In the first part of this series, we learned the key concepts to build an application that is as CMS-agnostic as possible. In this second and final part, we will proceed to abstract a WordPress application, making its code ready to be used with Symfony components, Laravel framework, and October CMS (which is based on Laravel).

Accessing Services

Before we start abstracting the code, we need to provide the layer of dependency injection to the application. As described in the first part of this series, this layer is satisfied through Symfony’s DependencyInjection component. To access the defined services, we create a class ContainerBuilderFactory which simply stores a static instance of the component’s ContainerBuilder object:

use SymfonyComponentDependencyInjectionContainerBuilder;

class ContainerBuilderFactory {
  private static $instance;
  public static function init()
  {
    self::$instance = new ContainerBuilder();
  }
  public static function getInstance()
  {
    return self::$instance;
  }
}

Then, to access a service called "cache", the application requests it like this:

$cacheService = ContainerBuilderFactory::getInstance()->get('cache');
// Do something with the service
// $cacheService->...

Abstracting WordPress Code

We have identified the following pieces of code and concepts from a WordPress application that need be abstracted away from WordPress’s opinionatedness:

• accessing functions
• function names
• function parameters
• states (and other constant values)
• CMS helper functions
• user permissions
• application options
• database column names
• errors
• hooks
• routing
• object properties
• global state
• entity models (meta, post types, pages being posts, and taxonomies —tags and categories—)
• translation
• media.

Let’s proceed to abstract them, one by one.

Note: For ease of reading, I have omitted adding namespaces to all classes and interfaces throughout this article. However, adding namespaces, as specified in PHP Standards Recommendation PSR-4, is a must! Among other advantages, the application can then benefit from autoloading, and Symfony’s dependency injection can rely on automatic service loading as to reduce its configuration to the bare minimum.

Accessing functions

The mantra “code against interfaces, not implementations” means that all those functions provided by the CMS cannot be accessed directly anymore. Instead, we must access the function from a contract (an interface), on which the CMS function will simply be the implementation. By the end of the abstraction, since no WordPress code will be referenced directly anymore, we can then swap WordPress with a different CMS.

For instance, if our application accesses function get_posts:

$posts = get_posts($args);

We must then abstract this function under some contract:

interface PostAPIInterface
{
  public function getPosts($args);
}

The contract must be implemented for WordPress:

class WPPostAPI implements PostAPIInterface
{
  public function getPosts($args) {
    return get_posts($args);
  }
}

A service "posts_api" must be added to the dependency injection services.yaml configuration file, indicating which class resolves the service:

services:
  posts_api:
    class: WPPostAPI

And finally, the application can reference the function through service "posts_api":

$postsAPIService = ContainerBuilderFactory::getInstance()->get('posts_api');
$posts = $postsAPIService->getPosts($args);

Function names

If you have noticed from the code demonstrated above, function get_posts is abstracted as getPosts. There are a couple of reasons why this is a good idea:

• By calling the function differently, it helps identify which code belongs to WordPress and which code belongs to our abstracted application.
• Function names must be camelCased to comply with PSR-2, which attempts to define a standard for writing PHP code.

Certain functions can be redefined, making more sense in an abstract context. For instance, WordPress function get_user_by($field, $value) uses parameter $field with values "id", "ID", "slug", "email" or "login" to know how to get the user. Instead of replicating this methodology, we can explicitly define a separate function for each of them:

interface UsersAPIInterface
{
  public function getUserById($value);
  public function getUserByEmail($value);
  public function getUserBySlug($value);
  public function getUserByLogin($value);
}

And these are resolved for WordPress:

class WPUsersAPI implements UsersAPIInterface
{
  public function getUserById($value)
  {
    return get_user_by('id', $value);
  }
  public function getUserByEmail($value)
  {
    return get_user_by('email', $value);
  }
  public function getUserBySlug($value)
  {
    return get_user_by('slug', $value);
  }
  public function getUserByLogin($value)
  {
    return get_user_by('login', $value);
  }
}

Certain other functions should be renamed because their names convey information about their implementation, which may not apply for a different CMS. For instance, WordPress function get_the_author_meta can receive parameter "user_lastname", indicating that the user’s lastname is stored as a “meta” value (which is defined as an additional property for an object, not originally mapped in the database model). However, other CMSs may have a column "lastname" in the user table, so it doesn’t apply as a meta value. (The actual definition of “meta” value is actually inconsistent in WordPress: function get_the_author_meta also accepts value "user_email", even though the email is stored on the user table. Hence, I’d rather stick to my definition of “meta” value, and remove all inconsistencies from the abstracted code.)

Then, our contract will implement the following functions:

interface UsersAPIInterface
{
  public function getUserDisplayName($user_id);
  public function getUserEmail($user_id);
  public function getUserFirstname($user_id);
  public function getUserLastname($user_id);
  ...
}

Which are resolved for WordPress:

class WPUsersAPI implements UsersAPIInterface
{
  public function getUserDisplayName($user_id)
  {
    return get_the_author_meta('display_name', $user_id);
  }
  public function getUserEmail($user_id)
  {
    return get_the_author_meta('user_email', $user_id);
  }
  public function getUserFirstname($user_id)
  {
    return get_the_author_meta('user_firstname', $user_id);
  }
  public function getUserLastname($user_id)
  {
    return get_the_author_meta('user_lastname', $user_id);
  }
  ...
}

Our functions could also be re-defined as to remove the limitations from WordPress. For instance, function update_user_meta($user_id, $meta_key, $meta_value) can receive one meta attribute at a time, which makes sense since each of these is updated on its own database query. However, October CMS maps all meta attributes together on a single database column, so it makes more sense to update all values together on a single database operation. Then, our contract can include an operation updateUserMetaAttributes($user_id, $meta) which can update several meta values at the same time:

interface UserMetaInterface
{
  public function updateUserMetaAttributes($user_id, $meta);
}

Which is resolved for WordPress like this:

class WPUsersAPI implements UsersAPIInterface
{
  public function updateUserMetaAttributes($user_id, $meta)
  {
    foreach ($meta as $meta_key => $meta_value) {
      update_user_meta($user_id, $meta_key, $meta_value);
    }
  }
}

Finally, we may want to re-define a function to remove its ambiguities. For instance, WordPress function add_query_arg can receive parameters in two different ways:

1. Using a single key and value: add_query_arg('key', 'value', 'http://example.com');
2. Using an associative array: add_query_arg(['key1' => 'value1', 'key2' => 'value2'], 'http://example.com');

This becomes difficult to keep consistent across CMSs. Hence, our contract can define functions addQueryArg (singular) and addQueryArgs (plural) as to remove the ambiguity:

public function addQueryArg(string $key, string $value, string $url);
public function addQueryArgs(array $key_values, string $url);

Function parameters

We must also abstract the parameters to the function, making sure they make sense in a generic context. For each function to abstract, we must consider:

• renaming and/or re-defining the parameters;
• renaming and/or re-defining the attributes passed on array parameters.

