Polymorphism, Design Patterns, and PHP Programming

Last week, we continued our discussion of the object-oriented features of PHP 5 by taking a first look at design patterns. This week, we will continue looking at design patterns, and examine polymorphism. This article, the third of four parts, is excerpted from chapter two of the book Advanced PHP Programming, written by George Schlossnagle (Sams; ISBN: 0672325616).

Polymorphism

The database wrappers developed in this chapter are pretty generic. In fact, if you look at the other database extensions built in to PHP, you see the same basic functionality over and over again—connecting to a database, preparing queries, executing queries, and fetching back the results. If you wanted to, you could write a similar DB_Pgsql or DB_Oracle class that wraps the PostgreSQL or Oracle libraries, and you would have basically the same methods in it.

In fact, although having basically the same methods does not buy you anything, having identically named methods to perform the same sorts of tasks is important. It allows for polymorphism, which is the ability to transparently replace one object with another if their access APIs are the same.

In practical terms, polymorphism means that you can write functions like this:

function show_entry($entry_id, $dbh) 
{
 $query = "SELECT * FROM Entries WHERE
entry_id = :1"; $stmt = $dbh->prepare($query)->execute($entry_id); $entry = $stmt->fetch_row(); // display entry }

This function not only works if $dbh is a DB_Mysql object, but it works fine as long as $dbh implements a prepare() method and that method returns an object that implements the execute() and fetch_assoc() methods.

To avoid passing a database object into every function called, you can use the concept of delegation. Delegation is an OO pattern whereby an object has as an attribute another object that it uses to perform certain tasks.

The database wrapper libraries are a perfect example of a class that is often delegated to. In a common application, many classes need to perform database operations. The classes have two options:

  • You can implement all their database calls natively. This is silly. It makes all the work you’ve done in putting together a database wrapper pointless.

  • You can use the database wrapper API but instantiate objects on-the-fly. Here is an example that uses this option:

    class Weblog {
     public function show_entry($entry_id)
     {
      $query = "SELECT * FROM Entries WHERE
    entry_id = :1"; $dbh = new Mysql_Weblog(); $stmt =
    $dbh->prepare($query)->execute($entry_id); $entry = $stmt->fetch_row(); // display entry } }

    On the surface, instantiating database connection objects on-the-fly seems like a fine idea; you are using the wrapper library, so all is good. The problem is that if you need to switch the database this class uses, you need to go through and change every function in which a connection is made.

  • You implement delegation by having Weblog contain a database wrapper object as an attribute of the class. When an instance of the class is instantiated, it creates a database wrapper object that it will use for all input/output (I/O). Here is a re-implementation of Weblog that uses this technique:

    class Weblog {
     protected $dbh;
    
     public function setDB($dbh)
     {
      $this->dbh = $dbh;
     }
     public function show_entry($entry_id)
     {
      $query = "SELECT * FROM Entries WHERE
    entry_id = :1"; $stmt = $this->dbh->prepare($query)->execute
    ($entry_id); $entry = $stmt->fetch_row(); // display entry } }

Now you can set the database for your object, as follows:

$blog = new Weblog;
$dbh = new Mysql_Weblog;
$blog->setDB($dbh);

Of course, you can also opt to use a Template pattern instead to set your database delegate:

class Weblog_Std extends Weblog {
 protected $dbh;
 public function _ _construct()
 {
  $this->dbh = new Mysql_Weblog;
 }
}
$blog = new Weblog_Std;

Delegation is useful any time you need to perform a complex service or a service that is likely to vary inside a class. Another place that delegation is commonly used is in classes that need to generate output. If the output might be rendered in a number of possible ways (for example, HTML, RSS [which stands for Rich Site Summary or Really Simple Syndication, depending on who you ask], or plain text), it might make sense to register a delegate capable of generating the output that you want.

{mospagebreak title=Interfaces and Type Hints}

A key to successful delegation is to ensure that all classes that might be dispatched to are polymorphic. If you set as the $dbh parameter for the Weblog object a class that does not implement fetch_row(), a fatal error will be generated at runtime. Runtime error detection is hard enough, without having to manually ensure that all your objects implement all the requisite functions.

