我知道这可能会引起不必要的争论，但是我知道您可能需要多个数据库连接，所以我承认单例可能不是最好的解决方案……然而，我发现单例模式还有其他非常有用的用途。

Here's an example: I decided to roll my own MVC and templating engine because I wanted something really lightweight. However, the data that I want to display contains a lot of special math characters such as ≥ and μ and what have you... The data is stored as the actual UTF-8 character in my database rather than pre-HTML-encoded because my app can deliver other formats such as PDF and CSV in addition to HTML. The appropriate place to format for HTML is inside the template ("view" if you will) that is responsible for rendering that page section (snippet). I want to convert them to their appropriate HTML entities, but PHPs get_html_translation_table() function is not super fast. It makes better sense to retrieve the data one time and store as an array, making it available for all to use. Here's a sample I knocked together to test the speed. Presumably, this would work regardless of whether the other methods you use (after getting the instance) were static or not.

class EncodeHTMLEntities {

    private static $instance = null;//stores the instance of self
    private $r = null;//array of chars elligalbe for replacement

    private function __clone(){
    }//disable cloning, no reason to clone

    private function __construct()
    {
        $allEntities = get_html_translation_table(HTML_ENTITIES, ENT_NOQUOTES);
        $specialEntities = get_html_translation_table(HTML_SPECIALCHARS, ENT_NOQUOTES);
        $this->r = array_diff($allEntities, $specialEntities);
    }

    public static function replace($string)
    {
        if(!(self::$instance instanceof self) ){
            self::$instance = new self();
        }
        return strtr($string, self::$instance->r);
    }
}
//test one million encodings of a string
$start = microtime(true);
for($x=0; $x<1000000; $x++){
    $dump = EncodeHTMLEntities::replace("Reference method for diagnosis of CDAD, but clinical usefulness limited due to extended turnaround time (≥96 hrs)");
}
$end = microtime(true);
echo "Run time: ".($end-$start)." seconds using singleton\n";
//now repeat the same without using singleton
$start = microtime(true);
for($x=0; $x<1000000; $x++){
    $allEntities = get_html_translation_table(HTML_ENTITIES, ENT_NOQUOTES);
    $specialEntities = get_html_translation_table(HTML_SPECIALCHARS, ENT_NOQUOTES);
    $r = array_diff($allEntities, $specialEntities);
    $dump = strtr("Reference method for diagnosis of CDAD, but clinical usefulness limited due to extended turnaround time (≥96 hrs)", $r);
}
$end = microtime(true);
echo "Run time: ".($end-$start)." seconds without using singleton";

基本上，我看到的典型结果如下:

php test.php
Run time: 27.842966794968 seconds using singleton
Run time: 237.78191494942 seconds without using singleton

因此，虽然我不是专家，但我认为没有更方便、更可靠的方法来减少对某些数据的缓慢调用的开销，同时使其超级简单(只需一行代码就可以完成所需的工作)。假设我的例子只有一个有用的方法，因此并不比一个全局定义的函数好，但是一旦你有了两个方法，你就想把它们组合在一起，对吗?我是不是太离谱了?

此外，我更喜欢真正做一些事情的例子，因为有时很难想象一个例子中包含“//在这里做一些有用的事情”这样的语句，这是我在搜索教程时经常看到的。

无论如何，我希望有任何关于为什么在这种情况下使用单例是有害的(或过于复杂)的反馈或评论。

protected  static $_instance;

public static function getInstance()
{
    if(is_null(self::$_instance))
    {
        self::$_instance = new self();
    }
    return self::$_instance;
}

这段代码可以应用于任何类，而无需关心类名。

/**
 * Singleton class
 *
 */
final class UserFactory
{
    private static $inst = null;

    // Prevent cloning and de-serializing
    private function __clone(){}
    private function __wakeup(){}


    /**
     * Call this method to get singleton
     *
     * @return UserFactory
     */
    public static function Instance()
    {
        if ($inst === null) {
            $inst = new UserFactory();
        }
        return $inst;
    }
    
    /**
     * Private ctor so nobody else can instantiate it
     *
     */
    private function __construct()
    {
        
    }
}

使用方法:

$fact = UserFactory::Instance();
$fact2 = UserFactory::Instance();

$fact == $fact2;

But:

$fact = new UserFactory()

抛出错误。

参见http://php.net/manual/en/language.variables.scope.php#language.variables.scope.static了解静态变量的作用域以及为什么设置static $inst = null;的工作原理。

所有这些复杂性(“后期静态绑定”……对我来说，harumph)只是PHP对象/类模型破碎的标志。如果类对象是一级对象(参见Python)，那么"$_instance"将是一个类实例变量——类对象的成员，而不是其实例的成员/属性，也与它的后代共享相反。在Smalltalk世界中，这就是“类变量”和“类实例变量”之间的区别。

在PHP中，在我看来，我们似乎需要牢记模式是编写代码的指南——我们可能会考虑Singleton模板，但试图编写从实际的“Singleton”类继承的代码对PHP来说是错误的(尽管我认为一些有进取心的人可以创建一个合适的SVN关键字)。

我将继续使用共享模板分别编写每个单例。

请注意，我绝对不会参与单身是邪恶的讨论，生命太短暂了。

这篇文章涵盖了相当广泛的主题: http://www.phptherightway.com/pages/Design-Patterns.html#singleton

Note the following: The constructor __construct() is declared as protected to prevent creating a new instance outside of the class via the new operator. The magic method __clone() is declared as private to prevent cloning of an instance of the class via the clone operator. The magic method __wakeup() is declared as private to prevent unserializing of an instance of the class via the global function unserialize(). A new instance is created via late static binding in the static creation method getInstance() with the keyword static. This allows the subclassing of the class Singleton in the example.

不幸的是，当有多个子类时，Inwdr的答案就失效了。

下面是一个正确的可继承单例基类。

class Singleton
{
    private static $instances = array();
    protected function __construct() {}
    protected function __clone() {}
    public function __wakeup()
    {
        throw new Exception("Cannot unserialize singleton");
    }

    public static function getInstance()
    {
        $cls = get_called_class(); // late-static-bound class name
        if (!isset(self::$instances[$cls])) {
            self::$instances[$cls] = new static;
        }
        return self::$instances[$cls];
    }
}

测试代码:

class Foo extends Singleton {}
class Bar extends Singleton {}

echo get_class(Foo::getInstance()) . "\n";
echo get_class(Bar::getInstance()) . "\n";