For instance, WordPress function get_posts receives a unique parameter $args, which is an array of attributes. One of its attributes is fields which, when given the value "ids", makes the function return an array of IDs instead of an array of objects. However, I deem this implementation too specific for WordPress, and for a generic context I’d prefer a different solution: Convey this information through a separate parameter called $options, under attribute "return-type".

To accomplish this, we add parameter $options to the function in our contract:

interface PostAPIInterface
{
  public function getPosts($args, $options = []);
}

Instead of referencing WordPress constant value "ids" (which we can’t guarantee will be the one used in all other CMSs), we create a corresponding constant value for our abstracted application:

class Constants
{
  const RETURNTYPE_IDS = 'ids';
}

The WordPress implementation must map and recreate the parameters between the contract and the implementation:

class WPPostAPI implements PostAPIInterface
{
  public function getPosts($args, $options = []) {
    if ($options['return-type'] == Constants::RETURNTYPE_IDS) {
      $args['fields'] = 'ids';
    }
    return get_posts($args);
  }
}

And finally, we can execute the code through our contract:

$options = [
  'return-type' => Constants::RETURNTYPE_IDS,
];
$post_ids = $postsAPIService->getPosts($args, $options);

While abstracting the parameters, we should avoid transferring WordPress’s technical debt to our abstracted code, whenever possible. For instance, parameter $args from function get_posts can contain attribute 'post_type'. This attribute name is somewhat misleading, since it can receive one element (post_type => "post") but also a list of them (post_type => "post, event"), so this name should be in plural instead: post_types. When abstracting this piece of code, we can set our interface to expect attribute post_types instead, which will be mapped to WordPress’s post_type.

Similarly, different functions accept arguments with different names, even though these have the same objective, so their names can be unified. For instance, through parameter $args, WordPress function get_posts accepts attribute posts_per_page, and function get_users accepts attribute number. These attribute names can perfectly be replaced with the more generic attribute name limit.

It is also a good idea to rename parameters to make it easy to understand which ones belong to WordPress and which ones have been abstracted. For instance, we can decide to replace all "_" with "-", so our newly-defined argument post_types becomes post-types.

Applying these prior considerations, our abstracted code will look like this:

class WPPostAPI implements PostAPIInterface
{
  public function getPosts($args, $options = []) {
    ...
    if (isset($args['post-types'])) {
      $args['post_type'] = $args['post-types'];
      unset($args['post-types']);
    }
    if (isset($args['limit'])) { 
      $args['posts_per_page'] = $args['limit'];
      unset($args['limit']);
    }
    return get_posts($args);
  }
}

We can also re-define attributes to modify the shape of their values. For instance, WordPress parameter $args in function get_posts can receive attribute date_query, whose properties ("after", "inclusive", etc) can be considered specific to WordPress:

$date = current_time('timestamp');
$args['date_query'] = array(
  array(
    'after' => date('Y-m-d H:i:s', $date),
    'inclusive' => true,
  )
);

To unify the shape of this value into something more generic, we can re-implement it using other arguments, such as "date-from" and "date-from-inclusive" (this solution is not 100% convincing though, since it is more verbose than WordPress’s):

class WPPostAPI implements PostAPIInterface
{
  public function getPosts($args, $options = []) {
    ...
    if (isset($args['date-from'])) {
      $args['date_args'][] = [
        'after' => $args['date-from'],
        'inclusive' => false,
      ];
      unset($args['date-from']);
    }
    if (isset($args['date-from-inclusive'])) {
      $args['date_args'][] = [
        'after' => $args['date-from-inclusive'],
        'inclusive' => true,
      ];
      unset($args['date-from-inclusive']);
    }
    return get_posts($args);
  }
}

In addition, we need to consider if to abstract or not those parameters which are too specific to WordPress. For instance, function get_posts allows to order posts by attribute menu_order, which I don’t think it works in a generic context. Then, I’d rather not abstract this code and keep it on the CMS-specific package for WordPress.

Finally, we can also add argument types (and, since here we are, also return types) to our contract fuction, making it more understandable and allowing the code to fail in compilation time instead of during runtime:

interface PostAPIInterface
{
  public function getPosts(array $args, array $options = []): array;
}

States (and other constant values)

We need to make sure that all states have the same meaning in all CMSs. For instance, posts in WordPress can have one among the following states: "publish", "pending", "draft" or "trash". To make sure that the application references the abstracted version of the states and not the CMS-specific one, we can simply define a constant value for each of them:

class PostStates {
  const PUBLISHED = 'published';
  const PENDING = 'pending';
  const DRAFT = 'draft';
  const TRASH = 'trash';
}

As it can be seen, the actual constant values may or may not be the same as in WordPress: while "publish" was renamed as "published", the other ones remain the same.

For the implementation for WordPress, we convert from the agnostic value to the WordPress-specific one:

class WPPostAPI implements PostAPIInterface
{
  public function getPosts($args, $options = []) {
    ...
    if (isset($args['post-status'])) {
      $conversion = [
        PostStates::PUBLISHED => 'publish',
        PostStates::PENDING => 'pending',
        PostStates::DRAFT => 'draft',
        PostStates::TRASH => 'trash',
      ];
      $args['post_status'] = $conversion[$args['post-status']];
      unset($args['post-status']);
    }
    return get_posts($args);
  }
}

Finally, we can reference these constants throughout our CMS-agnostic application:

$args = [
  'post-status' => PostStates::PUBLISHED,
];
$posts = $postsAPIService->getPosts($args);

This strategy works under the assumption that all CMSs will support these states. If any CMS does not support a particular state (eg: "pending") then it should throw an exception whenever a corresponding functionality is invoked.