To help catch these sorts of errors at an earlier stage, PHP5 introduces the concept of interfaces. An interface is like a skeleton of a class. It defines any number of methods, but it provides no code for them—only a prototype, such as the arguments of the function. Here is a basic interface that specifies the methods needed for a database connection:

interface DB_Connection {
 public function execute($query);
 public function prepare($query);
}

Whereas you inherit from a class by extending it, with an interface, because there is no code defined, you simply agree to implement the functions it defines in the way it defines them.

For example, DB_Mysql implements all the function prototypes specified by DB_Connection, so you could declare it as follows:

class DB_Mysql implements DB_Connection {
 /* class definition */
}

If you declare a class as implementing an interface when it in fact does not, you get a compile-time error. For example, say you create a class DB_Foo that implements neither method:

<?php
require "DB/Connection.inc";
class DB_Foo implements DB_Connection {}
?>

Running this class generates the following error:

Fatal error: Class db_foo contains 2 abstract
methods and must be declared abstract (db connection::execute, db
connection:: prepare) in /Users/george/Advanced
PHP/examples/chapter-2/14.php on line 3

PHP does not support multiple inheritance. That is, a class cannot directly derive from more than one class. For example, the following is invalid syntax:

class A extends B, C {}

However, because an interface specifies only a prototype and not an implementation, a class can implement an arbitrary number of interfaces. This means that if you have two interfaces A and B, a class C can commit to implementing them both, as follows:

<?php

interface A {
 public function abba();
}

interface B {
 public function bar();
}

class C implements A, B {
 public function abba()
 {
  // abba;
 }
 public function bar()
 {
  // bar;
 }
}
?>

An intermediate step between interfaces and classes is abstract classes. An abstract class can contain both fleshed-out methods (which are inherited) and abstract methods (which must be defined by inheritors). The following example shows an abstract class A, which fully implements the method abba() but defines bar() as an abstract:

abstract class A {
 public function abba()
 {
  // abba
 }
 abstract public function bar();
}

Because bar() is not fully defined, it cannot be instantiated itself. It can be derived from, however, and as long as the deriving class implements all of A‘s abstract methods, it can then be instantiated. B extends A and implements bar(), meaning that it can be instantiated without issue:

class B {
 public function bar()
 {
  $this->abba();
 }
}
$b = new B;

Because abstract classes actually implement some of their methods, they are considered classes from the point of view of inheritance. This means that a class can extend only a single abstract class.

Interfaces help prevent you from shooting yourself in the foot when you declare classes intended to be polymorphic, but they are only half the solution to preventing delegation errors. You also need to be able to ensure that a function that expects an object to implement a certain interface actually receives such an object.

You can, of course, perform this sort of computation directly in your code by manually checking an object’s class with the is_a() function, as in this example:

function addDB($dbh) 
{
 if(!is_a($dbh, "DB_Connection")) {
  trigger_error("$dbh is not a DB_Connection
object", E_USER_ERROR); } $this->dbh = $dbh; }

This method has two flaws:

  • It requires a lot of verbiage to simply check the type of a passed parameter.

  • More seriously, it is not a part of the prototype declaration for the function. This means that you cannot force this sort of parameter checking in classes that implement a given interface.

PHP5 addresses these deficiencies by introducing the possibility of type-checking/type hinting in function declarations and prototypes. To enable this feature for a function, you declare it as follows:

function addDB(DB_Connection $dbh)
{
 $this->dbh = $dbh;
}

This function behaves exactly as the previous example, generating a fatal error if $dbh is not an instance of the DB_Connection class (either directly or via inheritance or interface implementation).

{mospagebreak title=The Factory Pattern}

The Factory pattern provides a standard way for a class to create objects of other classes. The typical use for this is when you have a function that should return objects of different classes, depending on its input parameters.

