Thank you for taking the time to write to us, but this is not
a bug. Please double-check the documentation available at
http://www.php.net/manual/ and the instructions on how to report
a bug at http://bugs.php.net/how-to-report.php

This is how it works... you can always open up an API through a new extended interface, but not hide more.

is this the 'php'-dev definition?
i'm asking cause wraping a singleton pattern around a subclass makes sense
and the same example is valid for java and c++
to ask it differently: why is it working this way in php? (i'm interested in the background of this)
my point is that 2 languages allow it and there is an example which is valid, not?

I'm in the same situation as our friend "prometheus" here, I am trying to use a singleton pattern and logically, this should involve being able to encapsulate the subcalasses information particularly setting up a private constructor for the singleton subclass, IT DEFINITELY DOES MAKE SENSE, so please try to take this under consideration for the next php engine release.

Same problem here. You cannot actually apply the singleton pattern to subclasses with this behavior. I do not see any reason for not calling this a bug.

Apparently, we don't call this a "bug" because this is not caused by a programming error. So I'm calling it a "massive design flaw" instead.

Same "problem".

class A{
    public function __construct(){}
}

class B extends A {
    protected function __construct(){}
}

Please, can you tell us a reason why we can't put an private/protected access to B's constructor if A's constructor is public ?
Is there any logical reason to this ?


Thanks.

Most (all?) object oriented languages work this way. Java and C# both do. You can 
loosen visibility when you override a parent method, but you can never tighten 
it. This is part of what is known as the Liskov Substitution Principle and it is 
one of the fundamental principles of object oriented programming. You can read 
about at http://en.wikipedia.org/wiki/Liskov_substitution_principle

I can guarantee that abiding by LSP is not a bug

@rasmus:
Then why is there a strict warning regarding the compatibility of method signatures for this code:

--------------
class Foo {
    public function test() {
    }
}
 
class Bar extends Foo {
    public function test(ArrayObject $arrayObj, $number = 0) {
        /* do stuff with $arrayObj and $number */
    }
}

---------------

and not for this one:

---------------
class Foo {
    public function __construct() {
    }
}
 
class Bar extends Foo {
    public function __construct(ArrayObject $arrayObj, $number = 0) {
        /* do stuff with $arrayObj and $number */
    }
}
------------

No, as stated here : 
- http://stackoverflow.com/questions/5490824/should-constructors-comply-with-the-liskov-substitution-principle 
- and here: http://ralphschindler.com/2012/03/09/php-constructor-best-practices-and-the-prototype-pattern, the LSP does not apply to constructors for several reasons:
- When you call the constructor, you know whether you're using the type or one of its subtypes.
- You can't apply the LSP to an object that does not exist yet.

I think this is a design flaw. Not "massive", but a design flaw indeed.

@postmaster at greg0ire dot fr

__construct applies only to the class where it is declared.

On the other hand, a method, already defined in the parent class, must be as 
visible as the one declared in the parent class and with a compatible signature. 
The 2nd clause is not respected in your example as the 1st argument is not 
optional. It would work if it was optional, making $a->test() possible ass it is 
with the parent class.

All these details are very explained in the PHP manual and other various OOP 
references out there.

@pajoye : let me explain better what I tried to point out there:

Rasmus says the Fatal error we're getting is here to enforce the LSP.
What I'm saying here is that I'm not getting a signature-compatibility strict notice with the __construct() example , so obviously, the developer who wrote this notice thinks that LSP should not apply to __construct() (and I agree with him).

This is just one more argument to convince Rasmus that the LSP does not apply here : obviously, even some member of the php team are aware of that.

So to sum up the rules you are talking about, though "explained in the PHP manual", should not apply to __construct() (please read the references I gave in my previous comment to understand why).

This bug report isn't about constructors specifically though. See the bug title 
and the opening points (1) and (2). The fact that __construct was used as an 
example seemed entirely coincidental here.

For constructors specifically, I agree that Liskov can be mostly ignored, and we 
do so as you have discovered. The one case where we haven't is visibility. It 
seems a bit of an edge-case to tighten visibility on an inherited constructor 
and if that discourages the singleton pattern we might be doing the world a 
favour here.

I filed a separate bug report here: https://bugs.php.net/bug.php?id=61970 so that this bug report does not get too cluttered.

I agree that Singleton is not the best of the design patterns, but what if someone wants to use it anyway? And what if someone needs a protected / private constructor for some reason I cannot imagine?