CMS helper functions

WordPress implements several helper functions that must also abstracted, such as make_clickable. Because these functions are very generic, we can implement a default behavior for them that works well in an abstract context, and which can be overridden if the CMS implements a better solution.

We first define the contract:

interface HelperAPIInterface
{
  public function makeClickable(string $text);
}

And provide a default behaviour for the helper functions through an abstract class:

abstract class AbstractHelperAPI implements HelperAPIInterface
{
  public function makeClickable(string $text) {
    return preg_replace('!(((f|ht)tp(s)?://)[-a-zA-Zа-яА-Я()0-9@:%_+.~#?&;//=]+)!i', '<a href="$1">$1</a>', $text);
  }
}

Now, our application can either use this functionality or, if it runs on WordPress, use the WordPress-specific implementation:

class WPHelperAPI extends AbstractHelperAPI
{
  public function makeClickable(string $text) {
    return make_clickable($text);
  }
}

User permissions

For all CMSs which support user management, in addition to abstracting the corresponding functions (such as current_user_can and user_can in WordPress), we must also make sure that the user permissions (or capabilities) have the same effect across all CMSs. To achieve this, our abstracted application needs to explicitly state what is expected from the capability, and the implementation for each CMS must either satisfy it through one of its own capabilities or throw an exception if it can’t satisfy it. For instance, if the application needs to validate if the user can edit posts, it can represent it through a capability called "capability:editPosts", which is satisfied for WordPress through its capability "edit_posts".

This is still an instance of the “code against interfaces, not implementations” principle, however here we run against a problem: Whereas in PHP we can define interfaces and classes to model contracts and service providers (which works in compilation time, so that the code doesn’t compile if a class implementing an interface does not implement all functions defined in the interface), PHP offers no similar construct to validate that a contract capability (which is simply a string, such as "capability:editPosts") has been satisfied through a capability by the CMS. This concept, which I call a “loose contract”, will need to be handled by our application, on runtime.

To deal with “loose contracts”, I have created a service LooseContractService through which:

• the application can define what “contract names” must be implemented, through function requireNames.
• the CMS-specific implementations can satisfy those names, through function implementNames.
• the application can get the implementation of a name through function getImplementedName.
• the application can also inquire for all non-satisfied required names through function getNotImplementedRequiredNames, as to throw an exception or log the error if needed.

The service looks like this:

class LooseContractService
{
  protected $requiredNames = [];
  protected $nameImplementations = [];

  public function requireNames(array $names): void
  {
    $this->requiredNames = array_merge(
      $this->requiredNames,
      $names
    );
  }

  public function implementNames(array $nameImplementations): void
  {
    $this->nameImplementations = array_merge(
      $this->nameImplementations,
      $nameImplementations
    );
  }

  public function getImplementedName(string $name): ?string {
    return $this->nameImplementations[$name];
  }

  public function getNotImplementedRequiredNames(): array {
    return array_diff(
      $this->requiredNames,
      array_keys($this->nameImplementations)
    );
  }
}

The application, when initialized, can then establish loose contracts by requiring names:

$looseContractService = ContainerBuilderFactory::getInstance()->get('loose_contracts');
$looseContractService->requireNames([
  'capability:editPosts',
]);

And the CMS-specific implementation can satisfy these:

$looseContractService->implementNames([
  'capability:editPosts' => 'edit_posts',
]);

The application can then resolve the required name to the implementation from the CMS. If this required name (in this case, a capability) is not implemented, then the application may throw an exception:

$cmsCapabilityName = $looseContractService->getImplementedName('capability:editPosts');
if (!$cmsCapabilityName) {
  throw new Exception(sprintf(
    "The CMS has no support for capability "%s"",
    'capability:editPosts'
  ));
}
// Now can use the capability to check for permissions
$userManagementAPIService = ContainerBuilderFactory::getInstance()->get('user_management_api');
if ($userManagementAPIService->userCan($user_id, $cmsCapabilityName)) {
  ...
}

Alternatively, the application can also fail when first initialized if any one required name is not satisfied:

if ($notImplementedNames = $looseContractService->getNotImplementedRequiredNames()) {
  throw new Exception(sprintf(
    "The CMS has not implemented loose contract names %s",
    implode(', ', $notImplementedNames)
  ));
}

Application options

WordPress ships with several application options, such as those stored in table wp_options under entries "blogname", "blogdescription", "admin_email", "date_format" and many others. Abstracting application options involves:

• abstraction the function getOption;
• abstracting each of the required options, aiming to make the CMS satisfy the notion of this option (eg: if a CMS doesn’t have an option for the site’s description, it can’t return the site’s name instead).

Let’s solve these 2 actions in turn. Concerning function getOption, I believe that we can expect all CMSs to support storing and retrieving options, so we can place the corresponding function under a CMSCoreInterface contract:

interface CMSCoreInterface
{
  public function getOption($option, $default = false);
}

As it can be observed from the function signature above, I’m making the assumption that each option will also have a default value. However, I don’t know if every CMS allows setting default values for options. But it doesn’t matter since the implementation can simply return NULL then.

This function is resolved for WordPress like this:

class WPCMSCore implements CMSCoreInterface
{
  public function getOption($option, $default = false)
  {
    return get_option($option, $default);
  }
}

To solve the 2nd action, which is abstracting each needed option, it is important to notice that even though we can always expect the CMS to support getOption, we can’t expect it to implement each single option used by WordPress, such as "use_smiles" or "default_ping_status". Hence, we must first filter all options, and abstract only those that make sense in a generic context, such as "siteName" or "dateFormat".

Then, having the list of options to abstract, we can use a “loose contract” (as explained earlier on) and require a corresponding option name for each, such as "option:siteName" (resolved for WordPress as "blogname") or "option:dateFormat" (resolved as "date_format").

