Category Archives: PHP

OOP vs Procedural Code

After being the part of long discussion this evening, I felt that I should write a post about the real differences between OOP and Procedural coding styles. Hint: whether you use classes and objects or not has very little to do with the answer…

Procedural Programming

Wikipedia defines procedural programming as:

Procedural programming can sometimes be used as a synonym for imperative programming (specifying the steps the program must take to reach the desired state), but can also refer (as in this article) to a programming paradigm, derived from structured programming, based upon the concept of the procedure call.

That’s a decent definition, but let’s see if we can improve upon it. I’m going to assert here that procedural programming is really just the act of specifying a set of ordered steps needed to implement the requested functionality. How those steps are implemented is a detail that’s not related to the paradigm. The important thing is that it’s imperative in how it works. Let’s look at a few examples:

Obviously procedural:

$m = mysqli_connect(...);
$res = mysqli_query($m, $query);
$results = array();
while ($row = mysqli_fetch_assoc($res)) {
    $results[] = $row;
}

This is also procedural, even though it uses an object:

$m = new MySQLi(...);
$res = $m->query($query);
$results = array();
while ($row = $m->fetch_assoc($res)) {
    $results[] = $row;
}

This is still procedural, even though it uses a class:

class GetResults {
    public function getResults() {
        $m = new MySQLi(...);
        $res = $m->query($query);
        $results = array();
        while ($row = $m->fetch_assoc($res)) {
            $results[] = $row;
        }
        return $results;
    }
}

Note that all three of those examples use the exact same code structure. The only difference between them is the way the routines are resolved. But each is procedural. Each has discrete steps that must be taken. Let’s look at what OOP is and why this is different…

Object Oriented Programming

Wikipedia defines object oriented programming as:

Object-oriented programming (OOP) is a programming paradigm using “objects” – data structures consisting of data fields and methods together with their interactions – to design applications and computer programs. Programming techniques may include features such as data abstractionencapsulationmessagingmodularitypolymorphism, and inheritance.

Again, that’s a decent definition. But I only agree with the second part. The first sentence says that you must use object data structures to write OOP. That’s blatantly wrong. You can completely implement data abstraction, encapsulation, messaging, modularity, polymorphism and (to a limited extent) inheritance without using an object structure. What I’d argue makes code OOP is a few things. First, it must abstract the data concepts into modular units. Second, it must have some way to polymorphically execute code. Finally, it must at least partially encapsulate that code and functionality. Let’s look at a few examples before continuing further:

A classic OOP pattern:

class Mediator {
    protected $events = array();
    public function attach($eventName, $callback) {
        if (!isset($this->events[$eventName])) {
            $this->events[$eventName] = array();
        }
        $this->events[$eventName][] = $callback;
    }
    public function trigger($eventName, $data = null) {
        foreach ($this->events[$eventName] as $callback) {
            $callback($eventName, $data);
        }
    }
}
$mediator = new Mediator;
$mediator->attach('load', function() { echo "Loading"; });
$mediator->attach('stop', function() { echo "Stopping"; });
$mediator->attach('stop', function() { echo "Stopped"; });
$mediator->trigger('load'); // prints "Loading"
$mediator->trigger('stop'); // prints "StoppingStopped"

The same pattern, using functions.

$hooks = array();
function hook_register($eventName, $callback) {
    if (!isset($GLOBALS['hooks'][$eventName])) {
        $GLOBALS['hooks'][$eventName] = array();
    }
    $GLOBALS['hooks'][$eventName][] = $callback;
}
function hook_trigger($eventName, $data = null) {
    foreach ($GLOBALS['hooks'][$eventName] as $callback) {
        $callback($eventName, $data);
    }
}

As you can see, both follow the Mediator Pattern. Both are object oriented, because they both are designed to de-couple caller from sender. Both provide state, and are modular. The difference here, is that one is implemented using a traditional object (and is hence reusable, a very good advantage) and the other is not reusable since it depends on a global variable. I used the term “hook” here for a very important reason. It’s the name of the event system that Drupal uses.

