Roberto Diana Roberto Diana avatar

10 minute read

Google from zero to infinite

Sometimes I do a full immersion on topics of my interest, generally related to programming, topics on testing, good design, etc. I apply a technique invented by me named: “Google from zero to infinite”.

Basically I use one or more keywords in Google, then I literally follow all the links of all the pages. It is a very long and laborious activity. However, most of the time I find very interesting documents, otherwise difficult to reach. This time I decided to use the final keyword of PHP.

The concept behind final keyword

The keyword final was introduced in version 5 of php.

The manual says:

PHP 5 introduces the final keyword, which prevents child classes from overriding a method by prefixing the definition with final. If the class itself is being defined final then it cannot be extended.

PHP: Final Keyword - Manual

Java and C# use the same keyword but it is called sealed instead of final. These keywords have more or less the same behavior: they prevent a class from being inherited. In this article, we will discuss this bahavior only, even if some languages like Java support further behaviors in different contexts.

In this article we will start by explaining the purpose of the final keyword. Then, we will try to do some additional reasoning. Finally, we will talk about comments found on the net.

The concept expressed by the final keyword is that a class cannot be extended. The programmer must take this decision at the design stage by using this keyword.

In some cases, extension by inheritance is a good solution while in some others it is either not possible or wrong. The final keyword is a “tool” given to the programmer to express this constraint.

Why should someone set such a strong constraint? The answers are many and they all aim at having a robust and more manageable code-base. For example, to avoid the proliferation of infinite hierarchies of classes such as VCL (https://en.wikipedia.org/wiki/Visual_Component_Library), which are difficult to extend and maintain. In this case, even though it is a well-made library, inheritance has been used a lot.

Furthermore, I use the final keyword to reduce the API surface that needs to be covered by BC breaks avoidance in a distributed library.

Furthermore, using final forces the adoption of interfaces for example, for testing. This is a positive side effect indirectly caused by the use of this keyword.

If we encounter a class with the final keyword, it means that the designer of the class doesn’t want us to inherit from it. If we want to extend some behaviors, we must do it by composition or other mechanisms like events (PSR-14 https://www.php-fig.org/psr/psr-14 ) or plugins.

I think that this is a good feature of the language. In fact, by using the final keyword we provide important information on how the class should be used. Therefore, we reduce the effort of developers since they will spend less time analysing the design class.

S.O.L.I.D, Liskov and final

Developing an architecture isn’t a trivial task, many programmers use inheritance as solution to all problems.

There are many principles and guidelines that help to develop good design like S.O.L.I.D (https://en.wikipedia.org/wiki/SOLID), design patterns, (https://en.wikipedia.org/wiki/Design_pattern) etc. However, the problem is that these principles are difficult to apply, they almost always lead to writing more code. Not always this is possible, not always this is accepted.

In my opinion, one of the principles that best represents the use of final is LSP.

Let’s see what Liskov principle says in more detail. The canonical statement is the following:

“you should always be able to substitute the parent class with its derived class without any undesirable behaviour”

The concept of this principle was introduced by Barbara Liskov in a 1987 conference keynote and later published in a paper together with Jannette Wing in 1994.

This is a more precise definition:

“if S is a subtype of T, then objects of type T in a program may be replaced with objects of type S without altering any of the desirable properties of that program.”

Link to original paper: https://dl.acm.org/doi/pdf/10.1145/62138.62141

I used it as an example since it is specific to the design of hierarchies.

I do not want to write an article about developping in accordance with Liskov principle, there is already a lot of literature about it on the web. I would just like to emphasize that building a hierarchy of objects is not easy: the deeper the hierarchy, the more we are faced with design problems. final helps us to prevent these problems.

For this reason, before creating subclasses, we should ask ourselves questions like the following or similar:

  • is my derived class of the same type as the base class?
  • can I use a derived class instead of its base class without having “strange” behaviors in the code at runtime?
  • is my base class prepared to be derived?
  • are the enhancements made by the subclass primarily additive?

If the answers to these questions are negative, using inheritance could lead to complex and unmanageable hierarchies.

I will now present some examples of misuse of inheritance:

use App\Domain\Model;
class Stack extends ArrayList {
    public push(Object value) { … }
    public Object pop() { … }
}

Why is it wrong? Because the Stack class, inheriting from ArrayList, will have a lot of additional methods that are not related to the concept of ArrayList (e.g. push and pop).

In the previous example two main errors have been made:

  1. Stack is NOT an Array
  2. The cross-domain inheritance relationship

Our domain classes must use implementations and not inherit them. In the example, stack was a concept of domain (focusing only on reusing code can be a problem).

You will be probably thinking that the OOP is not keeping its promise. However, let us remember that we are programming in OOP (Object Oriented Programming) not in OOI (Object Oriented Inheritance). There is always a trade off between code reuse and good design. Inheritance should be mostly used for good design, for code reuse we choose composition.

I invite you, however, to do some tests with your classes to see if they adhere to these good practices and try to get a better design by avoiding inheritance.

The concrete example, break the immutable object

Let’s see some examples where final could be the right choice. Through final we create an immutable object with minimal effort. This is only a POC to show how final works. There are many ways to resolve this problem.