Database column names

In WordPress, when we are requesting data from function get_posts we can set attribute "orderby" in $args to order the results, which can be based on a column from the posts table (such as values "ID", "title", "date", "comment_count", etc), a meta value (through values "meta_value" and "meta_value_num") or other values (such as "post__in" and "rand").

Whenever the value corresponds to the table column name, we can abstract them using a “loose contract”, as explained earlier on. Then, the application can reference a loose contract name:

$args = [
  'orderby' => $looseContractService->getImplementedName('dbcolumn:orderby:posts:date'),
];
$posts = $postsAPIService->getPosts($args);

And this name is resolved for WordPress:

$looseContractService->implementNames([
  'dbcolumn:orderby:posts:date' => 'date',
]);

Now, let’s say that in our WordPress application we have created a meta value "likes_count" (which stores how many likes a post has) to order posts by popularity, and we want to abstract this functionality too. To order results by some meta property, WordPress expects an additional attribute "meta_key", like this:

$args = [
  'orderby' => 'meta_value',
  'meta_key' => 'likes_count',
];

Because of this additional attribute, I consider this implementation WordPress-specific and very difficult to abstract to make it work everywhere. Then, instead of generalizing this functionality, I can simply expect every CMS to add their own, specific implementation.

Let’s do that. First, I create a helper class to retrieve the CMS-agnostic query:

class QueryHelper
{
  public function getOrderByQuery()
  {
    return array(
      'orderby' => $looseContractService->getImplementedName('dbcolumn:orderby:posts:likesCount'),
    );
  }
}

The OctoberCMS-specific package can add a column "likes_count" to the posts table, and resolve name "dbcolumn:orderby:posts:likesCount" to "like_count" and it will work. The WordPress-specific package, though, must resolve "dbcolumn:orderby:posts:likesCount" as "meta_value" and then override the helper function to add the additional property "meta_key":

class WPQueryHelper extends QueryHelper
{
  public function getOrderByQuery()
  {
    $query = parent::getOrderByQuery();
    $query['meta_key'] = 'likes_count';
    return $query;
  }
}

Finally, we set-up the helper query class as a service in the ContainerBuilder, configure it to be resolved to the WordPress-specific class, and we obtain the query for ordering results:

$queryHelperService = ContainerBuilderFactory::getInstance()->get('query_helper');
$args = $queryHelperService->getOrderByQuery();
$posts = $postsAPIService->getPosts($args);

Abstracting the values for ordering results that do not correspond to column names or meta properties (such as "post__in" and "rand") seems to be more difficult. Because my application doesn’t use them, I haven’t considered how to do it, or even if it is possible. Then I took the easy way out: I have considered these to be WordPress-specific, hence the application makes them available only when running on WordPress.

Errors

When dealing with errors, we must consider abstracting the following elements:

• the definition of an error;
• error codes and messages.

Let’s review these in turn.

Definition of an error:

An Error is a special object, different than an Exception, used to indicate that some operation has failed, and why it failed. WordPress represents errors through class WP_Error, and allows to check if some returned value is an error through function is_wp_error.

We can abstract checking for an error:

interface CMSCoreInterface
{
  public function isError($object);
}

Which is resolved for WordPress like this:

class WPCMSCore implements CMSCoreInterface
{
  public function isError($object)
  {
    return is_wp_error($object);
  }
}

However, to deal with errors in our abstracted code, we can’t expect all CMSs to have an error class with the same properties and methods as WordPress’s WP_Error class. Hence, we must abstract this class too, and convert from the CMS error to the abstracted error after executing a function from the CMS.

The abstract error class Error is simply a slightly modified version from WordPress’s WP_Error class:

class Error {

  protected $errors = array();
  protected $error_data = array();

  public function __construct($code = null, $message = null, $data = null) 
  {
    if ($code) {
      $this->errors[$code][] = $message;
      if ($data) {
        $this->error_data[$code] = $data;
      }
    }
  }

  public function getErrorCodes()
  {
    return array_keys($this->errors);
  }

  public function getErrorCode()
  {    
    if ($codes = $this->getErrorCodes()) {
      return $codes[0];
    }

    return null;
  }

  public function getErrorMessages($code = null)
  {    
    if ($code) {
      return $this->errors[$code] ?? [];
    }

    // Return all messages if no code specified.
    return array_reduce($this->errors, 'array_merge', array());
  }

  public function getErrorMessage($code = null)
  {
    if (!$code) {
      $code = $this->getErrorCode();
    }
    $messages = $this->getErrorMessages($code);
    return $messages[0] ?? '';
  }

  public function getErrorData($code = null)
  {
    if (!$code) {
      $code = $this->getErrorCode();
    }

    return $this->error_data[$code];
  }

  public function add($code, $message, $data = null)
  {
    $this->errors[$code][] = $message;
    if ($data) {
      $this->error_data[$code] = $data;
    }
  }

  public function addData($data, $code = null)
  {
    if (!$code) {
      $code = $this->getErrorCode();
    }

    $this->error_data[$code] = $data;
  }

  public function remove($code)
  {
    unset($this->errors[$code]);
    unset($this->error_data[$code]);
  }
}

We implement a function to convert from the CMS to the abstract error through a helper class:

class WPHelpers
{
  public static function returnResultOrConvertError($result)
  {
    if (is_wp_error($result)) {
      // Create a new instance of the abstracted error class
      $error = new Error();
      foreach ($result->get_error_codes() as $code) {
        $error->add($code, $result->get_error_message($code), $result->get_error_data($code));
      }
      return $error;
    }
    return $result;
  }
}

And we finally invoke this method for all functions that may return an error:

class UserManagementService implements UserManagementInterface
{
  public function getPasswordResetKey($user_id)
  {
    $result = get_password_reset_key($user_id);
    return WPHelpers::returnResultOrConvertError($result);
  }
}

Error codes and messages:

Every CMS will have its own set of error codes and corresponding explanatory messages. For instance, WordPress function get_password_reset_key can fail due to the following reasons, as represented by their error codes and messages:

1. "no_password_reset": Password reset is not allowed for this user.
2. "no_password_key_update": Could not save password reset key to database.

