I disagree with this fix. class-string isn't always literal string.
Rather than that, $x::class should be an intersection with literal-string. You can implement that in MutatingScope::resolveType() and also verify that in NodeScopeResolverTest.

I disagree with this fix. class-string isn't always literal string.

@craigfrancis didn't give an example of class-string which are not literal-string in phpstan/phpstan-doctrine#332 (comment)

But I think we should follow the way it's handled by psalm
https://psalm.dev/r/16085bfd30

[I’m out atm, and using phone, will try to check properly later]

The source of the variable has to be a literal, so if you got it from example::class, that’s fine… but if you’re checking the value contains a string that happens to match a class name (e.g. ‘class_exists()’), that doesn’t flip the variable to being a literal.

That's what I thought, so this PR has to be fixed with my suggestion. Thank you.

thanks for the input guys. I just committed an update.

The spurious out of memory error could also be observed on other PRs. I don‘t think it is related to the actual change

adjusted per feedback.

I am thinking whether a object typed parameter on which ::class is appended should also be considered literal?

/**
 * @param object $o
 */
function a($o): void
{
	assertType('class-string&literal-string', $o::class);
	sayHello($o::class);
}

Of course :)

Please rebase and solve the conflict, I'll merge it then.

Thank you!