use conditional parameter for is_a() via stub

[staabm]

closes phpstan/phpstan#4371
closes #440

see https://www.php.net/is_a

[staabm]

stub taken 1:1 from psalm

[staabm]

update: psalm was missing the type for $allow_string, which will be added with vimeo/psalm#7951

[ondrejmirtes]

Please read the title of the linked issue: is_a: "if third argument is false then do not allow string in first argument"

Which means that there should be a new test in CallToFunctionParametersRuleTest (or ImpossibleCheckTypeFunctionCallRuleTest, that depends) that reports a new error here https://phpstan.org/r/44b40ab5-7c7f-4ced-8048-e084d3d63f1d on line 10.

[staabm]

yeah, it seems the initially reported bug in 4371 is different from the one I had in phpstan/phpstan#4371 (comment)

but I am fine, with also covering the original issue here then

[staabm]

is no longer considered "fine" because Foo is not a subclass of Throwable

[ondrejmirtes]

But also please test Foo::class (which is different thanks to namespace resolution), the class isn't final so a subclass might implement Throwable.

[staabm]

hm I guess you mean I should test class-string<Foo> as a parameter not Foo::class.. at least thats what I did for now.. see added tests at the end of this file

[staabm]

this gets a few rough edges, therefore I pushed the current development state and hope for feedback.

[staabm]

hm I guess you mean I should test class-string<Foo> as a parameter not Foo::class.. at least thats what I did for now.. see added tests at the end of this file

[staabm]

with 3e955af I dropped the conditional return type stub again, as I am no longer sure its worth to report about string vs. class-string type errors for $class.. might be tedious.

since we already have extensions for is_a in place the stub itself was nevertheless only partly usefull, I think.

[staabm]

because I dropped this early-return, I had to compensate a few false positives in the IsAFunctionTypeSpecifyingExtension - added a few early returns.

[staabm]

especially this condition is pretty ugly .. would love feedback on how to improve it.

at first I only had the $classType instanceof ClassStringType part, but later on needed to add !$classType instanceof GenericClassStringType to prevent regressions of existing tests

[ondrejmirtes]

This is already quite messy again :/ I think the stub solved a problem so I really don't want to have it reverted here 3e955af.

But this is truly a minor problem. I'd rather be if you worked on the NamedArgumentsHelper improvements :) Thanks.

[staabm]

I started workin on this topic, because we had a production issue, which would be great if beeing covered by phpstan the next time.

it looked easy at first, but it seems I have no luck in picking topics to work on, since most of the time it turns into walking thru rabbit holes, like we can see here :-/.

will do NamedArguments stuff, when free time allows again.

[ondrejmirtes]

You can just send the conditional stub in a separate PR alongside with the tests as you initially did, without marking phpstan/phpstan#4371 as closing :)

[ondrejmirtes]

Basically this commit 6bbb2a6 :)

[ondrejmirtes]

I have no luck in picking topics to work on

That's normal, I just spent several hours today trying to clean up something about named arguments and I'm not happy with the result either. The combination of php-8-stubs, phpstorm-stubs and php-8-stubs is a mess in some situations. And PHP itself is a mess too :)