In order to unify errors so that an error code and message is consistent across CMSs, we will need to inspect these and replace them with our custom ones (possibly in function returnResultOrConvertError explained above).

Hooks

Abstracting hooks involves:

• the hook functionality;
• the hooks themselves.

Let’s analyze these in turn.

Abstracting the hook functionality

WordPress offers the concept of “hooks”: a mechanism through which we can change a default behavior or value (through “filters”) and execute related functionality (through “actions”). Both Symfony and Laravel offer mechanisms somewhat related to hooks: Symfony provides an event dispatcher component, and Laravel’s mechanism is called events; these 2 mechanisms are similar, sending notifications of events that have already taken place, to be processed by the application through listeners.

When comparing these 3 mechanisms (hooks, event dispatcher and events) we find that WordPress’s solution is the simpler one to set-up and use: Whereas WordPress hooks enable to pass an unlimited number of parameters in the hook itself and to directly modify a value as a response from a filter, Symfony’s component requires to instantiate a new object to pass additional information, and Laravel’s solution suggests to run a command in Artisan (Laravel’s CLI) to generate the files containing the event and listener objects. If all we desire is to modify some value in the application, executing a hook such as $value = apply_filters("modifyValue", $value, $post_id); is as simple as it can get.

In the first part of this series, I explained that the CMS-agnostic application already establishes a particular solution for dependency injection instead of relying on the solution by the CMS, because the application itself needs this functionality to glue its parts together. Something similar happens with hooks: they are such a powerful concept that the application can greatly benefit by making it available to the different CMS-agnostic packages (allowing them to interact with each other) and not leave this wiring-up to be implemented only at the CMS level. Hence, I have decided to already ship a solution for the “hook” concept in the CMS-agnostic application, and this solution is the one implemented by WordPress.

In order to decouple the CMS-agnostic hooks from those from WordPress, once again we must “code against interfaces, not implementations”: We define a contract with the corresponding hook functions:

interface HooksAPIInterface
{
  public function addFilter(string $tag, $function_to_add, int $priority = 10, int $accepted_args = 1): void;
  public function removeFilter(string $tag, $function_to_remove, int $priority = 10): bool;
  public function applyFilters(string $tag, $value, ...$args);
  public function addAction(string $tag, $function_to_add, int $priority = 10, int $accepted_args = 1): void;
  public function removeAction(string $tag, $function_to_remove, int $priority = 10): bool;
  public function doAction(string $tag, ...$args): void;
}

Please notice that functions applyFilters and doAction are variadic, i.e. they can receive a variable amount of arguments through parameter ...$args. By combining this feature (which was added to PHP in version 5.6, hence it was unavailable to WordPress until very recently) with argument unpacking, i.e. passing a variable amount of parameters ...$args to a function, we can easily provide the implementation for WordPress:

class WPHooksAPI implements HooksAPIInterface
{
  public function addFilter(string $tag, $function_to_add, int $priority = 10, int $accepted_args = 1): void
  {
    add_filter($tag, $function_to_add, $priority, $accepted_args);
  }

  public function removeFilter(string $tag, $function_to_remove, int $priority = 10): bool
  {
    return remove_filter($tag, $function_to_remove, $priority);
  }

  public function applyFilters(string $tag, $value, ...$args)
  {
    return apply_filters($tag, $value, ...$args);
  }

  public function addAction(string $tag, $function_to_add, int $priority = 10, int $accepted_args = 1): void
  {
    add_action($tag, $function_to_add, $priority, $accepted_args);
  }

  public function removeAction(string $tag, $function_to_remove, int $priority = 10): bool
  {
    return remove_action($tag, $function_to_remove, $priority);
  }

  public function doAction(string $tag, ...$args): void
  {
    do_action($tag, ...$args);
  }
}

As for an application running on Symfony or Laravel, this contract can be satisfied by installing a CMS-agnostic package implementing WordPress-like hooks, such as this one, this one or this one.

Finally, whenever we need to execute a hook, we do it through the corresponding service:

$hooksAPIService = ContainerBuilderFactory::getInstance()->get('hooks_api');
$title = $hooksAPIService->applyFilters("modifyTitle", $title, $post_id);

Abstracting the hooks themselves

We need to make sure that, whenever a hook is executed, a consistent action will be executed no matter which is the CMS. For hooks defined inside of our application that is no problem, since we can resolve them ourselves, most likely in our CMS-agnostic package. However, when the hook is provided by the CMS, such as action "init" (triggered when the system has been initialized) or filter "the_title" (triggered to modify a post’s title) in WordPress, and we invoke these hooks, we must make sure that all other CMSs will process them correctly and consistently. (Please notice that this concerns hooks that make sense in every CMS, such as "init"; certain other hooks can be considered too specific to WordPress, such as filter "rest_{$this->post_type}_query" from a REST controller, so we don’t need to abstract them.)

The solution I found is to hook into actions or filters defined exclusively in the application (i.e. not in the CMS), and to bridge from CMS hooks to application hooks whenever needed. For instance, instead of adding an action for hook "init" (as defined in WordPress), any code in our application must add an action on hook "cms:init", and then we implement the bridge in the WordPress-specific package from "init" to "cms:init":

$hooksAPIService->addAction('init', function() use($hooksAPIService) {
  $hooksAPIService->doAction('cms:init');
});

Finally, the application can add a “loose contract” name for "cms:init", and the CMS-specific package must implement it (as demonstrated earlier on).

Routing

Different frameworks will provide different solutions for routing (i.e. the mechanism of identifying how the requested URL will be handled by the application), which reflect the architecture of the framework:

• In WordPress, URLs map to database queries, not to routes.
• Symfony provides a Routing component which is independent (any PHP application can install it and use it), and which enables to define custom routes and which controller will process them.
• Laravel’s routing builds on top of Symfony’s routing component to adapt it to the Laravel framework.

As it can be seen, WordPress’s solution is the outlier here: the concept of mapping URLs to database queries is tightly coupled to WordPress’s architecture, and we would not want to restrict our abstracted application to this methodology (for instance, October CMS can be set-up as a flat-file CMS, in which case it doesn’t use a database). Instead, it makes more sense to use Symfony’s approach as its default behavior, and allow WordPress to override this behavior with its own routing mechanism.