Drupal in a lot of ways is very object oriented. Their module system, their hook system, their form system, etc are all object oriented. But none of them use objects for that. They use functions and dynamic dispatch. This leads to some really awkward tradeoffs, so I’m not suggesting that it’s a good thing, just that it’s a proof that you don’t need classes to write OOP.

Why does it matter?

It matters for a very simple reason. A lot of developers think that just because they use classes, they are writing OOP. And others think that because they use functions, they are using procedural programming. And that’s not true. Procedural vs OOP is an approach to writing code, not how you write it. Are you focusing on “Steps” and an ordered way of writing a program? You’re likely writing procedural code. But if you’re focusing on state transformations and encapsulated abstractions, you’re writing OOP.

Classes are just a tool that make writing real OOP easier. They aren’t a requirement or a indicator that you’re writing OOP.

Just My $0.02…

Update: OOP Database Access

So, some of you are asking what database access would look like in OOP code. The reason that I didn’t include an example is that it’s all abstracted away. In reality, the way that I would do that query could be:

$mapper = new PersonDataMapper(new MySQLi(...));
$people = $mapper->getAll();

Where $people is an array of person objects. Note that it’s a responsibility that’s abstracted away. So in your business objects you’d never access the database directly. You’d use a mapper to translate back and forth from your business objects to the data store. Internally, a specific mapper will build a query, execute it, and fetch results, but that’s all abstracted away. We can change the implementation detail of the database layer by simply swapping out a mapper.

The responsibility of data persistence becomes an encapsulated abstraction. And that’s why it’s not procedural but object oriented…

'Coz sharing is caring

Coding conventions: PHP

Code structure

Assignment expressions

Using assignment as an expression is surprising to the reader and looks like an error. Do not write code like this:

if ( $a = foo() ) {
    bar();
}

Space is cheap, and you’re a fast typist, so instead use:

$a = foo();
if ( $a ) {
    bar();
}

Using assignment in a while() clause used to be legitimate, for iteration:

$res = $dbr->query( 'SELECT * FROM some_table' );
while ( $row = $dbr->fetchObject( $res ) ) {
    showRow( $row );
}

This is unnecessary in new code; instead use:

$res = $dbr->query( 'SELECT * FROM some_table' );
foreach ( $res as $row ) {
    showRow( $row );
}

Spaces

MediaWiki favors a heavily-spaced style for optimum readability.

Put spaces on either side of binary operators, for example:

// No:
$a=$b+$c;

// Yes:
$a = $b + $c;

Put spaces next to parentheses on the inside, except where the parentheses are empty. Do not put a space following a function name.

$a = getFoo( $b );
$c = getBar();

Opinions differ as to whether control structures  if, while, for, foreach etc. should be followed by a space; the following two styles are acceptable:

// Spacey
if ( isFoo() ) {
        $a = 'foo';
}

// Not so spacey
if( isFoo() ) {
        $a = 'foo';
}

In comments there should be one space between the # or // character and the comment, and a comment should be put on its own line.

// No:
        public static function getFoo( $bar ) {
                if ( $bar !== false ) { //because this and that..
                        return $bar; //already defined, return it
                }
        }

// Yes:
        public static function getFoo( $bar ) {
                // Because this and that..
                if ( $bar !== false ) {
                        // Already defined, return it.
                        return $bar;
                }
        }

To help developers fix code with an inadequately spacey style, a tool called stylize.php has been created, which uses PHP’s tokenizer extension to enforce most whitespace conventions automatically.

Ternary operator

The ternary operator can be used profitably if the expressions are very short and obvious:

$swat = isset( $this->mParams['swat'] ) ? $this->mParams['swat'] : false;

But if you’re considering a multi-line expression with a ternary operator, please consider using an if() block instead. Remember, disk space is cheap, code readability is everything, “if” is English and ?: is not.

PHP-v5.3 shorthand

Since we still support PHP 5.2.x, use of the shorthand ternary operator (?:) introduced in PHP 5.3 is not allowed.

String literals

For simple string literals, single quotes are slightly faster for PHP to parse than double quotes. Perhaps more importantly, they are easier to type, since you don’t have to press shift. For these reasons, single quotes are preferred in cases where they are equivalent to double quotes.

