Skip to content

TESTING.md

Latest commit

 

History

History
79 lines (64 loc) · 3.28 KB

TESTING.md

File metadata and controls

79 lines (64 loc) · 3.28 KB
Raw

Testing

The murmuration app has an automated test suite built on Vitest. It covers the server API, the Postgres storage layer, the client API wrapper, eligibility helpers, the wallet gate, and the main UI flows.

Running

pnpm test            # run the whole suite once
pnpm test:watch      # watch mode
pnpm test:coverage   # run with a coverage report (text + HTML in ./coverage)

No database or wallet is required — Postgres, on-chain reads, IPFS, GitHub, and the wallet client are all mocked. EIP-712 signatures are produced with real viem test keys so signature verification is genuinely exercised.

Layout

tests/
  setup.ts              # global setup (silences logs, seeds a test admin)
  helpers.ts            # test accounts + EIP-712 signing matching the server
  server/               # Fastify API tests, run via app.inject()
    create / vote / read / options / commit-github / db / smoke
  client/               # jsdom component + lib tests
    votingApi / eligibility / wagmi / main / app*

Server tests mock server/db.mjs (in-memory) and the viem chain client. Client component tests mock wagmi, @rainbow-me/rainbowkit, and src/votingApi. A few app.jsx flows live in their own files because app.jsx keeps module-level state that leaks between sequential renders in one file.

Coverage (current)

Area Lines
server/api.mjs ~95%
server/db.mjs ~99%
src/votingApi.ts 100%
src/eligibility.ts 100%
src/wagmi.ts 100%
src/main.tsx ~80%
src/app.jsx ~52%
Overall ~67%

The business logic — vote casting (budget, eligibility, scheduling, admin bypass), proposal CRUD with EIP-712 admin auth, the tally/merkle math, the storage layer, and the wallet-gate role derivation — is covered at 95–100%.

The gap to 100% is almost entirely src/app.jsx, a single ~4,100-line auto-generated UI module. Its main flows (rounds list with the active / upcoming / past split, round detail + scheduling gate, the allocations panel, admin editor, the propose-a-direction screen + back button) are tested via F2App integration renders, but the long tail of UI branches (every admin form path, the slider drag interaction, the badge PFP modal, the verify panel, every error toast) is not.

Reaching ~100% there would mean either a large number of brittle interaction tests or — better — extracting app.jsx's pure helpers (date/phase utilities, QV cost math, coin/PFP mapping) and smaller components into separately-exported, unit-testable modules. That refactor is the recommended next step.

Testability changes

Two production modules were given small, behavior-preserving hooks so they can be imported by tests without side effects:

  • server/api.mjs exports app and only calls app.listen() when run directly (node server/api.mjs); the admin allowlist accepts EXTRA_ADMIN_ADDRESSES (mirroring the client's VITE_EXTRA_ADMIN_ADDRESSES) and the staging vote-bypass flag is read per-request.
  • src/main.tsx exports WalletGate / useCurrentPath / WalletLockedHint / AppErrorBoundary and only mounts the React root in a browser (not under the test runner).