(Indeed, while WordPress’s approach works well for retrieving content, it is rather inappropriate when we need to access some functionality, such as displaying a contact form. In this case, before the launch of Gutenberg, we were forced to create a page and add a shortcode "[contact_form]" to it as content, which is not as clean as simply mapping the route to its corresponding controller directly.)

Hence, the routing for our abstracted application will not be based around the modeled entities (post, page, category, tag, author) but purely on custom-defined routes. This should already work perfectly for Symfony and Laravel, using their own solutions, and there is not much for us to do other than injecting the routes with the corresponding controllers into the application’s configuration.

To make it work in WordPress, though, we need to take some extra steps: We must introduce an external library to handle routing, such as Cortex. Making use of Cortex, the application running on WordPress can have it both ways:

• if there is a custom-defined route matching the requested URL, use its corresponding controller.
• if not, let WordPress handle the request in its own way (i.e. retrieving the matched database entity or returning a 404 if no match is successful).

To implement this functionality, I have designed the contract CMSRoutingInterface to, given the requested URL, calculate two pieces of information:

• the actual route, such as contact, posts or posts/my-first-post.
• the nature of the route: core nature values "standard", "home" and "404", and additional nature values added through packages such as "post" through a “Posts” package or "user" through a “Users” package.

The nature of the route is an artificial construction that enables the CMS-agnostic application to identify if the route has extra qualities attached to it. For instance, when requesting the URL for a single post in WordPress, the corresponding database object post is loaded into the global state, under global $post. It also helps identify which case we want to handle, to avoid inconsistencies. For instance, we could have defined a custom route contact handled by a controller, which will have nature "standard", and also a page in WordPress with slug "contact", which will have nature "page" (added through a package called “Pages”). Then, our application can prioritize which way to handle the request, either through the controller or through a database query.

Let’s implement it. We first define the service’s contract:

interface CMSRoutingInterface
{
  public function getNature();
  public function getRoute();
}

We can then define an abstract class which provides a base implementation of these functions:

abstract class AbstractCMSRouting implements CMSRoutingInterface
{
  const NATURE_STANDARD = 'standard';
  const NATURE_HOME = 'home';
  const NATURE_404 = '404';

  public function getNature()
  {
    return self::NATURE_STANDARD;
  }

  public function getRoute()
  {
    // By default, the URI path is already the route (minus parameters and trailing slashes)
    $route = $_SERVER['REQUEST_URI'];
    $params_pos = strpos($route, '?');
    if ($params_pos !== false) {
       $route = substr($route, 0, $params_pos);
    }
    return trim($route, '/');
  }
}

And the implementation is overriden for WordPress:

class WPCMSRouting extends AbstractCMSRouting
{
  const ROUTE_QUERY = [
    'custom_route_key' => 'custom_route_value',
  ];
  private $query;
  private function init()
  {
    if (is_null($this->query)) {
      global $wp_query;
      $this->query = $wp_query;
    }
  }

  private function isStandardRoute() {
    return !empty(array_intersect($this->query->query_vars, self::ROUTE_QUERY));
  }

  public function getNature()
  {
    $this->init();
    if ($this->isStandardRoute()) {
      return self::NATURE_STANDARD;
    } elseif ($this->query->is_home() || $this->query->is_front_page()) {
      return self::NATURE_HOME;
    } elseif ($this->query->is_404()) {
      return self::NATURE_404;
    }

    // Allow components to implement their own natures
    $hooksAPIService = ContainerBuilderFactory::getInstance()->get('hooks_api');
    return $hooksAPIService->applyFilters(
      "nature",
      parent::getNature(),
      $this->query
    );
  }
}

In the code above, please notice how constant ROUTE_QUERY is used by the service to know if the route is a custom-defined one, as configured through Cortex:

$hooksAPIService->addAction(
  'cortex.routes', 
  function(RouteCollectionInterface $routes) {  
    // Hook into filter "routes" to provide custom-defined routes
    $appRoutes = $hooksAPIService->applyFilters("routes", []);
    foreach ($appRoutes as $route) {
      $routes->addRoute(new QueryRoute(
        $route,
        function (array $matches) {
          return WPCMSRouting::ROUTE_QUERY;
        }
      ));
    }
  }
);

Finally, we add our routes through hook "routes":

$hooksAPIService->addFilter(
  'routes',
  function($routes) {
    return array_merge(
      $routes,
      [
        'contact',
        'posts',
      ]
    );
  }
);

Now, the application can find out the route and its nature, and proceed accordingly (for instance, for a "standard" nature invoke its controller, or for a "post" nature invoke WordPress’s templating system):

$cmsRoutingService = ContainerBuilderFactory::getInstance()->get('routing');
$nature = $cmsRoutingService->getNature();
$route = $cmsRoutingService->getRoute();
// Process the requested route, as appropriate
// ...

Object properties

A rather inconvenient consequence of abstracting our code is that we can’t reference the properties from an object directly, and we must do it through a function instead. This is because different CMSs will represent the same object as containing different properties, and it is easier to abstract a function to access the object properties than to abstract the object itself (in which case, among other disadvantages, we may have to reproduce the object caching mechanism from the CMS). For instance, a post object $post contains its ID under $post->ID in WordPress and under $post->id in October CMS. To resolve this property, our contract PostObjectPropertyResolverInterface will contain function getId:

interface PostObjectPropertyResolverInterface {
  public function getId($post);
}

Which is resolved for WordPress like this:

class WPPostObjectPropertyResolver implements PostObjectPropertyResolverInterface {
  public function getId($post)
  {
    return $post->ID;
  }
}

Similarly, the post content property is $post->post_content in WordPress and $post->content in October CMS. Our contract will then allow to access this property through function getContent:

interface PostObjectPropertyResolverInterface {
  public function getContent($post);
}

Which is resolved for WordPress like this:

class WPPostObjectPropertyResolver implements PostObjectPropertyResolverInterface {
  public function getContent($post)
  {
    return $post->post_content;
  }
}

