[chubes4]

Related issues

• CLI E2E on Linux failed twice in the last 200 builds (23113 and 23272) with Failed to listen on 127.0.0.1:44047 (reason: Address already in use), followed by PHP worker 3/4 exited unexpectedly and Failed to start WordPress server.

How AI was used in this PR

Claude (via the Pi coding agent in a Roadie session) found the failure in the CI logs, traced it to the worker port allocation, and wrote the change and the test. I reviewed the diff and the test results.

Proposed Changes

A native PHP site runs a pool of 4 php -S workers. php-server-child.ts picked their ports one at a time: bind a probe server to port 0, read the port, close the probe, then pick the next. Once a probe closes, its port is free again, so a later probe can be handed the same port. Linux picks ephemeral ports at random, so this happens rarely: about 1 in 1,000 site starts, measured on a Linux machine. When it does, two workers get the same port, the second fails to bind it, and the site does not start.

• getAvailablePorts( count ) in a new lib/native-php/worker-ports.ts keeps every probe bound until all ports are picked, then releases them together. Ports picked this way cannot repeat.
• php-server-child.ts uses it in place of the one-at-a-time loop.

Ports are still released before php -S binds them, so another process could in principle take one in that gap. That race is unchanged, and it is much rarer than one pool reusing its own port.

Testing Instructions

• npx vitest run --project cli lib/native-php/tests/worker-ports.test.ts. The test fakes an OS that always hands out the lowest free port. Trunk's loop gets the same port 4 times there (expected 1 to be 4); the new function returns 4 distinct ports and releases them all.
• Real kernel, outside the test: picking 4 ports the way trunk does repeated a port in 98 of 100,000 pools on Linux (0 of 2,000 on macOS). With this change a pool can't repeat a port.
• npm run cli:build && npm test -- --project cli --tagsFilter=e2e --no-file-parallelism .e2e.test.ts: 42 passed locally.

Pre-merge Checklist

• Have you checked for TypeScript, React or other console errors?

[wpmobilebot]

📊 Performance Test Results

Comparing b3d8eec vs trunk

app-size

Metric	trunk	b3d8eec	Diff	Change
App Size (Mac)	954.28 MB	954.28 MB	0.00 MB	⚪ 0.0%

site-editor

Metric	trunk	b3d8eec	Diff	Change
load	1255 ms	1201 ms	54 ms	🟢 -4.3%

site-startup

Metric	trunk	b3d8eec	Diff	Change
siteCreation	5840 ms	5606 ms	234 ms	🟢 -4.0%
siteStartup	2858 ms	2888 ms	+30 ms	⚪ 0.0%

---

Results are median values from multiple test runs.

Legend: 🟢 Improvement (faster) | 🔴 Regression (slower) | ⚪ No change (<50ms diff)