PHPStan wrongly assumes that $_SERVER cannot be null

[sebastianbergmann]

Bug report

In PHPUnit, I have this line of code:

$env = $_SERVER ?? [];

PHPStan wrongfully complains about this line:

Variable $_SERVER on left side of ?? always exists and is not nullable.

However, this line of code exists for a reason: some people have test code or test code that sets $_SERVER to null. See https://3v4l.org/t1PTW for proof that this is possible.

Code snippet that reproduces the problem

https://phpstan.org/r/fa40be63-50df-4254-86e8-d02f26654e49

Expected output

No error.

Did PHPStan help you today? Did it make you happy in any way?

Yes!

[ondrejmirtes]

This is a hard issue to tackle correctly and practically. Apparently you can set these global variables to anything: https://3v4l.org/8H9ap

So ideally we should make them just mixed because it's possible they might be anything.

But if you're in a codebase that doesn't do anything weird like this, they're going to be array<string, mixed>. So complaining that you're passing mixed into array after such a fix would be annoying for most people.

We could also suppress some errors reported to these global variables when rechecking their type, but it's hard to do such a thing consistently without changing type of these variables.

I'm going to keep this open but I'd say the best course of action is to ignore this kind of error - "hey, I know what I'm doing, I'm going to ignore this here".

[sebastianbergmann]

"hey, I know what I'm doing, I'm going to ignore this here"

Okay :)