Fix ImpossibleCheckTypeFunctionCallRule for is_subclass_of and is_a

[VincentLanglet]

Closes phpstan/phpstan#3979
which can be simplified to https://phpstan.org/r/ef72a996-8fbe-4a4e-831f-4d78c542ad19

I'm not sure there is a better (easy ?) fix for this issue. is_subclass_of($string, $bar) specify the type of $string to class-string when true, but it doesn't mean that the call is useless if the param is already a class-string.

I discovered there was an exclusion of is_a in the ImpossibleCheckTypeFunctionCallRule (maybe for the same reason)
so I think it could be the same for is_subclass_of since the behavior is similar.

[ondrejmirtes]

Changing the rule is not going to help you against "Result of && is always false.".

I'd much rather change the logic in IsSubclassOfFunctionTypeSpecifyingExtension.

[VincentLanglet]

if (is_subclass_of($object, $stringOrNull, false))

specified $object as object but this was already known so it's not a big loss.

[phpstan-bot]

This pull request has been marked as ready for review.

[VincentLanglet]

I'd much rather change the logic in IsSubclassOfFunctionTypeSpecifyingExtension.

I think it was possible indeed, I get something working.

[VincentLanglet]

Friendly ping @ondrejmirtes ; this PR is ready to be re-reviewed :)

[ondrejmirtes]

Thank you.

[ondrejmirtes]

Unfortunately I have to revert this. it leads to false positives like: https://phpstan.org/r/5941cd95-26d7-49cf-a2e0-ffc21c4a7697

[VincentLanglet]

Unfortunately I have to revert this. it leads to false positives like: phpstan.org/r/5941cd95-26d7-49cf-a2e0-ffc21c4a7697

No problem, I'll try to improve the idea in #3788
Did you had another false positive I have to add as non-regression test ?