However, do not be afraid of using PHP’s double-quoted string interpolation feature: $elementId = “myextension-$index”; This has slightly better performance characteristics than the equivalent using the concatenation (dot) operator, and it looks nicer too.

Heredoc-style strings are sometimes useful:

$s = <<<EOT
<div class="mw-some-class">
$boxContents
</div>
EOT;

Some authors like to use END as the ending token, which is also the name of a PHP function. This leads to IRC conversations like the following:

<Simetrical>      vim also has ridiculously good syntax highlighting.
<TimStarling>     it breaks when you write <<<END in PHP
<Simetrical>      TimStarling, but if you write <<<HTML it syntax-highlights as HTML!
<TimStarling>     I have to keep changing it to ENDS so it looks like a string again
<brion-codereview>        fix the bug in vim then!
<TimStarling>     brion-codereview: have you ever edited a vim syntax script file?
<brion-codereview>        hehehe
<TimStarling>     http://tstarling.com/stuff/php.vim
<TimStarling>     that's half of it...
<TimStarling>     here's the other half: http://tstarling.com/stuff/php-syntax.vim
<TimStarling>     1300 lines of sparsely-commented code in a vim-specific language
<TimStarling>     which turns out to depend for its operation on all kinds of subtle inter-pass effects
<werdnum> TimStarling: it looks like some franken-basic language.

Functions and parameters

Avoid passing huge numbers of parameters to functions or constructors:

//Constructor for Block.php as of 1.17. *DON'T* do this!
function __construct( $address = '', $user = 0, $by = 0, $reason = '',
        $timestamp = 0, $auto = 0, $expiry = '', $anonOnly = 0, $createAccount = 0, $enableAutoblock = 0,
        $hideName = 0, $blockEmail = 0, $allowUsertalk = 0 )
{
        ...
}

It quickly becomes impossible to remember the order of parameters, and you will inevitably end up having to hardcode all the defaults in callers just to customise a parameter at the end of the list. If you are tempted to code a function like this, consider passing an associative array of named parameters instead.

In general, using boolean parameters is discouraged in functions. In $object->getSomething( $input, true, true, false ), without looking up the documentation for MyClass::getSomething(), it is impossible to know what those parameters are meant to indicate. Much better is to either use class constants, and make a generic flag parameter:

$myResult = MyClass::getSomething( $input, MyClass::FROM_DB & MyClass::PUBLIC_ONLY );

Or to make your function accept an array of named parameters:

$myResult = MyClass::getSomething( $input, array( 'fromDB', 'publicOnly' ) );

Try not to repurpose variables over the course of a function, and avoid modifying the parameters passed to a function (unless they’re passed by reference and that’s the whole point of the function, obviously).

C borrowings

The PHP language was designed by people who love C and wanted to bring souvenirs from that language into PHP. But PHP has some important differences from C.

In C, constants are implemented as preprocessor macros and are fast. In PHP, they are implemented by doing a runtime hashtable lookup for the constant name, and are slower than just using a string literal. In most places where you would use an enum or enum-like set of macros in C, you can use string literals in PHP.

PHP has three special literals: true, false and null. Homesick C developers write null as NULL because they want to believe that it is a macro defined as ((void*)0). This is not necessary.

Use elseif not else if. They have subtly different meanings:

// This:
if( $foo == 'bar' ) {
        echo 'Hello world';
} else if( $foo == 'Bar' ) {
        echo 'Hello world';
} else if( $baz == $foo ) {
        echo 'Hello baz';
} else {
        echo 'Eh?';
}

// Is actually equivalent to:
if( $foo == 'bar' ) {
        echo 'Hello world';
} else {
        if( $foo == 'Bar' ) {
                echo 'Hello world';
        } else  {
                if( $baz == $foo ) {
                        echo 'Hello baz';
                } else {
                        echo 'Eh?';
                }
        }
}

And the latter has poorer performance.

