-compatible matchers backed by chai:

Value matchers: toBe, toEqual, toStrictEqual, toContain, toContainEqual, toMatch, toMatchObject, toSatisfy, toBeCloseTo

Type/state matchers: toBeDefined, toBeUndefined, toBeNull, toBeTruthy, toBeFalsy, toBeNaN, toBeTypeOf, toBeInstanceOf

Comparison matchers: toBeGreaterThan, toBeGreaterThanOrEqual, toBeLessThan, toBeLessThanOrEqual

Collection matchers: toHaveLength, toHaveProperty

Error matcher: toThrow

Spy matchers (sinon-chai): toHaveBeenCalled, toHaveBeenCalledTimes, toHaveBeenCalledWith, toHaveBeenCalledExactlyOnceWith, toHaveBeenLastCalledWith, toHaveBeenNthCalledWith, toHaveReturned, toHaveReturnedTimes, toHaveReturnedWith, toHaveLastReturnedWith, toHaveNthReturnedWith

Negation: expect(x).not.toBe(y)

Promise matchers (legacy, avoid in new code): expect(promise).rejects.toThrow(), expect(promise).resolves.toBe(42). Prefer awaiting the promise first, then asserting on the result or caught error directly.

Asymmetric matchers (for use inside toHaveBeenCalledWith): expect.stringContaining, expect.stringMatching, expect.objectContaining, expect.arrayContaining, expect.anything, expect.any

Test output format

Every log line carries the full suite hierarchy as a tag chain. The leftmost tag is the outermost describe, the rightmost tag is the current it (or innermost describe); empty-name suites contribute no tag.

Default verbosity is compact: each describe emits one info-level line listing its fulfilled children's names. Per-test PASS lines are at debug (hidden unless DEBUG=true). Failures remain at error and are always visible.

[info]  [iso] [outer] [inner] PASS childA, childB, ... (Nms)   -- happy-path enumeration
[info]  [iso] [outer] [inner] PASS childA                       -- mixed-result: passing siblings only
[error] [iso] [outer] [inner] [bad] FAIL (Nms)                  -- per-test failure with full hierarchy
[error] [iso] [outer] [inner] FAIL (Nms)                        -- suite rollup with failure
[debug] [iso] [outer] [inner] [test-name] PASS (Nms)            -- per-test detail with DEBUG=true

Parent-children relationships are visible per line by default; no DEBUG flag needed. Levels:

• info -- per-suite PASS childA, childB (Nms) enumeration line and SKIP notices from it.
• error -- FAIL for tests and suites; always visible regardless of verbosity.
• debug -- per-test PASS (Nms), per-suite start (concurrency: N) traces, and the rollup line for empty-name (invisible) suites. Hidden by default; set DEBUG=true or pass --verbose to surface.

When a failure surfaces, read the rightmost (deepest) tag plus the FAIL line to identify the failing test; walk leftward through tags to follow the suite nesting back to the file's top-level describe. The full Error.cause chain mirrors the tag chain.

Linting test code

Testing intentional violations

When tests intentionally violate a lint rule to verify behavior:

// BAD: Adding data to satisfy the linter
expect(isError(new Error('test message',),),).toBe(true,);

// GOOD: Use disable comments for intentional violations
// oxlint-disable-next-line unicorn/error-message -- Testing error without message
expect(isError(new Error(),),).toBe(true,);

For known testing issues (duplicate describe blocks, missing test output, misattributed logs), see TROUBLESHOOTING.testing.md in the repository root.