Dealing With Legacy PHP Applications

Dealing with Legacy PHP ApplicationsPrepared by Clinton R. Nixon, Viget Labs2007 July

Dealing with Legacy PHP Applications2007 Jul

What is a legacy application?Code you didn't writeCode you wouldn't writeUntested code

Code with competing visions


What do we do with legacy code?We refactor!

Refactoring is safely changing the implementation of code without changing the behavior of code.


Bad code smellsWhat are some specific problems in legacy

PHP code?‣ No separation between PHP and HTML‣ Lots of requires, few method calls‣ Global variables


No separation between PHP and HTML<h1>Orders</h1>

<?php

$account = new Account($account_id);

$account->loadOrders();

foreach ($account->getOrders() as $order) {

echo '<h2>' . $order['id'] . '</h2>';

echo '<p>Status: ' . lookup_status($order['status_id']) . '<br />;

echo 'Total: ';

$total = array_reduce($order['purchases'],

create_function('$a, $b', '$a += $b; return $a'));

echo $total . '</p>';

}

?>


Separating controllers and views Even without a solid MVC architecture, this helpsYou can do this in several safe and easy stepsYou absolutely will find pain points


Why do I need to do this?Your code complexity will decrease.

echo isn't as fun as it looks.

You will find hidden bugs and mistakes.


The simplest view classclass View {

protected static $VIEW_PATH = '/wherever/views/';

public function assign($name, $value) {

return $this->$name = $value;

}

public function render($filename) {

$filename = self::$VIEW_PATH . $filename;

if (is_file($filename)) {

ob_start();

include($filename);

return ob_get_clean();

}

}

}


Obvious improvements to makeError handlingAssignment by referenceChanging view path

Display convenience methodUse-specific subclasses with helper methods


The separation processGather all your code

Sift and separate controller from view code

Assign variables to the view object

Change all variable references in the view code

Split the files

Find duplicated views


The rules of view codeAllowed:‣Control structures‣ echo, or <?= $var ?>‣Display-specific functions, never nested

Not allowed:‣Assignment‣Other function calls


Gather and sift codeThe step you won't like: gather all code for this

controllerWipe browDraw a line at the top of the code

Move controller code above this line, fixing as necessary‣At this point, everything is view code


Code gathered<?php // View code goes below here ?>

<h1>Orders</h1>

<?php



foreach ($account->getOrders() as $order) {

echo '<h2>' . $order['id'] . '</h2>';

echo '<p>Status: ' . lookup_status($order['status_id']) .

'<br />';

echo 'Total: ';

$total = array_reduce($order['purchases'],

create_function('$a, $b', '$a += $b; return $a'));

echo $total . '</p>';

}


Some controller code moved<?php



?>

<?php // View code goes below here ?>

<h1>Orders</h1>

<?php foreach ($account->getOrders() as $order) { ?>

<h2><?= $order['id'] ?></h2>

<p>Status: <?= lookup_status($order['status_id']) ?>

<br />

Total: <?= array_reduce($order['purchases'], create_function(

'$a, $b', '$a += $b; return $a')) ?>

</p>

<?php } ?>


Alternative control structures<?php if ($foo): ?>...<?php endif; ?>

<?php foreach ($this as $that): ?>...<?php endforeach; ?>


Using alternative control structures<?php



?>


<h1>Orders</h1>

<?php foreach ($account->getOrders() as $order): ?>

<h2><?= $order['id'] ?></h2>

<p>Status: <?= lookup_status($order['status_id']) ?>

<br />

Total: <?= array_reduce($order['purchases'], create_function(

'$a, $b', '$a += $b; return $a')) ?>

</p>

<?php endforeach; ?>


A frustrating problem<?php foreach ($account->getOrders() as $order): ?>

<h2><?= $order['id'] ?></h2>

<p>Status: <?= lookup_status(

$order['status_id']) ?>

<br />

Total: <?= array_reduce($order['purchases'],

create_function('$a, $b', '$a += $b; return $a')) ?>

</p>



Dealing with this problemThere are two approaches.‣You can create a new array of variables for

your view.‣Or, you can encapsulate this logic in an object.


Our new order object<?php

class Order {

...

public function getStatus() {

return lookup_status($this->getStatusId());

}

public function getTotal() {

return array_reduce($this->getPurchases(),create_function('$a, $b', '$a += $b; return $a'));

}

}

?>


Logic removed from view code<?php



$orders = $account->getOrders();

?>


<h1>Orders</h1>

<?php foreach ($orders as $order): ?>

<h2><?= $order->getId() ?></h2>

<p>Status: <?= $order->getStatus() ?>

<br />

Total: <?= $order->getTotal() ?>

</p>



Change all variables to view object variables

Assign variables to the view object:$view->assign('foo', $foo);

One-by-one, change variables in view code.Test to convince yourself.You will probably iterate back to the

previous step.

Document inputs to the view.


View object created<?php




$view = new View();

$view->assign('orders', $orders);

?>


<h1>Orders</h1>

<?php foreach ($view->orders as $order): ?>

<h2><?= $order->getId() ?></h2>


<br />


</p>



Separate the filesCreate a new file for the view code.

Important! Search and replace $view with $this.Test one more time.