Let’s go and see some code.

class SomeImmutableObject
{
    private $someString;

    public function __construct(string $value)
    {
        $this->someString = $value;
    }

    public function getValue(): string
    {
        return 'the value is:' . $this->someString . ' - ';
    }
}

Let’s concentrate: can we modify this object after it has been created? …. Yes we can!

$one = new SomeImmutableObject('Foo');
echo $one->getValue(); //Foo
$one->__construct('Bar'); 
echo $one->getValue(); //Bar

try it! break the immutable object

This problem is easy to solve: we must put a flag in the constructor and, if it is true, it throws an exception. Another way to fix it would be to create a named constuctor and make the __constructor private. Let’s do it using the first solution!

The concrete example, break the immutable object: first fix

<?php declare(strict_types=1);

class SomeImmutableObject
{
    private $someString;
    private $flagCreate = false;

    public function __construct(string $value)
    {
        if (true === $this->flagCreate) {
            throw new \BadMethodCallException('This immutable object has already been created.');
        }

        $this->someString = $value;
        $this->flagCreate = true;
    }

    public function getValue(): string
    {
        return 'the value is:' . $this->someString . ' - ';
    }
}

$two = new SomeImmutableObject('Foo');
echo $two->getValue(); //Foo
$two->__construct('Bar'); // \BadMethodCallException 
echo $two->getValue(); //Bar

Now we are happy and we have our immutable object. Are we sure? Mmm no… have a look at here!

The concrete example, break the immutable object: the second way

<?php
class SomeImmutableObject
{
    private $someString;
    private $flagCreate = false;

    public function __construct(string $value)
    {

        if ($this->flagCreate === true) {
            throw new \BadMethodCallException('This immutable object has already been created.');
        }

        $this->someString = $value;
        $this->flagCreate = true;
    }

    public function getValue(): string
    {
        return 'the value is:' . $this->someString . ' - ';
    }
}

class AnotherClassToBreakImmutableObject extends SomeImmutableObject
{

    public function __construct(string $value)
    {
        $this->someString = $value;
    }

    public function getValue(): string
    {
        return 'the value is:' . $this->someString . ' - ';
    }

    public function change(): void
    {
        $this->someString .= ' and Baz';
    }
}

$three = new AnotherClassToBreakImmutableObject('Foo');
echo $three->getValue(); //Foo
echo $three->change();
echo $three->getValue(); //the value is: Foo and Baz -

try it! break the immutable object

It doesn’t work yet. Ok, now we’re going to fix it, I promise! The problem is that inheritance breaks encapsulation. Therefore, in this case using final is a good solution!.

the definitive version

The concrete example: a definitive fix with final keyword

<?php

final class SomeImmutableObject
{
    private $someString;
    private $flagCreate = false;

    public function __construct(string $value)
    {

        if ($this->flagCreate === true) {
            throw new \BadMethodCallException('This is an immutable object has already create.');
        }

        $this->someString = $value;
        $this->flagCreate = true;
    }

    public function getValue(): string
    {
        return 'the value is:' . $this->someString . ' - ';
    }
}

class TryToBreakImmutableObject extends SomeImmutableObject
{
    public $someString;

    public function __construct(string $value)
    {
        $this->someString = $value;
    }

    public function getValue(): string
    {
        return 'the value is:' . $this->someString . ' - ';
    }

    public function change(): void
    {
        $this->someString .= ' and Minny';
    }
}

$four = new TryToBreakImmutableObject('Foo');
echo $four->getValue(); //Foo
echo $four->change();
echo $four->getValue();

PHP Fatal error: Class BreakImmutableObject may not inherit from final class (SomeImmutableObject)

try it! immutable object

… Here are some links to deepen the topic

Obviously, it isn’t all puppy dogs and rainbows. Technical literature does not perfectly suit to all needs, especially in our field.

What I want to point out is that in some cases final can be a good ally and can help us to simplify the design.
Using final leads programmers to make a round of additional reasoning, to pay more attention to the use of the class. For example, during review, a diff with the removal of final could lead to useful comments to find different solutions.

A downside of the final keyword is that it is considered too coarse an instrument to be effective. A developer can simply remove the final keyword and then do wrong things!.

I do not agree with this. I’ve seen misuse of inheritance many times. Therefore, I think that using the final keyword is useful both for beginners and senior developers.

Another problem is that some mocking libraries don’t work with final class. This problem is solved in php by using the annotation @Final, even if it does not have the same validity as a language keyword. However, nowadays, with Ide inspectors, and with static analysis tools like PhpStan (https://phpstan.org/blog), Psalm (https://psalm.dev/docs/) the annotation is fully supported and it works properly.

To support the last topics I add some links on discussions against the use of final.

Other links that I used as a starting point for this article:

Conclusion

The topic is very complex, this article aims at giving a general overview of it. I hope it will be useful as a starting point to deepen the subject and to evaluate when to use final in our classes.

From my point of view the use of final can help to improve design. I would not write all my class with final but i’d start from the simplest cases: value objects, algorithms, patterns like the template method etc.
In any case, the final keyword can be removed with due precautions and, if nothing else, using it has forcibly added one more step to the analysis process.

comments powered by Disqus