fix(cli): bind-probe 0.0.0.0 and loopback so macOS sees a busy port

[peterbussch]

Summary

On macOS, probePortFree() says a port is free while another OmniRoute holds it. The probe binds the wildcard address, and macOS allows that bind while another socket listens on the same port at 0.0.0.0 (the default serve host), 127.0.0.1 or ::1. Linux rejects it, so CI stays green, but two tests in tests/unit/cli-serve-port-in-use-preflight.test.mjs fail on a Mac.

To reproduce (macOS 27.0, Apple M4 Pro, Node 22.23.0, release/v3.8.51 at a58000c768):

$ node -e '
const net = require("node:net");
const s = net.createServer().listen(0, "0.0.0.0", async () => {
  const { probePortFree } = await import("./bin/cli/utils/pid.mjs");
  console.log("held on 0.0.0.0, probePortFree:", await probePortFree(s.address().port));
  s.close();
});'
held on 0.0.0.0, probePortFree: true

Cause: bin/cli/utils/pid.mjs:107-119 binds only the wildcard address. On macOS only a bind to the held address itself fails with EADDRINUSE.

Fix:

• Keep the wildcard bind, then also bind 0.0.0.0, 127.0.0.1 and ::1. Any EADDRINUSE means busy.
• Every other error still reads as free (for example EADDRNOTAVAIL on a host without ::1), so the probe still never blocks a start it cannot observe (fix(cli): port-in-use guard silently passes when lsof is unavailable (Termux/slim containers) — user gets an EADDRINUSE restart loop instead of the clear message #14518).

Related Issues

• Related to fix(startup): serve crashes on macOS with a free port — "Cannot read properties of null (reading 'length')" (regression from #14538) #14800
• Related to fix(cli): prevent serve crash when port is free on macOS #14812

Note: #14812 fixes the lsof crash in #14800 and does not touch probePortFree. This is the remaining macOS part. The two branches merge cleanly in either order, and the combined test file passes 15/15.

Validation

• Change type: CLI
• Focused tests: node --import tsx/esm --test tests/unit/cli-serve-port-in-use-preflight.test.mjs: 13 passed on macOS. Before the fix: 10 passed, 3 failed (the new test and the two existing macOS failures).
• npm run lint: I ran eslint and prettier --check on the changed files only. Both are clean.
• Reconciled with the current active release base (release/v3.8.51 at a58000c768)
• Production-code changes include a new or updated automated test in this PR

Tests Added Or Updated

• tests/unit/cli-serve-port-in-use-preflight.test.mjs
  • probePortFree sees a held port when the wildcard bind succeeds (macOS) stubs deps.net with macOS bind semantics, so without the fix it fails on Linux CI too.
  • probePortFree treats a missing loopback address as free covers the fail-open path.

Coverage Notes

The two new tests reach every branch of the new loop in probePortFree without opening a socket.

Reviewer Notes

• Linux behavior does not change. The first probe is still the wildcard bind, and it already fails there when the port is held.
• A free port now costs four binds instead of one.
• The probe now counts a listener on any of these addresses as busy, even if serve would bind a different one. findListeningPids() already works that way: lsof -ti :PORT ignores the address.

⚠️ base-red inherited: #14866

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.