This pull request has been marked as ready for review.

This pull request has been marked as ready for review.

@rvanvelzen
Cherry picked your tests from #1929 because the refactoring now includes your fix too.

I tried to break the logic in #1929 and #1934 and I did it. This example breaks in these PRs: https://phpstan.org/r/704ec47a-e5b0-4d5b-9bf6-037c4b563278

Relying on Variable being in there and invalidating those expressions is pretty smart, but there are expressions that don't contain a Variable and need to be invalidated too.

I guess that's what those failures in phpstan-webmozart-assert are about too.

Thank you for the example. I see, this logic isn't correct...
I'll have to investigate them further.

For the phpstan-webmozart-assert failure, I think rvanvelzen's explanation #1929 (comment) is right. (so it should fail as same as arrow function fails)

@ondrejmirtes
Do you think the logic of 93ef0b8 is correct enough?

I don't think that any changes around the Variables are correct. There's only one bug reported, and that's about function_exists(). So we need to retain this expression, and everything else needs to stay the same.

Also class_exists maybe.

I added the variable related specification logic in the migration of variableTypes and moreSpecificTypes
rajyan@3e20db3

to keep the logic before
https://github.com/rajyan/phpstan-src/blob/76be530e341d6452fbd4a910b8302438904caa55/src/Analyser/MutatingScope.php#L2910-L2922

but I think it's not correct enough
because of the #1934 (comment) case.

This pull request was first intended to fix that issue.

so, I think I can reset this pull request to 93ef0b8
and leave the phpstan/phpstan#8205 open

Made your example explicit that there are non variable expressions that should be invalidated.
https://3v4l.org/9bXe9

so, I think I can reset this pull request to 93ef0b8

Yeah, sure :)

@ondrejmirtes
Reset and added your comment as a test case 😄

Thank you!