[PHP] Fix "Controller is already closed" crash in CLI server

[adamziel]

Summary

This PR fixes two related Node-side crash paths in PHP-WASM that could bring down the Playground CLI server instead of failing a single request cleanly.

The first crash happened when PHP had already produced output, which closes the headers stream, and then execution failed afterwards. In that case, Node could still try to error() a ReadableStream controller that had already been closed, which threw:

TypeError [ERR_INVALID_STATE]: Invalid state: Controller is already closed

The validated repro for that path is:

1. PHP writes output
2. closeHeadersStream() closes the headers controller
3. a later WASM/runtime error occurs
4. error handling tries to signal the already-closed controller

A simple echo ...; exit; by itself is not the exact repro.

The second crash happened when a stream consumer cancelled the response stream before PHP finished, for example because the HTTP pipeline completed or the client disconnected. In that case, the runtime could still try to close() an already-cancelled controller and hit the same Controller is already closed failure.

The third crash path was in the proc_open() bridge used by the Node PHP runtime. If a spawned process timed out or exited while Node still had buffered stdout/stderr chunks to forward, the generated php.js bridge could try to write into a PIPEFS stream that PHP had already closed. That crashed the host with an error like:

TypeError: Cannot read properties of null (reading 'length')

Example:

<?php
$res = proc_open(
    "hanging_command",
    [
        ["pipe", "r"],
        ["pipe", "w"],
        ["pipe", "w"],
    ],
    $pipes
);
fread($pipes[1], 1024);

What changed

• Added defensive stream-controller handling in the universal runtime and stream bridge so already-closed controllers no longer crash the host process.
• Fixed child stdout/stderr teardown in the Emscripten Node bridge by detaching listeners on exit and ignoring late chunks after PHP has already closed the child-side pipe.
• Added off() to the shared EventEmitter interface so the child-process typing matches the APIs used by the runtime cleanup logic.
• Regenerated all checked-in Node PHP binaries (asyncify and jspi, PHP 7.4 through 8.5) from the updated build source.

Test plan

CI

[Copilot]

Pull request overview

Fixes a Node CLI server crash caused by attempting to operate on ReadableStreamDefaultControllers after the consumer has cancelled the stream (e.g., client disconnect), resulting in "Invalid state: Controller is already closed".

Changes:

• Add safeStreamError() / safeStreamClose() helpers and use them in PHP execution error/finally paths.
• Guard child-process stdout/stderr stream enqueues and errors against cancellation.
• Wrap portToStream() controller operations in a try/catch and add a regression test reproducing the cancellation-before-exit scenario.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
packages/php-wasm/universal/src/lib/php.ts	Adds safe controller close/error helpers and applies them to crash/finally and child process stream handling.
packages/php-wasm/universal/src/lib/api.ts	Wraps ReadableStream controller operations in portToStream() to avoid exceptions after cancellation.
packages/php-wasm/node/src/test/php-crash.spec.ts	Adds a regression test for cancellation before PHP exits.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This mock returns undefined for any ccall other than wasm_sapi_handle_request. That can make the test brittle (or accidentally change behavior) if runStream() relies on other ccall results during setup/teardown. Prefer delegating to the original ccall for all other c_func values (store the original before spying), and only override the one function needed for the scenario.