Everything

php

Bug Wrong resolution of "Late Static Binding" after self::method()

php.internals

Marc

6 years ago

Hi,

I have reported a bug in Mai 2019 but did not get any response here
https://bugs.php.net/bug.php?id=77985.


The problem is that "static::class" as well as "get_called_class()" used
within a method resolves to a different class if this method was called
with "self::method". Interestingly it works as expected if the method
got called with "ClassName::method".


I send it here because fixing this could result in a hard to find BC
break and it could be a good idea to handle this in 8.0.


https://3v4l.org/NSdB8

class A {
    static function call() {
        self::method();

        $self = self::class;
        $self::method();
    }
    static function method() { echo static::class; }
}

class B extends A {}

B::call(); // displays "BA" instead of "AA"


In case this is "Not a Bug" please explain to me why the result of
"self::method()" is different then "$self = self::class; $self::method();".


Thanks,

Marc

Rowan Collins

6 years ago

On 8 March 2020 09:47:37 GMT+00:00, Marc <marc@mabe.berlin> wrote:

>https://3v4l.org/NSdB8
>
>class A {
>     static function call() {
>         self::method();
>
>         $self = self::class;
>         $self::method();
>     }
>     static function method() { echo static::class; }
>}
>
>class B extends A {}
>
>B::call(); // displays "BA" instead of "AA"

Someone else may be able to confirm if this is a bug or working as designed, but what seems to be happening is that the value of "static" is not reset when calling a method using self::

This doesn't just apply to getting the class name, but to actual static calls; for instance:

# https://3v4l.org/p2g8E
class A {
static function call() {
self::method1();
}
static function method1() {
static::method2();
}
static function method2() {
echo 'Base definition';
}
}
class B extends A {
static function method2() {
echo 'Override';
}
}
B::call();

If the call to self::method1() reset the called class, this would run A::method2() and echo 'Base definition'; instead, the called class is remembered, and it calls B::method2() and echoes 'Override'.

If you explicitly call A::method1(), as you are effectively doing in your self::class example, the "called class" information is reset, and the call goes through to A::method2().

Regards,
Hi Marc,

--
Rowan Tommins
[IMSoP]

Matthew Brown

6 years ago

This is expected behaviour given my understanding of how late static binding works:

If there is a chain of “self::” calls that ultimately ends in “static::someMethod”, then PHP behaves as if every preceding call was a call to “static::”, not “self::”. In your second call you’re explicitly overriding that resolution by changing using (self::class)::someMethod, which PHP always treats as A::someMethod, thus producing the expected output.

Marc

6 years ago

Hi,

On 08.03.20 17:54, Matthew Brown wrote:

> This is expected behaviour given my understanding of how late static binding works:
>
> If there is a chain of “self::” calls that ultimately ends in “static::someMethod”, then PHP behaves as if every preceding call was a call to “static::”, not “self::”. In your second call you’re explicitly overriding that resolution by changing using (self::class)::someMethod, which PHP always treats as A::someMethod, thus producing the expected output.

Thanks for explaining.

I couldn't find any information for this behavior in the documentation -
Is that documented anywhere?

Does it make sense? -> I have read "self::" all time as a shortcut for
"MyClass::" until I noticed this is not the case and I expect most PHP
devs would explain it this way.

Is there a reason why self:: doesn't reset the internal "static" reference?

Sorry for all these questions.

Sometimes the world could be so simple until it turns out it isn't for
no reason.

Thanks,

Marc

Nikita Popov

6 years ago

On Mon, Mar 9, 2020 at 6:38 AM Marc <marc@mabe.berlin> wrote:

> Hi,
>
>
> On 08.03.20 17:54, Matthew Brown wrote:
> > This is expected behaviour given my understanding of how late static
> binding works:
> >
> > If there is a chain of “self::” calls that ultimately ends in
> “static::someMethod”, then PHP behaves as if every preceding call was a
> call to “static::”, not “self::”. In your second call you’re explicitly
> overriding that resolution by changing using (self::class)::someMethod,
> which PHP always treats as A::someMethod, thus producing the expected
> output.
>
> Thanks for explaining.
>
> I couldn't find any information for this behavior in the documentation -
> Is that documented anywhere?
>
> Does it make sense? -> I have read "self::" all time as a shortcut for
> "MyClass::" until I noticed this is not the case and I expect most PHP
> devs would explain it this way.
>
> Is there a reason why self:: doesn't reset the internal "static" reference?
>
>
> Sorry for all these questions.
>
> Sometimes the world could be so simple until it turns out it isn't for
> no reason.
>
>
> Thanks,
>
> Marc
>

See https://wiki.php.net/rfc/lsb_parentself_forwarding, which is the RFC
that added this behavior.

Nikita

Rowan Collins

6 years ago

On Mon, 9 Mar 2020 at 05:38, Marc <marc@mabe.berlin> wrote:

>
> Does it make sense? -> I have read "self::" all time as a shortcut for
> "MyClass::" until I noticed this is not the case and I expect most PHP
> devs would explain it this way.
>
> Is there a reason why self:: doesn't reset the internal "static" reference?
>

A reasonably intuitive parallel that occurred to me is that $this is also
maintained through self:: calls (as long as they're not to a method
declared static).

So if you replace the static:: call in my previous example with a $this->
call, you get the same result:


# https://3v4l.org/deRcD
class A {
function call() {
self::method1();
}
function method1() {
$this->method2();
}
function method2() {
echo 'Base definition';
}
}
class B extends A {
function method2() {
echo 'Override';
}
}
(new B)->call(); # echoes 'Override'


Regards,

--
Rowan Tommins
[IMSoP]

Marc

6 years ago

Thank you all for explaining.

This helps a lot!

Marc

On 09.03.20 10:27, Rowan Tommins wrote: