Flaky daemon test: KUBO_RPC_URL test still uses a one-shot fetch that races the pkc-js kubo restart (lineage #95)

[Rinse12]

Symptom

test/cli/daemon.test.ts › bitsocial daemon KUBO_RPC_URL env var › daemon uses KUBO_RPC_URL env var to configure kubo bind address failed on macOS CI:

TypeError: fetch failed  ->  connect ECONNREFUSED 127.0.0.1:<kuboPort>
 ❯ test/cli/daemon.test.ts:757:25

Ubuntu passed in the same run; the suite passes locally. Seen on the CI run for PR #132 (a pkc-js 0.0.82 -> 0.0.83 bump whose lockfile diff touches nothing but pkc-js itself), so this is not caused by that upgrade.

Root cause

Identical to issue #95. startPkcDaemon resolves as soon as the daemon prints its Communities in data path banner, which does not guarantee kubo's HTTP API is currently listening. Since pkc-js 0.0.46, pkc-js rewrites kubo's Routing config and POSTs /shutdown on first connect, so the daemon restarts kubo around init — a brief window where the kubo API port refuses connections.

The test verifies kubo with a single, un-retried fetch that can land in that window:

const res = await fetch(`${daemon.kuboApiUrl}/bitswap/stat`, { method: "POST" });
expect(res.status).toBe(200);

The #95 fix only patched the two tests in the kills kubo on its own shutdown describe. This is the last remaining bare one-shot kubo check in the suite — everything else polls waitForKuboReady.

Fix

Use the existing retry helper:

const kuboReady = await waitForKuboReady(daemon.kuboApiUrl, 45000);
expect(kuboReady).toBe(true);

Fixed on PR #132.