One of the major challenges in migrating services to a different database is finding all the places where the old wrapper object is used and supplying the new one. For example, say you have a reporting database that is backed against an Oracle database that you access exclusively through a class called DB_Oracle_Reporting:

class DB_Oracle_Reporting extends DB_Oracle { /*
... */}

and because you had foresight DB_Oracle uses our standard database API.

class DB_Oracle implements DB_Connection { /* ...
*/ }

Scattered throughout the application code, whenever access to the reporting database is required, you have wrapper instantiations like this:

$dbh = new DB_Oracle_Reporting;

If you want to cut the database over to use the new wrapper DB_Mysql_Reporting, you need to track down every place where you use the old wrapper and change it to this:

$dbh = new DB_Mysql_Reporting;

A more flexible approach is to create all your database objects with a single factory. Such a factory would look like this:

function DB_Connection_Factory($key)
{
 switch($key) {
  case "Test":
   return new DB_Mysql_Test;
  case "Prod":
   return new DB_Mysql_Prod;
  case "Weblog":
   return new DB_Pgsql_Weblog;
  case "Reporting":
   return new DB_Oracle_Reporting;
  default:
   return false;
 }
}

Instead of instantiating objects by using new, you can use the following to instantiate objects:

$dbh = DB_Connection_factory("Reporting");

Now to globally change the implementation of connections using the reporting interface, you only need to change the factory.

{mospagebreak title=The Singleton Pattern}

One of the most lamented aspects of the PHP4 object model is that it makes it very difficult to implement singletons. The Singleton pattern defines a class that has only a single global instance. There are an abundance of places where a singleton is a natural choice. A browsing user has only a single set of cookies and has only one profile. Similarly, a class that wraps an HTTP request (including headers, response codes, and so on) has only one instance per request. If you use a database driver that does not share connections, you might want to use a singleton to ensure that only a single connection is ever open to a given database at a given time.

There are a number of methods to implement singletons in PHP5. You could simply declare all of an object’s properties as static, but that creates a very weird syntax for dealing with the object, and you never actually use an instance of the object. Here is a simple class that implements the Singleton pattern:

<?php
class Singleton {
 static $property;
 public function _ _construct() {}
}

Singleton::$property = "foo";
?>

In addition, because you never actually create an instance of Singleton in this example, you cannot pass it into functions or methods.

One successful method for implementing singletons in PHP5 is to use a factory method to create a singleton. The factory method keeps a private reference to the original instance of the class and returns that on request. Here is a Factory pattern example. getInstance() is a factory method that returns the single instance of the class Singleton.

class Singleton {
 private static $instance = false;
 public $property;

 private function _ _construct() {}
 public static function getInstance()
 {
  if(self::$instance === false) {
   self::$instance = new Singleton;
  }
  return self::$instance;
 }
}

$a = Singleton::getInstance();
$b = Singleton::getInstance();
$a->property = "hello world";
print $b->property;

Running this generates the output "hello world", as you would expect from a singleton. Notice that you declared the constructor method private. That is not a typo; when you make it a private method, you cannot create an instance via new Singleton except inside the scope of the class. If you attempt to instantiate outside the class, you get a fatal error.

Some people are pathologically opposed to factory methods. To satisfy developers who have such leanings, you can also use the _ _get() and _ _set() operators to create a singleton that is created through a constructor:

class Singleton {
 private static $props = array();

 public function _ _construct() {}
 public function _ _get($name)
 {
  if(array_key_exists($name, self::$props)) {
   return self::$props[$name];
  }
 }
 public function _ _set($name, $value)
 {
  self::$props[$name] = $value;
 }
}

$a = new Singleton;
$b = new Singleton;
$a->property = "hello world";
print $b->property;

In this example, the class stores all its property values in a static array. When a property is accessed for reading or writing, the _ _get and _ _set access handlers look into the static class array instead of inside the object’s internal property table.

Personally, I have no aversion to factory methods, so I prefer to use them. Singletons are relatively rare in an application and so having to instantiate them in a special manner (via their factory) reinforces that they are different. Plus, by using the private constructor, you can prevent rogue instantiations of new members of the class.

Chapter 6, “Unit Testing,” uses a factory method to create a pseudo-singleton where a class has only one global instance per unique parameter.

Please check back next week for the conclusion of this article.

[gp-comments width="770" linklove="off" ]
antalya escort bayan antalya escort bayan