Avoid false error on is_subclass_of

[VincentLanglet]

Closes phpstan/phpstan#13713

Solves phpstan/phpstan#6305 (comment)

[phpstan-bot]

This pull request has been marked as ready for review.

[VincentLanglet]

This issue was more complicated than expected... I'm not against your opinion/review @staabm :)

[VincentLanglet]

I dunno what to do about the webmozart-assert failure because I don't report anymore

Call to static method Webmozart\Assert\Assert::implementsInterface() with class-string<WebmozartAssertImpossibleCheck\Bar>|WebmozartAssertImpossibleCheck\Bar and 'WebmozartAssertImpossibleCheck\\Bar' will always evaluate to true.

Technically, because a implementsInterface check is called before it's indeed always true.
But strictly looking at the type, calling

class-string<WebmozartAssertImpossibleCheck\Bar>|WebmozartAssertImpossibleCheck\Bar

And currently, phpstan/phpstan-webmozart-assert#194 is showing the issue

[ondrejmirtes]

This needs to be fixed in phpstan-webmozart-assert before we can merge this: https://github.com/phpstan/phpstan-webmozart-assert/blob/b2221d38ee5338531c0bcaa985268ae5bcef452e/src/Type/WebMozartAssert/AssertTypeSpecifyingExtension.php#L401-L419

[VincentLanglet]

This needs to be fixed in phpstan-webmozart-assert before we can merge this: phpstan/phpstan-webmozart-assert@b2221d3/src/Type/WebMozartAssert/AssertTypeSpecifyingExtension.php#L401-L419

It's been a long time I worked on this PR, but the implementation looks fine to me.
PHPStan will still report most of always-true, always-false implementsInterface calls.

The only issue is the fact it cannot detect

if (\is_subclass_of($object, \stdClass::class, false)) {
      // $object is class-string<\stdClass>|\stdClass
		if (\is_subclass_of($object, \stdClass::class, false)) { // This is not detected as always true
		}
	}

Because I dunno currently a way to say it cannot be the class exactly but only a child or for the call to be remembered.
Maybe you know @staabm ?

And so far I thought it was better to have a missing "always true" report rather than a wrong one.

[staabm]

Because I dunno currently a way to say it cannot be the class exactly but only a child or for the call to be remembered.
Maybe you know @staabm ?

just a guess:

$a->isSuperTypeOf($b)->yes() && $b->isSuperTypeOf($a)->yes() && !$a->equals($b)

or maybe isSubTypeOf()

And so far I thought it was better to have a missing "always true" report rather than a wrong one.

absolutely

[VincentLanglet]

Because I dunno currently a way to say it cannot be the class exactly but only a child or for the call to be remembered.
Maybe you know @staabm ?

just a guess:

$a->isSuperTypeOf($b)->yes() && $b->isSuperTypeOf($a)->yes() && !$a->equals($b)

or maybe isSubTypeOf()

It's more with the type instances after

if (\is_subclass_of($object, \stdClass::class, false)) {

how to specify better than class-string<\stdClass>|\stdClass

I feel like we cannot do this without a ChildOfObjectType() or a boolean in ObjectType to require child only and then we specify to class-string<child-of<\stdClass>>|child-of<\stdClass>

So right now @ondrejmirtes, I feel like the best we can is

• Avoid false error on is_subclass_of where there is uncertainty
• Lose some "always true" error reported

And a next step could be to introduce a childOnly bool in ObjectType constructor (?)