Fix E2E tests on macOS when STUDIO_RUNTIME=native-php

[fredrikekelund]

Related issues

• Fixes RSM-1725

How AI was used in this PR

Primarily to help me debug the problem. I alternated between Claude and Codex.

Proposed Changes

While working on #3271, @bcotrim discovered that the E2E tests were failing when run with STUDIO_RUNTIME=native-php. This PR fixes that by implementing the following fixes:

• When Studio clones a site, explicitly update the site URL. Playground can handle this internally, but the native PHP runtime obviously cannot.
• On Posix platforms, spawn process daemon children with detached: true to assign them to a new process group. This is so we can pass a negative PID to process.kill to terminate the entire process group (the Node.js process and the associated PHP process). This approach isn't bulletproof, but it's alright, and it works for the purpose of the E2E tests.
• Favor SIGTERM over SIGKILL when killing child processes to give them a chance to clean up.

Testing Instructions

Run the E2E tests locally on macOS by following these steps:

1. npm run package
2. STUDIO_RUNTIME=native-php npm run e2e
3. Ensure that they pass successfully

Pre-merge Checklist

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

[wpmobilebot]

📊 Performance Test Results

Comparing e82c8a0 vs trunk

app-size

Metric	trunk	e82c8a0	Diff	Change
App Size (Mac)	1667.60 MB	1652.64 MB	14.95 MB	🟢 -0.9%

site-editor

Metric	trunk	e82c8a0	Diff	Change
load	1499 ms	1494 ms	5 ms	⚪ 0.0%

site-startup

Metric	trunk	e82c8a0	Diff	Change
siteCreation	8079 ms	8109 ms	+30 ms	⚪ 0.0%
siteStartup	4935 ms	4953 ms	+18 ms	⚪ 0.0%

---

Results are median values from multiple test runs.

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

[bcotrim]

I had trouble running the e2e tests locally.
First issue was checksum mismatches — looks like upstream static-php-cli re-publishes archives, so the pinned hashes go stale. Should be moot once #3297 lands and we build binaries ourselves.

Second issue: even with hashes fixed, e2e teardown hung for 3 minutes with one PHP child orphaned per session. Root cause is that php-server-child.ts only handles exit/disconnect, not SIGTERM. When the daemon SIGTERMs the Node fork, it dies without firing exit, so killPhpProcess() never runs and the PHP grandchild leaks. Adding these unblocks teardown for me:

process.on( 'SIGTERM', () => {
	killPhpProcess();
	process.exit( 0 );
} );
process.on( 'SIGINT', () => {
	killPhpProcess();
	process.exit( 0 );
} );

Could you reproduce + confirm the fix on your end?

[bcotrim]

Should we run this only for native-php?

[fredrikekelund]

It's obviously not required for Playground, but that's only because Playground enforces the siteurl internally. I think it makes sense to follow the more standard approach of explicitly updating the siteurl rather than relying on some runtimes to do it for us.

[fredrikekelund]

Sorry about that, @bcotrim. I did confirm this working previously… It seems you're correct that process.on( 'exit' ) isn't called in response to SIGINT or SIGTERM signals, so I'm not sure what made the cleanup work before. I may have broken it by messing up a git merge.

In any case, I've updated php-server-child to explicitly handle SIGINT or SIGTERM signals. I've also made it so process-manager-daemon sends SIGTERM signals to the entire process group (previously we SIGTERM only to the Node.js child process and followed up with SIGKILL signals to the group if needed).

I've also updated the PHP binary checksums for macOS with ARM chips to make the tests pass.

[fredrikekelund]

I'm playing around with Windows changes here, too. I could move those commits to a new PR, though.

@bcotrim, if you review this again, feel free to ignore the changes explicitly targeting Windows. I'll let you know when those are ready for review.

[bcotrim]

Thanks for following up on the changes @fredrikekelund
e2e now run successfully on macOS.

I would suggest moving the Windows to another PR so we can land this one.

Nit: A nice workflow would be to control the STUDIO_RUNTIME via a PR label (for example) so the CI runs native PHP e2e. But we can do it in a follow-up PR, and maybe only after we have the binaries in our own CDN.

[fredrikekelund]

A nice workflow would be to control the STUDIO_RUNTIME via a PR label (for example) so the CI runs native PHP e2e

Good idea 👍 I added this in #3343