Support for class-string $class parameter in is_subclass_of()

[arnaud-lb]

This adds support for $class parameter of type class-string in is_subclass_of().

I tried to unify the logic for the cases when $class is a constant string, a class string, or something else.

I also tried to narrow the specified type when we know that the $object parameter is either a string or an object. For example if the parameter is a string, it will no longer be specified as class-string|object.

There is one caveat: the type specifier extension may now specify the $object parameter as *NEVER* in some impossible cases. Is this acceptable ?

This fixes phpstan/phpstan#6698

[staabm]

There is one caveat: the type specifier extension may now specify the $object parameter as NEVER in some impossible cases. Is this acceptable ?

when does this happen? Could it use $overwrite=true In this cases?

This would work, when the *NEVER* type is generated because of type validation in

phpstan-src/src/Rules/Comparison/ImpossibleCheckTypeHelper.php

Line 163 in dbbb721

if ($specifiedTypes->shouldOverwrite()) {

[herndlm]

Hi, I found some inconsistencies with falsey contexts and is_a that I wanted to look into. Or my constant type tests are just odd.
Anyways, since is_a and is_subclass_of are extremely similar (https://3v4l.org/WDt9u) and both have a specifying extension I wanted to mention this. Maybe we can duplicate some logic there or even re-use it and fix bugs for both :)

UPDATE: phpstan/phpstan#6704 is what I was referring to.

[arnaud-lb]

@staabm this happens when the $object param is a string, and the $allow_string param is false, or if $object is not a super type of object|string. Both of these cases are not possible in a truthy context and IsSubclassOfFunctionTypeSpecifyingExtension will specify $object to *NEVER* because it intersects between the original $object type and the inferred type.

I took more time to have a better understanding of TypeSpecifyingExtension / TypeSpecifier, and I've found that the intersect is not needed as it's already done by TypeSpecifier::create(). So I've just removed the intersect.

[ondrejmirtes]

Thank you!