Naming

Use lowerCamelCase when naming functions or variables. For example:

private function doSomething( $userPrefs, $editSummary )

Use UpperCamelCase when naming classes: class ImportantClass. Use uppercase with underscores for global and class constants: DB_MASTER, Revision::REV_DELETED_TEXT. Other variables are usually lowercase or lowerCamelCase; avoid using underscores in variable names.

There are also some prefixes used in different places:

Functions

  • wf (wiki functions) – top-level functions, e.g.
function wfFuncname() { ... }

Verb phrases are preferred: use getReturnText() instead of returnText().

Variables

  • $wg – global variables, e.g. $wgVersion, $wgTitle. Always use this for new globals, so that it’s easy to spot missing “global $wgFoo” declarations. In extensions, the extension name should be used as a namespace delimiter. For example, $wgAbuseFilterConditionLimit, not $wgConditionLimit.

It is common to work with an instance of the Database class; we have a naming convention for these which helps keep track of the nature of the server to which we are connected. This is of particular importance in replicated environments, such as Wikimedia and other large wikis; in development environments there is usually no difference between the two types, which can conceal subtle errors.

  • $dbw – a Database object for writing (a master connection)
  • $dbr – a Database object for non-concurrency-sensitive reading (this may be a read-only slave, slightly behind master state, so don’t ever try to write to the database with it, or get an “authoritative” answer to important queries like permissions and block status)

The following may be seen in old code but are discouraged in new code:

  • $ws – Session variables, e.g. $_SESSION[‘wsSessionName’]
  • $wc – Cookie variables, e.g. $_COOKIE[‘wcCookieName’]
  • $wp – Post variables (submitted via form fields), e.g. $wgRequest->getText( ‘wpLoginName’ )
  • $m – object member variables: $this->mPage. This is discouraged in new code, but try to stay consistent within a class.

Pitfalls

  • Understand and read the documentation for isset() and empty(). Use them only when appropriate.
    • empty() is inverted conversion to boolean with error suppression. Only use it when you really want to suppress errors. Otherwise just use !. Do not use it to test if an array is empty, unless you simultaneously want to check if the variable is unset.
    • Do not use isset() to test for null. Using isset() in this situation could introduce errors by hiding mis-spelled variable names. Instead, use $var === null
  • Study the rules for conversion to boolean. Be careful when converting strings to boolean.
  • Be careful with double-equals comparison operators. Triple-equals is often more intuitive.
    • ‘foo’ == 0 is true
    • ‘000’ == ‘0’ is true
    • ‘000’ === ‘0’ is false
  • Array plus does not renumber the keys of numerically-indexed arrays, so array(‘a’) + array(‘b’) === array(‘a’). If you want keys to be renumbered, use array_merge(): array_merge( array( ‘a’ ), array( ‘b’ ) ) == array( ‘a’, ‘b’ )
  • Make sure you have error_reporting set to E_ALL for PHP 5. This will notify you of undefined variables and other subtle gotchas that stock PHP will ignore. See also Manual:How to debug.
  • When working in a pure PHP environment, remove any trailing ?> tags. These tags often cause issues with trailing white-space and “headers already sent” error messages (cf. bugzilla:17642 and http://news.php.net/php.general/280796).
  • Do not use the ‘goto’ syntax introduced in 5.3. PHP may have introduced the feature, but that does not mean we should use it.

Comments and Documentation

The Doxygen documentation style is used (it is very similar to PHPDoc for the subset that we use). A code documentation example: giving a description of a function or method, the parameters it takes (using @param), and what the function returns (using @return), or the @ingroup or @author tags.

Use @ rather than \ as the escape character (i.e. use @param rather than \param) – both styles work in Doxygen, but for backwards and future compatibility MediaWiki uses has chosen the @param style as convention).

Use /** to begin the comments, instead of the Qt-style formatting /*!.

General format for parameters is such: @param type $varname: description. Multiple types can be listed by separating with a pipe character.