Our two files<?php




$view = new View();

$view->assign('orders', $orders);

$view->display('orders.tpl');

?>

<h1>Orders</h1>

<?php foreach ($view->orders as $order): ?>

<h2><?= $order->getId() ?></h2>


<br />


</p>



Find duplicated viewsAs you do this to multiple controllers, you will see

repetition.There will probably be subtle differences.Take the time to re-work these so you can re-use

view files.

Note! You can include views in other views with:$this->render('included_file.tpl');


Using nested requires instead of function calls<?php

require_once('db_setup_inc.php');

require_once('account_auth_inc.php');

require_once('i18n_inc.php');

echo '<h1>Orders for account #' . $account_id .

'</h1>';

require('get_all_orders_inc.php');

...


Untangling a require webRequire statements which call other require statements.

Can be very complex.

Dependent on application structure.


Reasons to untangle this webRemove unneeded complexity.

Create less procedural code.

Prior to PHP 5.2, require_once and include_once are more expensive than you would think.

If you are requiring class definitions, and you have a standard file naming method, use __autoload().


The untangling processIdentify inputsIdentify outputsWrap the file in a method

Refactor methodMove method to correct location


Identify inputs and outputsFind all variables expected to be set before this

file is included.One possible way: execute this file by itself.Find all variables expected to be set or mutated

by this file.

Set variables are easy: comment out the require and watch the errors.

Mutated is the set of inputs changed. Learn to search for these!


account_auth_inc.php<?php

$auth_token = $_COOKIE['token'];

if ($auth_token) {

$acct_id = $db->GetOne('SELECT acct_id FROM

logins WHERE auth_token = ?', array($auth_token));

}

if ($acct_id) {

$acct = new Account($acct_id);

} else {

$acct = null;

}

$_COOKIE['token'] = gen_new_token($auth_token);


Wrap the file in a functionWrap the entire include in a function.

Pass all input variables.

Return all output variables as an array.

And then, call that function at the bottom of the required file!

This is a mess!


Function-wrapped<?php

function account_auth($db, $auth_token) {

if ($auth_token) {

$acct_id = $db->GetOne('SELECT acct_id FROM

logins WHERE auth_token = ?',

array($auth_token));

}

if ($acct_id) {

$acct = new Account($acct_id);

} else {

$acct = null;

}

return array($acct, gen_new_token($auth_token));

}

list($acct, $_COOKIE['token']) = account_auth($db, $_COOKIE['token']);


Refactor until completeTease out the functions, or objects, inside this

function.If you are returning a lot of data, see if it can be an

object.Leave your temporary big function in place, so that

your outside code doesn't break. Keep updating it to deal with your refactoring.


Moved token handling to Account<?php

function account_auth($db, $auth_token) {

// Instead of null, we now return an unloaded Account.

$acct = new Account();

if ($auth_token) {

// SQL code from before

$acct->loadFromToken($auth_token);

// Token generation and cookie setting

$acct->genNewToken($auth_token);

}

return $acct;

}

$acct = account_auth($db, $_COOKIE['token']);


Move to correct locationFinally!Figure out where these functions or objects should

live in your application.Move them there.

Find where the require is called throughout your application, and replace that with your new function call or object method.


Global variables everywhere<?php

$account_id = $_POST['acct_id'];


function getPurchases() {

global $account;

global $database;

...

}

function getLanguage() {

global $account;

global $database;

global $i18n;

...

}


Removing globals one by oneCommon globals:‣ $_POST and $_GET‣Session or cookie data‣Database handles‣User account‣ Language


Do you still have register_globals on?You may have heard: this is a bad idea.You may think that it will be impossible to fix.It's not. Turn on E_ALL.

Spider your site and grep for uninitialized variables.

It's some work, but not as hard as you think. It's worth it.


$_POST and $_GETThese aren't horrible.But not horrible isn't a very high standard.

class InputVariable {public function __construct($name) {...}public function isSet() {...}public function isGet() {...}public function isPost() {...}public function getAsString() {...}public function getAsInt() {...}...

}


The database global objectVery common in PHP codeAgain, not horriblePrevents testing

Prevents multiple databases


Parameterizing the DB handleDoes it need to be everywhere?Can you pass it in to a function or to a

constructor?The process is simple.

‣Add database parameter.

‣Pass in that global variable.

‣ If the call is not in global scope, find out how to pass in that variable to the current scope.

‣Repeat.


Parameterizing globals<?php

$account_id = $_POST['acct_id'];

$account = new Account($database, $account_id);

function getPurchases($account) {

global $account;

global $database;

...

}

function getLanguage($account, $i18n) {

global $account;

global $database;

global $i18n;

...

}


Maybe it does have to be everywhereUse a singleton.But not really.Make a way to change the singleton

instance.‣Global define or environment variable.‣Static mutator.


A quick recapWhat are some specific problems in legacy

PHP code?‣Mixed PHP and HTML – confusion between

controller and view‣Use of require statements instead of function

calls‣Unnecessary global variables causing

dependencies


Further readingWorking Effectively With Legacy Code,

Michael FeathersRefactoring, Martin Fowler


[email protected]

Slides available at http://www.slideshare.net/viget and http://clintonrnixon.net.

Dealing With Legacy PHP Applications

Business

legacy php applications

legacy php code

legacy code

legacy application

theimplementation of

view codeorders status

bad code smellswhat

codemove controller