Writing tests for pota

Tests run in a real Chromium browser via the custom Puppeteer runner (architecture and the full #test helper contract: tools/test-runner/readme.md). Conventions, timing rules, and the coverage inventory live in tests/readme.md. This skill is the procedure for adding coverage.

Procedure

1. Read the source first. Find the unit's real signature and behavior in src/ — never test from memory of a similar API.
2. Place the file mirroring src/: tests/api/components/ for built-in components, reactivity/ for primitives, dom/ for renderer/props/events, store/, use/ (one file per pota/use/* module), forms/ for native form behavior, jsx/ for transform and tracking semantics. Type-level assertions go in tests/typescript/*.tsx instead (typecheck-only, npm run test:ts-tests).
3. Write the tests (anatomy below), one file at a time. Cover the behaviors the source actually shows, including edge cases.
4. Run filtered, then full: npm run test:api -- <path-substring>, then npm run test:api.
5. Update the inventory — the matching table row in tests/readme.md.
6. If a test you wrote fails, don't delete, weaken, or rewrite it to pass. Surface the failing assertion, the observed behavior, and a hypothesis — it may be a real library bug (see Workflow in documentation/AGENTS.md). To demonstrate an alternative expectation, add a new test alongside the existing one.

Anatomy

/** @jsxImportSource pota */
// Tests for the Show component: static truthy/falsy, signal
// toggling, fallback.
import { test, body, microtask } from '#test'

import { render, signal } from 'pota'
import { Show } from 'pota/components'

await test('Show - renders children when `when` is true', expect => {
	const dispose = render(
		<Show when={true}>
			<p>hello</p>
		</Show>,
	)
	expect(body()).toBe('<p>hello</p>')

	dispose()
})

await test('Show - signal: toggles from true to false', expect => {
	const visible = signal(true)
	const dispose = render(
		<Show
			when={visible.read}
			fallback={<p>fallback</p>}
		>
			<p>hello</p>
		</Show>,
	)
	expect(body()).toBe('<p>hello</p>')
	visible.write(false)
	expect(body()).toBe('<p>fallback</p>')

	dispose()
})

• Files start with the @jsxImportSource pota pragma and a 1–3 line comment saying what the file covers.
• Tests are top-level await test(title, fn) calls, run sequentially in file order; the harness reports automatically — no run() call, no registration step, no build step before running.
• fn receives expect (toBe, toEqual, not.*). Prefer exact output assertions (expect(body()).toBe('…')) when browser serialization is stable.
• Call the dispose returned by render at the end of the test. The harness asserts a clean document around every test, so a leaked node fails the next test — this doubles as a disposal-leak check.
• No try/catch around assertions, no helper mini-frameworks — inline the setup.

Timing

Signal writes update the DOM synchronously — assert right after the write, no await needed (see the toggle test above). When an await is needed, prefer microtask() over macrotask() over sleep(ms) — the ladder and the cases each covers are in tests/readme.md (Timing Considerations). The two common cases: promise/async content needs await microtask() after the promise settles; a use:* plugin registered with the default onMicrotask=true needs await microtask() after render().

Running

npm run test:api -- <substring> filters by path; --bail stops on the first failure; npm run watch:test re-runs on change; npm run coverage adds a c8 report under generated/coverage/. The full pre-release gate is npm test (types + browser + babel-preset).