Doxygen documentation states that @param should have the same format as phpDocumentor:

@param  datatype1|datatype2 $paramname description

For every public interface (method, class, variable, whatever) you add or change, a @since tag should be provided, so people extending the code via this interface know they are breaking compatibility with older versions of the code.

class Foo {

        /**
         * @var array $bar: Description here
         * @example array( 'foo' => Bar, 'quux' => Bar, .. )
         */
        protected $bar;

        /**
         * Short decription here, following by documentation of the parameters.
         *
         * @since 1.42
         *
         * @param FooContext $context
         * @param array|string $options: Optionally pass extra options. Either a string or an array of strings.
         * @return Foo|null: New instance of Foo or null of quuxification failed.
         *
         * Some example:
         * @code
         * ...
         * @endcode
         */
        public function makeQuuxificatedFoo( FooContext $context = null, $options = array() ) {
                /* .. */
        }

}

PHPDoc was used at the very beginning but got replaced with Doxygen for performance reason. We should probably drop PHPDoc compatibility.

@var: documenting class members

There is a ‘bug’ in Doxygen which affects MediaWiki’s documentation: using @var to specify the class members’ type only works if the variable name is appended:

       /**
         * Some explanation about the variable
         *
         * @var string $msg
         */
        protected $msg;

If you don’t append the variable name Doxygen will ignore the entire comment block and it will not be included in the docs.

Integration

There are a few pieces of code in the MediaWiki codebase which are intended to be standalone and easily portable to other applications; examples include the UTF normalisation in /includes/normal and the libraries in /includes/libs. Apart from these, code should be integrated into the rest of the MediaWiki environment, and should allow other areas of the codebase to integrate with it in return.

Global objects

Do not access the PHP superglobals $_GET, $_POST, etc, directly; use $request->get*( ‘param’ ) instead; there are various functions depending on what type of value you want. You can get a WebRequest from the nearest RequestContext, or if absolutely necessary $wgRequest. Equally, do not access $_SERVER directly; use $request->getIP() if you want to get the IP address of the current user.

Static methods and properties

Static methods and properties are useful for programmers because they act like globals without polluting the global namespace. However, they make subclassing and reuse more difficult for other developers. Generally, you should avoid introducing static functions and properties when you can, especially if the sole purpose is to just save typing.

For example, lots of developers would prefer to write something like:

Foo::bar();

This is because it is shorter and takes less keystrokes. However, by doing this you’ve made the Foo class much harder to subclass and reuse. Instead of introducing a static method, you could just type:

$f = new Foo();
$f->bar();

Remember, shorter does not always mean better, and you should take the time to design your classes in a way that makes them easy to reuse.

Late static binding

In PHP 5.3, a new feature called “Late Static Binding” (LSB) was added to help work around this perceived lack of functionality in static functions. However, the usefulness of LSB is debatable among MediaWiki developers and should be avoided for the time being.

Classes

Encapsulate your code in an object-oriented class, or add functionality to existing classes; do not add new global functions or variables. Try to be mindful of the distinction between ‘backend’ classes, which represent entities in the database (eg User, Block, Revision, etc), and ‘frontend’ classes, which represent pages or interfaces visible to the user (SpecialPage, Article, ChangesList, etc. Even if your code is not obviously object-oriented, you can put it in a static class (eg IP or Html).

As a holdover from PHP 4’s lack of private class members and methods, older code will be marked with comments such as /** @private */ to indicate the intention; respect this as if it were enforced by the interpreter.

Mark new code with proper visibility modifiers, including public if appropriate, but do not add visibility to existing code without first checking, testing and refactoring as required. It’s generally a good idea to avoid visibility changes unless you’re making changes to the function which would break old uses of it anyway.

Error handling

Don’t suppress errors with PHP’s @ operator, for any reason ever. It’s broken when E_STRICT is enabled and it causes an unlogged, unexplained error if there is a fatal, which is hard to support. Use wfSuppressWarnings() and wfRestoreWarnings() instead. The checkSyntax.php maintenance script can check for this error for you.

When your code encounters a sudden error, you should throw a MWException (or an appropriate subclass) rather than using PHP’s trigger_error. The exception handler will display this as nicely as possible to the end user and wiki administrator, and also provides a stack trace to developers.

'Coz sharing is caring