This pull request has been marked as ready for review.

Thank you!

oh cool, this seems to be a simpler version of #3762 which I can close now :)

Ohh.. this wasn't intentional 😅

that's ok, this solution is simpler / more elegant :)

Hi, I remember there was a weird issue when I tried to use my prepared superglobals tests here: https://github.com/phpstan/phpstan-src/pull/3762/files#diff-51a54cb478a71bc6cb00a1276e9df23606258635fe80bfff8843430cc254505c

basically, the truthy/falsey scope bled into the null one. I should have debugged this more.

I believe that is related to this change here causing issues like phpstan/phpstan#12771 and the ones linked to it. @staabm are you available and want to / can look into this? If not I can try later too

hmm, looks like the issue is fixed when I set the expression directly as in https://github.com/phpstan/phpstan-src/pull/3762/files#diff-dc817f2bab8672057a95375a542e68599164f6ab2380e4cccd1b50583e295d85R562-R564, which is kind of the only core difference to this PR.

but I'm not a 100% sure how it even worked before tbh. and if I propose the expression fix, then I could also propose the rest I'm doing in that MR (taking over the expressions when entering functions, classes, namespace)? not sure. an opinion from @ondrejmirtes would be interesting before I prepare anything :)

This introduced a regression: https://phpstan.org/r/86811bd5-067b-476e-958a-bf621ad773c5

Please look into it, thanks.

This introduced a regression: phpstan.org/r/86811bd5-067b-476e-958a-bf621ad773c5

for posterity: fixed via #3901