Fix GH-12895: Incorrect ReflectionClass::getNamespaceName for anonymous class

[nielsdos]

Because paths can contain backslashes, searching for the first backslash starting on the right for detecting namespaces yields the wrong result. Furthermore, because the namespace of the class is not actually in the name (only the interface or parent class is) the namespace can be wrong.

This fix consists of two parts:

• Fix how a prefix is chosen such that the correct namespace is included in the name.
• If ZEND_ACC_ANON_CLASS is set, check the position of @ and search for the backslash before that.

While the issue seems Windows-specific, I could even reproduce this on Linux because Linux allows directories to contain backslashes.

Important I'm not so sure we can fully fix this in stable branches. The previous time the name scheme for anonymous classes changed was targeted to the master branch. However, the check for the ZEND_ACC_ANON_CLASS flag is safe for stable branches I think.

[mvorisek]

I would be happy if the "1st parent non-anonymous class" can be kept dumped, as from anonymous class it is often not easy to tell quickly what the anonymous class is about and the "1st parent non-anonymous class" was helpful, compare:

DateTime@anonymous vs. class@anonymous (or Ns\@anonymous) - https://3v4l.org/Bj2qg

[nikic]

This should be a reflection-only change, not changing the generated name for anonymous classes.

[nikic]

It's not super clear what the behavior for anonymous classes should actually be -- I could see either returning the namespace portion before the @, or just always saying that the namespace name for anon classes is empty. The concept is not really well-defined for anon classes.

[nielsdos]

Returning the namespace portion before @ without changing the generated name seems definitely wrong to me because then it'll return the namespace portion of the first interface or parent class.
Returning an empty string could make some sense and is consistent with the result for getNamespaceName for an anonymous class without interfaces/parent class.

[mvorisek]

IMHO NS should be returned exactly the same as for non-anonymous class - https://3v4l.org/5RLAY and https://3v4l.org/bal7C

I belive so as NS (ReflectionClass::getNamespaceName()) is at least used to resolve relative names in phpdocs.

[iluuu1994]

Also worth noting that __NAMESPACE__ currently refers to the lexical namespace in an anonymous class. Having (new \ReflectionClass($this))->getNamespaceName() diverge is a bit odd.

[nielsdos]

Right, it should be consistent with at least __NAMESPACE__ in which case I can't avoid changing the generated name.

[nikic]

Also worth noting that __NAMESPACE__ currently refers to the lexical namespace in an anonymous class. Having (new \ReflectionClass($this))->getNamespaceName() diverge is a bit odd.

It's a bit odd, but it's not necessarily wrong. The lexical and logical scope are not necessarily the same. A very similar example would be the behavior of __CLASS__ vs self::class inside a closure:

class A {}

class B {
    public function test() {
        $fn = function() {
            var_dump(__CLASS__);
            var_dump(self::class);
        };
        $fn->bindTo(new A, 'A')();
    }
}
(new B)->test();

Here __CLASS__ always refers to the lexical scope, which will be "B", as the closure was declared lexically within class B. However, this does not match the self scope of the closure. self::class will return "A" in this example.

[derickr]

Are anonymous classes really part of a namespace? Shouldn't they be "anonymous" and hence not even have any name logically?

[mvorisek]

The namespace is not about only naming, but also about how names not startwing with \ are resolved. In anonymous classes, there are resolved (correctly) using the actual namespace. The same for PHPDoc.