test(lib): add unit tests for src/lib/stellar.ts (closes #382)

[Vyacheslav-Tomashevskiy]

Closes #382

src/lib/stellar.ts had zero test coverage. This adds src/lib/__tests__/stellar.test.ts — 24 unit tests, no network access.

Approach

Horizon, Friendbot and the rate engine are mocked; Asset, Operation, TransactionBuilder and Keypair stay real, so buildSendTransaction assertions run against genuine XDR that is decoded back with TransactionBuilder.fromXDR() rather than against a stubbed builder. The module is re-imported per test (vi.resetModules()) because STELLAR_NETWORK / STELLAR_HORIZON_URL are read at module load.

What is covered

• Network config — testnet defaults, and STELLAR_NETWORK=public switching NETWORK_PASSPHRASE to Networks.PUBLIC.
• createTestnetAccount — the returned secret really is the secret for the returned public key (Keypair.fromSecret(...).publicKey()), the exact Friendbot URL, non-OK Friendbot response and a thrown fetch both swallowed (registration must not fail because Friendbot is down), and Friendbot not called on a non-testnet network.
• fetchRate — delegates to getRate() and returns only .rate; a rate-engine failure propagates instead of yielding a fabricated rate.
• getAccountBalances — native → XLM, credit assets → asset_code, Horizon 404 → [] (unfunded), and non-404 errors rethrown rather than being silently reported as an empty wallet.
• getNetworkStatus — field mapping plus the .order("desc").limit(1) query shape.
• buildSendTransaction — the 1% destMin slippage floor (123.4567891 → 122.2222212, 7dp), the 180s timebound, malformed recipient rejected before the Horizon round-trip (loadAccount asserted not called), insufficient balance, a missing trustline treated as a zero balance, a configured STELLAR_<CODE>_ISSUER resolving to the right Asset (with lower-case input, since resolveAsset upper-cases), and both issuer-misconfiguration errors (missing env var, malformed key).
• submitTransaction — success hash, Horizon result_codes surfaced as a JSON string instead of throwing, unknown_error fallback, and an unparseable XDR rejecting before any submission.

Verification

• npx vitest run src/lib/__tests__/stellar.test.ts → 24/24 pass.
• Negative control (mutation check) — these tests actually fail when the code regresses. Flipping * 0.99 to * 1.0 and deleting the status === 404 branch in stellar.ts turned 3 tests red; reverted afterwards, so stellar.ts is unchanged in this PR (test file only, +1 file).
• npx eslint src/lib/__tests__/stellar.test.ts → clean. npx tsc --noEmit → 0 errors in this file.
• Live server (npm run dev): /api/health reports "stellar": true, i.e. the real Horizon path in this module works against testnet, and /api/stellar/status answers normally.

Pre-existing failures, disclosed

npx vitest run on the full suite still fails to collect 4 unrelated test files, and /api/stellar/rate 500s on a live dev server, both with ReferenceError: isValidStellarPublicKey is not defined from src/lib/validations.ts:40. That reproduces on a clean origin/main and is the bug introduced by #527; the fix is waiting in the still-open #529. Nothing in this PR touches it. The 34 tests that do run all pass.

Scope note per the issue's contributor guidance: the issue also mentions "route-finding", but there is no route-finding logic inside src/lib/stellar.ts today (/api/stellar/routes builds its list elsewhere), so I did not invent tests for a function that does not exist.

[vercel]

Someone is attempting to deploy a commit to the codex723's projects Team on Vercel.

A member of the Team first needs to authorize it.