Template type is not narrowed down in conditional return types

[canvural]

Bug report

Hello 👋🏽

I have the following snippet:

<?php declare(strict_types = 1);

class Model {}

/**
 * @template TKey of array-key
 * @template TModel
 *
 */
class SupportCollection {}

/**
 * @template TKey of array-key
 * @template TModel of Model
 *
 */
class Collection
{
    /**
     * Run a map over each of the items.
     *
     * @template TMapValue
     *
     * @param  callable(TModel, TKey): TMapValue  $callback
     * @return (TMapValue is Model ? static<TKey, TMapValue> : SupportCollection<TKey, TMapValue>)
     */
    public function map(callable $callback) {} // @phpstan-ignore return.missing
}

I was expecting the TMapValue to be narrowed down to Model in static<TKey, TMapValue> because we are in the truthy branch of the if condition. But as it can be seen in the playground, it doesn't work like that.

Am I wrong to expect that, or is it a bug?

Code snippet that reproduces the problem

https://phpstan.org/r/92226828-0e09-4210-9906-6b1df14395c0

Expected output

No errors.

Did PHPStan help you today? Did it make you happy in any way?

It helps me every day!

[ondrejmirtes]

Even if we eliminate static<TKey, TMapValue> which isn't supported, it still does not work: https://phpstan.org/r/385b8e60-9e39-474f-963b-2a6559a975d8

I think we need some logic to make it understand that TMapValue in the then part of the conditional type always has Model bound which it currently doesn't understand.

If you try to use this signature does it work as expected? You can try some dumpType calls.

[canvural]

Yes it works, the return type is narrowed down correctly. But this code is in the stubs, so it throws the same error you see in the playground, when it validates the stub files.

For now I'll try to do the same thing with return type extensions.

[rvanvelzen]

When the conditional type is resolved the type is narrowed correctly, but GenericObjectTypeCheck does not do this as there is no real way to do it without resolving the type (which in turn could cause false negatives).

[ondrejmirtes]

I think it could apply the condition in the then branch and apply it inversely in the else branch. So the type in the then branch would be an intersection with the subject type.

[rvanvelzen]

Yep, that's basically what's happening in ConditionalType::getResult() already. That logic should be reused for this.

[ondrejmirtes]

Fixed phpstan/phpstan-src#2948

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.