Kévin Dunglas

Actually, merging the $_SERVER superglobals of the worker and the request wasn't expected, we even have a test (the failing one) for that. As showcased by the test, it's entirely...

I've an idea for the `after:`. I'll take a look. The crash and the leak are weird, I'm not sure of why it behaves like this.

Maybe related: https://github.com/dunglas/frankenphp/pull/81

@TimWolla @withinboredom would you mind testing/reviewing this patch one more time? The current implementation looks ok-ish to me.

In the last commit, I fixed the crash without changing the existing behavior (the docs change is now irrelevant, I'll revert that). I think it's good enough for now.

I cleaned the code and updated the docs. I think I've found a way to keep the userland and internal structures in sync but I didn't create a test case....

@TimWolla are the two data structures synced automatically in a normal script? I may be wrong, but I don't think that we do anything special FrankenPHP-side that would cause this...

@TimWolla according to my tests, this behaves the same even when not in worker mode, so it looks "expected" to me, at least let's say that it's not related to...

Yes, for instance Laravel Sail runs FrankenPHP using supervisord: https://laravel.com/docs/10.x/octane#frankenphp-via-laravel-sail

Definitely!