plugin-calendar: object-calendar-renderer.propsContract.test.tsx escapes to the real network on 127.0.0.1:3000 (4 ECONNREFUSED, same class as #3339 / #5225 / #5280) #5438

Found while working #5421 — this is the stderr noise that the coverage-lane logs show and that a reader could mistake for the shard failure. It is not a failure: the run stays green. Filed unassigned.

Measured

From the repo root of a clean worktree on main (9f7d786bd), the file standing alone:

pnpm exec vitest run packages/plugin-calendar/src/object-calendar-renderer.propsContract.test.tsx

emits exactly four of these and then passes:

Error: connect ECONNREFUSED 127.0.0.1:3000
    at TCPConnectWrap.afterConnect [as oncomplete] (node:net:1637:16) {
  errno: -111,
  code: 'ECONNREFUSED',
  syscall: 'connect',
  address: '127.0.0.1',
  port: 3000
}

Attribution was confirmed twice: a single-worker sequential run of a 27-file subset (--maxWorkers=1 --fileParallelism=false) prints the four errors in the middle of this file's own case list, between the locale-rejection cases and the onEventClick cases; and the file alone reproduces all four.

Coverage is irrelevant to it — the same four appear with and without --coverage, so the green PR lane carries this noise today as well. On CI the same four appear as AggregateError with both ::1:3000 and 127.0.0.1:3000 legs, because the runner has IPv6.

Mechanism

happy-dom's default document origin is http://localhost:3000/, so any relative-URL fetch that nothing intercepts resolves against that origin and leaves the process for real. No test file in the shard stubs fetch for this path — the request comes from the component graph under test, on the branch where the node does not carry the declared data pre-fetch array and the component therefore runs its own fetch.

Same defect class as #3339 (plugin-detail), #5225 (plugin-report) and #5280 (plugin-dashboard DatasetWidget), all closed. This is a new member of it, not a regression of those.

Why it is worth closing

• It is a real, unhandled outbound connection from a unit test: on any machine where something is listening on port 3000, the assertion outcome depends on what that server answers.
• It makes every coverage-lane log carry four scary-looking network stack traces that have nothing to do with the run's verdict — which is precisely how The suite is still red under v8 coverage on main after #5402 — shard 3/4 fails deterministically and the unsharded run leaves no coverage/ at all #5421 came to be filed against the wrong mechanism.

Fix shape follows the closed siblings: give the component's own fetch path a stub (or author the data pre-fetch array on the nodes that do not need to exercise the fetching branch), so nothing in the file reaches the network.