Please notice that function getContent receives the object itself through parameter $post. This is because we are assuming the content will be a property of the post object in all CMSs. However, we should be cautious on making this assumption, and decide on a property by property basis. If we don’t want to make the previous assumption, then it makes more sense for function getContent to receive the post’s ID instead:

interface PostObjectPropertyResolverInterface {
  public function getContent($post_id);
}

Being more conservative, the latter function signature makes the code potentially more reusable, however it is also less efficient, because the implementation will still need to retrieve the post object:

class WPPostObjectPropertyResolver implements PostObjectPropertyResolverInterface {
  public function getContent($post_id)
  {
    $post = get_post($post_id);
    return $post->post_content;
  }
}

In addition, some properties may be needed in their original value and also after applying some processing; for these cases, we will need to implement a corresponding extra function in our contract. For instance, the post content needs be accessed also as HTML, which is done through executing apply_filters('the_content', $post->post_content) in WordPress, or directly through property $post->content_html in October CMS. Hence, our contract may have 2 functions to resolve the content property:

interface PostObjectPropertyResolverInterface {
  public function getContent($post_id); // = raw content
  public function getHTMLContent($post_id);
}

We must also be concerned with abstracting the value that the property can have. For instance, a comment is approved in WordPress if its property comment_approved has the value "1". However, other CMSs may have a similar property with value true. Hence, the contract should remove any potential inconsistency or ambiguity:

interface CommentObjectPropertyResolverInterface {
  public function isApproved($comment);
}

Which is implemented for WordPress like this:

class WPCommentObjectPropertyResolver implements CommentObjectPropertyResolverInterface {
  public function isApproved($comment)
  {
    return $comment->comment_approved == "1";
  }
}

Global state

WordPress sets several variables in the global context, such as global $post when querying a single post. Keeping variables in the global context is considered an anti-pattern, since the developer could unintentionally override their values, producing bugs that are difficult to track down. Hence, abstracting our code gives us the chance to implement a better solution.

An approach we can take is to create a corresponding class AppState which simply contains a property to store all variables that our application will need. In addition to initializing all core variables, we enable components to initialize their own ones through hooks:

class AppState
{
  public static $vars = [];

  public static function getVars()
  {
    return self::$vars;
  }

  public static function initialize()
  {
    // Initialize core variables
    self::$vars['nature'] = $cmsRoutingService->getNature();
    self::$vars['route'] = $cmsRoutingService->getRoute();

    // Initialize $vars through hooks
    self::$vars = $hooksAPIService->applyFilters("AppState:init", self::$vars);

    return self::$vars;
  }
}

To replace global $post, a hook from WordPress can then set this data through a hook. A first step would be to set the data under "post-id":

$hooksAPIService->addFilter(
  "AppState:init", 
  function($vars) {
    if (is_single()) {
      global $post;
      $vars['post-id'] => $post->ID;
    }
    return $vars;
  }
);

However, we can also abstract the global variables: instead of dealing with fixed entities (such as posts, users, comments, etc), we can deal with the entity in a generic way through "object-id", and we obtain its properties by inquiring the nature of the requested route:

$hooksAPIService->addFilter(
  "AppState:init", 
  function($vars) {
    if ($vars['nature'] == 'post') {
      global $post;
      $vars['object-id'] => $post->ID;
    }
    return $vars;
  }
);

From now own, if we need to display a property of the current post, we access it from the newly defined class instead of the global context:

$vars = AppState::getVars();
$object_id = $vars['object-id'];
// Do something with it
// ...

Entity models (meta, post types, pages being posts, and taxonomies —tags and categories—)

We must abstract those decisions made for WordPress concerning how its entities are modeled. Whenever we consider that WordPress’s opinionatedness makes sense in a generic context too, we can then replicate such a decision for our CMS-agnostic code.

Meta:

As mentioned earlier, the concept of “meta” must be decoupled from the model entity (such as “post meta” from “posts”), so if a CMS doesn’t provide support for meta, it can then discard only this functionality.

Then, package “Post Meta” (decoupled from, but dependent on, package “Posts”) defines the following contract:

interface PostMetaAPIInterface
{
  public function getMetaKey($meta_key);
  public function getPostMeta($post_id, $key, $single = false);
  public function deletePostMeta($post_id, $meta_key, $meta_value = '');
  public function addPostMeta($post_id, $meta_key, $meta_value, $unique = false);
  public function updatePostMeta($post_id, $meta_key, $meta_value);
}

Which is resolved for WordPress like this:

class WPPostMetaAPI implements PostMetaAPIInterface
{
  public function getMetaKey($meta_key)
  {
    return '_'.$meta_key;
  }
  public function getPostMeta($post_id, $key, $single = false)
  {
    return get_post_meta($post_id, $key, $single);
  }
  public function deletePostMeta($post_id, $meta_key, $meta_value = '')
  {
    return delete_post_meta($post_id, $meta_key, $meta_value);
  }
  public function addPostMeta($post_id, $meta_key, $meta_value, $unique = false)
  {
    return add_post_meta($post_id, $meta_key, $meta_value, $unique);
  }
  public function updatePostMeta($post_id, $meta_key, $meta_value)
  {
    return update_post_meta($post_id, $meta_key, $meta_value);
  }
}

Post types:

I have decided that WordPress’s concept of a custom post type, which allows to model entities (such as an event or a portfolio) as extensions of posts, can apply in a generic context, and as such, I have replicated this functionality in the CMS-agnostic code. This decision is controversial, however, I justify it because the application may need to display a feed of entries of different types (such as posts, events, etc) and custom post types make such implementation feasible. Without custom post types, I would expect the application to need to execute several queries to bring the data for every entity type, and the logic would get all muddled up (for instance, if fetching 12 entries, should we fetch 6 posts and 6 events? but what if the events were posted much earlier than the last 12 posts? and so on).

What happens when the CMS doesn’t support this concept? Well, nothing serious happens: a post will still indicate its custom post type to be a “post”, and no other entities will inherit from the post. The application will still work properly, just with some slight overhead from the unneeded code. This is a trade-off that, I believe, is more than worth it.

