Is it possible to overuse late static binding in PHP?

Question

Starting with version 5.3, PHP supports late binding for static methods. While it's an undoubtedly useful feature, there are only several cases where its use is really necessary (e.g. the Active Record pattern).

Consider these examples:

1. Convenience constructors (::create())

class SimpleObject
{
    public function __construct() { /* ... */ }

    public static function create()
    {
        return new static; // or: return new self;
    }
}

If this class may be extended (however, it's not extended by any class in the same package), should late static binding be used just to make extending it easier (without having to rewrite the ::create() method, and, more importantly, without having to remember to do that)?

Note: this idiom is used to work around the impossibility to call methods on just constructed objects: new SimpleObject()->doStuff() is invalid in PHP.

---

2. Class constants

class TagMatcher
{
    const TAG_PATTERN = '/\<([a-z\-]+?)\>/i';

    private $subject;

    public function construct($subject) { $this->subject = $subject; }

    public function getAllTags()
    {
        $pattern = static::TAG_PATTERN;
        preg_match_all($pattern, $this->subject);
        return $pattern[1];
    }
}

The reason to use static:: in this example is similar to the previous one. It's used just because this class can be made to match differently formed tags just by extending it and overriding the constant.

---

So, to wrap it all up, are these uses (and similar ones) of late static binding are an overkill? Is there any noticeable performance hit? Also, does frequent use of late binding reduce the overall performance boost given by opcode caches?

Answer 1

So, to wrap it all up, are these uses (and similar ones) of late static binding are an overkill? Is there any noticeable performance hit? Also, does frequent use of late binding reduce the overall performance boost given by opcode caches?

The introduction of late static binding fixes a flaw in PHP's object model. It's not about performance, it's about semantics.

For example, I like to use static methods whenever the implementation of the method doesn't use $this. Just because a method is static doesn't mean to say that you don't want to override it sometimes. Prior to PHP 5.3, the behavior was that no error was flagged if you overrode a static method, but PHP would just go ahead and silently use the parent's version. For example, the code below prints 'A' before PHP 5.3. That's highly unexpected behavior.

Late static binding fixes it, and now the same code prints 'B'.

<?php
class A {
  public static function who() {
    echo __CLASS__;
  }
  public static function test() {
    static::who();
  }
}

class B extends A {
  public static function who() {
    echo __CLASS__;
  }
}

B::test();
?>

Answer 2

static methods (early- or late-bound) create tight coupling and (thus) reduce testability. you can create large programs in PHP without using more than a few static calls. for me, late static methods are a non-feature.

edit to answer Marco Demaio's question, how do static method reduce testability?

i'm sorry if this is all obvious to you, static members (both data and methods) are useful and do no harm if used responsibly, i was alluding to their prevalent misuse.

say you have a web application that uses an SQL database. your business objects may retrieve data using a static interface or through polymorphism. either

class MyBusinessObject
extends...
{
  public function doThisOrThat(...)
  {
    $results = db::query('sql string...');
    ...
  }
}

or

class MyBusinessObject
extends...
{
  public function __construct(dbconn $db)
  {
    $this->db = $db;
  }
  private $db;
  public function doThisOrThat(...)
  {
    $results = $this->db->query('sql string...');
    ...
  }
}

the latter is easier to test (as in: i want to test that the sql string constructed from such-and-such inputs is such-and-such) because it's easier to create another implementation of the dbconn interface than it is to change the meaning of db::. why you'd want either? because you don't need a real database to test the sql-composing behavior, and in fact it's easier to test for without a real database. also, it's easier to stub out the sql consumer if your tests are concerned with another aspect of the CUT (Code Under Test).

testing always implies lying to the tested code about its collaborators, and abstaining from static interfaces (the "doublecolon" or "quadridot") means the lie need not be a massive surgery, which is a plus, since the farther the tested code is from the production code, the less meaningful the test results are.