calling static methods on strings

[SignpostMarv]

• 0.8.x no errors: https://phpstan.org/r/27b31b75cf52496d8d4809036b61bc66
• master has errors: https://phpstan.org/r/27b31b75cf52496d8d4809036b61bc66
• 3v4l only has errors on php5: https://3v4l.org/5dlg9

<?php declare(strict_types = 1);

interface foo
{
	public static function bar();
}

function baz1(string $thing)
{
	$thing::bar();
}

function baz2(string $thing)
{
	if (is_a($thing, foo::class, true)) {
		$thing::bar();
	}
}

function baz3(string $thing)
{
	if (is_a($thing, foo::class, true) === false) {
		throw new Exception('fail');
	}
	
	$thing::bar();
}

ideally, this should only cause an error in the case of baz1, since there's nothing in particular to indicate that $thing is an implementation of foo

[SignpostMarv]

the workaround on 74c3cf8 seems to be to @var foo $thing then (string) $foo for any function that requires a string parameter.

[ondrejmirtes]

There are two things going on at once:

1. There's currently a bug on dev-master because PHPStan thinks that static methods cannot be called on a string, which isn't true.
2. There's no type-specifying code when is_a is used, unlike for is_int (and other functions) and instanceof keyword which ARE supported. So PHPStan does not recognize your type check when calling is_a.

About 1), I'm torn whether PHPStan should complain when a method is called on a string without any is_a check, forcing the developer to have a sanity check:

public function doFoo(string $className)
{
    $className::doSomething(); // dangerous, report Cannot call static method doSomething() on string.
}

versus:

public function doFoo(string $className)
{
    if (!is_a($className, DoesSomething::class, true)) {
        throw new \InvalidArgumentException));
    }
    $className::doSomething(); // this is fine
}

It probably should report that, but the error message is not a convenient place to explain that the developer should use is_a check beforehand.

About 2) - I will try to implement this and I hope the current type system will allow this. If not, it will have to be postponed until later. If everything goes well, both 1) and 2) will be adressed at once.

[SignpostMarv]

no rush on it, although mildly inellegant, the workaround to (string) $className after * @var SomeClass $className works for now, the static analysis I've got running doesn't complain about typecasting to string something that's already string :)

p.s. see the changes here corresponding to (mostly phpstan) today's update on static analysis.

[ondrejmirtes]

Fixee the "cannot call on string" problem: a7c931c

I'm removing the "0.9" milestone and leaving this open for the is_a support.

[JanTvrdik]

@ondrejmirtes Removing the milestone seems unnecessary as is_a is already implement as part of #507

[ondrejmirtes]

@JanTvrdik Sorry, didn't notice that, the milestone is back :)

[ondrejmirtes]

#507 got merged so this issue should be completely solved. Feel free to test dev-master.