By name: unit-tests description: | Guardrails for adding unit tests in bklit-ui without over-testing. Use when the user mentions unit test, unit tests, tests, test coverage, add tests, write tests, vitest, jest, or asks whether something should be tested.
Unit Tests (bklit-ui)
Read this skill before proposing or writing tests. Default stance: fewer, higher-signal tests.
When to add tests
Add tests when they lock in behavior that is easy to break silently:
| Worth testing | Why |
|---|---|
| Pure functions / modules | Stable inputs → outputs; fast; no DOM |
| Formatters, parsers, scale math, bounds | Regression on string/number output is user-visible |
| Codegen / export / registry helpers | Output shape is the contract |
| Non-obvious edge cases | Empty data, reversed ranges, clamping |
Examples in this repo: chart-formatters.test.ts, highlight-segment-bounds.test.ts, animation.test.ts, apps/web/lib/studio/__tests__/*.
When NOT to add tests (unless explicitly requested)
Do not add tests just to increase coverage or “be thorough”:
| Skip | Why |
|---|---|
| React component render smoke tests | visx, motion, portals → brittle; manual/docs check is cheaper |
memo() / guard extractions (#65-style) |
Structural perf refactors; output unchanged; RTL mount assertions are heavy |
| Default prop passthrough | formatValue = intFmt — types + formatter tests cover it |
| Third-party library behavior | Don’t re-test d3, visx, or React |
| Trivial getters / one-line wrappers | No regression signal |
| Snapshot entire chart SVG/JSX | High churn, low signal |
If the user asks “should we test X?” — say no when X falls in this table, and suggest a lighter alternative (pure helper test, manual check, CI build).
Repo conventions
Runner: Node built-in test runner + tsx (not Jest/Vitest unless the repo adopts them later).
pnpm test # root — turbo runs packages with a test script
pnpm --filter @bklitui/ui test
cd apps/web && pnpm test
Place tests: **/__tests__/**/*.test.ts next to the code under test.
Pattern:
import assert from "node:assert/strict";
import { describe, it } from "node:test";
import { myFn } from "../my-module";
describe("myFn", () => {
it("handles empty input", () => {
assert.equal(myFn([]), expected);
});
});
Equivalence tests (preferred for formatters): assert shared module output matches the previous inline call (e.g. toLocaleDateString with same locale/options) so tests stay timezone-safe and prove no visual regression.
New package test script: add to package.json:
"test": "node scripts/run-tests.mjs"
Use a small scripts/run-tests.mjs that collects *.test.ts from __tests__ and invokes node --import tsx --test — shell globs break on Linux CI. Add tsx as a devDependency if missing. Wire into root turbo.json test task; CI runs pnpm test.
Decision checklist (run before writing)
- Is the logic pure or extractable to pure functions? → Test that. Consider extracting first.
- Would a test fail on a real user-visible bug? → Good candidate.
- Does it need jsdom / RTL / Playwright? → Stop; justify to user or defer to manual/visual check.
- Did the user ask for tests? → Still apply this skill; don’t over-deliver.
- How many cases? → 3–10 focused cases, not exhaustive matrices.
What to tell the user
When recommending tests, be explicit:
- Add: “Test
chart-formatters.ts— pure, high regression value.” - Skip: “Skip component tests for the memo split — no output change; build + manual chart docs are enough.”
- CI: Mention
pnpm testin PR test plan when adding or changing tests.
Anti-patterns
- Adding Jest/Vitest/Testing Library for a one-off without team buy-in
- Mocking entire chart context to assert
useMemocall counts - Testing implementation details (hook order, internal component names)
- Duplicating type-checker work (
expect(typeof x).toBe('function')) - Committing tests that only pass locally due to hard-coded timezone/locale strings