Native type does not know anything about purity

[VincentLanglet]

Closes phpstan/phpstan#12458

[ondrejmirtes]

I disagree with this on a conceptual level. With a closure like fn () => sleep(5), only native types are involved and we still know it's impure.

With one-off fixes like that, we'd be chaising our own tail for a long time (finding more stuff to fix all the time, which is frustrating). I propose a different fix. In https://github.com/phpstan/phpstan-src/blob/2.1.x/src/Rules/PhpDoc/VarTagTypeRuleHelper.php there are about 9 isSuperTypeOf calls.

1. Instead of calling them directly on a type, introduce a private method that accepts $type, $varTagType and calls isSuperTypeOf inside. That way we only need to change a single place.
2. After calling this isSuperTypeOf, in case an error would be reported, run one more check. Take the $type (the native type of the expression), do Type::toPhpDocNode() and then create a Type again with TypeNodeResolver. This will "clean up" everything that's not expressible with PHPDocs.
3. Call isSuperTypeOf again with the new $type that's the result of operation from 2).

[VincentLanglet]

2. After calling this isSuperTypeOf, in case an error would be reported, run one more check. Take the $type (the native type of the expression), do Type::toPhpDocNode() and then create a Type again with TypeNodeResolver. This will "clean up" everything that's not expressible with PHPDocs.

TypeNodeResolver::resolve needs a NameScope. How do I get one here ?

[ondrejmirtes]

I think null is fine? Or an empty one. We won't be resolving any relative names, I hope.

[ondrejmirtes]

Anyway, this is just a guess what could work, maybe it's still going to break for some cases.

[VincentLanglet]

I think null is fine? Or an empty one. We won't be resolving any relative names, I hope.

Indeed.

I saw there was 5 occurences of new NameScope(null, []) do you think it could be useful to provide a
NameScope::createEmpty() method ?

[VincentLanglet]

After calling this isSuperTypeOf, in case an error would be reported, run one more check.

Seems like if I do this every time it works. I would say it's avoid two calls to isSuperTypeOf (?)

Also since sometimes checks are

!$this->isSuperTypeOfVarType($type, $varTagType)->yes()

and sometimes it's

$this->isSuperTypeOfVarType($type, $varTagType)->no()

It was unclear to me if I should have call toPhpDocNode for maybe results. So I did it every time.

[ondrejmirtes]

Please do what I suggested. I don't want to break any existing use-cases, that's why I want to do the "new" thing right before when the original error message would appear. To decrease the number of reported errors, not to potentially incraese it.

[VincentLanglet]

Sure, but I don't understand how I'll know "an error would be reported" easily since sometime the usage will be
!$this->isSuperTypeOfVarType($type, $varTagType)->yes() and sometimes it will be
$this->isSuperTypeOfVarType($type, $varTagType)->no().

The following implementation

private function isSuperTypeOfVarType(Type $type, Type $varTagType): TrinaryLogic
	{
	    $result = $type->isSuperTypeOf($varTagType);
        if ($result->yes()) {
             return $result;
        }
	
		$type = $this->typeNodeResolver->resolve($type->toPhpDocNode(), new NameScope(null, []));

		return $type->isSuperTypeOf($varTagType);
	}

is not satisfying since for case where the usage is

$this->isSuperTypeOfVarType($type, $varTagType)->no()

I use the strategy toPhpDocNode for nothing since the first isSuperTypeOf wouldn't have report any error.

[VincentLanglet]

I have the implementation
d8b9c48

But I dunno if it's satisfying enough

[ondrejmirtes]

Please do what I suggested. I don't want to break any existing use-cases, that's why I want to do the "new" thing right before when the original error message would appear. To decrease the number of reported errors, not to potentially incraese it.

[ondrejmirtes]

We don't need to be too smart. We csn have two different methods - one for !yes(), one for no ().

[VincentLanglet]

We don't need to be too smart. We csn have two different methods - one for !yes(), one for no ().

I created isSuperTypeOfVarType and isAtLeastMaybeSuperTypeOfVarType