To support custom post types, we simply add a function getPostType in our contract:

interface PostAPIInterface
{
  public function getPostType($post_id);
}

Which is resolved for WordPress like this:

class WPPostAPI implements PostAPIInterface
{
  public function getPostType($post_id) {
    return get_post_type($post_id);
  }
}

Pages being posts:

While I justify keeping custom post types in order to extend posts, I don’t justify a page being a post, as it happens in WordPress, because in other CMSs these entities are completely decoupled and, more importantly, a page may have higher rank than a post, so making a page extend from a post would make no sense. For instance, October CMS ships pages in its core functionality, but posts must be installed through plugins.

Hence we must create separate contracts for posts and pages, even though they may contain the same functions:

interface PostAPIInterface
{
  public function getTitle($post_id);
}

interface PageAPIInterface
{
  public function getTitle($page_id);
}

To resolve these contracts for WordPress and avoid duplicating code, we can implement the common functionality through a trait:

trait WPCommonPostFunctions
{
  public function getTitle($post_id)
  {
    return get_the_title($post_id);
  }
}

class WPPostAPI implements PostAPIInterface
{
  use WPCommonPostFunctions;
}

class WPPageAPI implements PageAPIInterface
{
  use WPCommonPostFunctions;
}

Taxonomies (tags and categories):

Once again, we can’t expect all CMSs to support what is called taxonomies in WordPress: tags and categories. Hence, we must implement this functionality through a package “Taxonomies”, and, assuming that tags and categories are added to posts, make this package dependent on package “Posts”.

interface TaxonomyAPIInterface
{
  public function getPostCategories($post_id, $options = []);
  public function getPostTags($post_id, $options = []);
  public function getCategories($query, $options = []);
  public function getTags($query, $options = []);
  public function getCategory($cat_id);
  public function getTag($tag_id);
  ...
}

We could have decided to create two separate packages “Categories” and “Tags” instead of “Taxonomies”, however, as the implementation in WordPress makes evident, a tag and a category are basically the same concept of entity with only a tiny difference: categories are hierarchical (i.e. a category can have a parent category), but tags are not. Then, I consider that it makes sense to keep this concept for a generic context, and shipped under a single package “Taxonomies”.

We must pay attention that certain functionalities involve both posts and taxonomies, and these must be appropriately decoupled. For instance, in WordPress we can retrieve posts that were tagged "politics" by executing get_posts(['tag' => "politics"]). In this case, while function getPosts must be implemented in package “Posts”, filtering by tags must be implemented in package “Taxonomies”. To accomplish this separation, we can simply execute a hook in the implementation of function getPosts for WordPress, allowing any component to modify the arguments before executing get_posts:

class WPPostAPI implements PostAPIInterface
{
  public function getPosts($args) {
    $args = $hooksAPIService->applyFilters("modifyArgs", $args);
    return get_posts($args);
  }
}

And finally we implement the hook in package “Taxonomies for WordPress”:

$hooksAPIService->addFilter(
  'modifyArgs',
  function($args) {
    if (isset($args['tags'])) {
      $args['tag'] = implode(',', $args['tags']);
      unset($args['tags']);
    }
    if (isset($args['categories'])) {
      $args['cat'] = implode(',', $args['categories']);
      unset($args['categories']);
    }
    return $args;
  }
);

Please notice that in the abstracted code the attributes were re-defined (following the recommendations for abstracting function parameters, explained earlier on): "tag" must be provided as "tags" and "cat" must be provided as "categories" (shifting the connotation from singular to plural), and these values must be passed as arrays (i.e. removed accepting comma-separated strings as in WordPress, to add consistency).

Translation

Because calls to translate strings are spread all over the application code, translation is not a functionality that we can opt out from, and we should make sure that the other frameworks are compatible with our chosen translation mechanism.

In WordPress, which implements internationalization through gettext, we are required to set-up translation files for each locale code (such as ‘fr_FR’, which is the code for french language from FRance), and these can be set under a text domain (which allows themes or plugins to define their own translations without fear of collision with the translations from other pieces of code). We don’t need to check for support for placeholders in the string to translate (such as when doing sprintf(__("Welcome %s"), $user_name)), because function sprintf belongs to PHP and not to the CMS, so it will always work.

Let’s check if the other frameworks support the required two properties, i.e. getting the translation data for a specific locale composed of language and country, and under a specific text domain:

• Symfony’s translation component supports these two properties.
• The locale used in Laravel’s localization involves the language but not the country, and text domains are not supported (they could be replicated through overriding package language files, but the domain is not explicitly set, so the contract and the implementation would be inconsistent with each other).

However, luckily there is library Laravel Gettext which can replace Laravel’s native implementation with Symfony’s translation component. Hence, we got support for all frameworks, and we can rely on a WordPress-like solution.

We can then define our contract mirroring the WordPress function signatures:

interface TranslationAPIInterface
{
  public function __($text, $domain = 'default');
  public function _e($text, $domain = 'default');
}

The implementation of the contract for WordPress is like this:

class WPTranslationAPI implements TranslationAPIInterface
{
  public function __($text, $domain = 'default')
  {
    return __($text, $domain);
  }
  public function _e($text, $domain = 'default')
  {
    _e($text, $domain);
  }
}

And to use it in our application, we do:

$translationAPI = ContainerBuilderFactory::getInstance()->get('translation_api');
$text = $translationAPI->__("translate this", "my-domain");

Media

WordPress has media management as part of its core functionality, which represents a media element as an entity all by itself, and allows to manipulate the media element (such as cropping or resizing images), but we can’t expect all CMSs to have similar functionality. Hence, media management must be decoupled from the CMS core functionality.

For the corresponding contract, we can mirror the WordPress media functions, but removing WordPress’s opinionatedness. For instance, in WordPress, a media element is a post (with post type "attachment"), but for the CMS-agnostic code it is not, hence the parameter must be $media_id (or $image_id) instead of $post_id. Similarly, WordPress treats media as attachments to posts, but this doesn’t need to be the case everywhere, hence we can remove the word “attachment” from the function signatures. Finally, we can decide to keep the $size of the image in the contract; if the CMS doesn’t support creating multiple image sizes for an image, then it can just fall back on its default value NULL and nothing grave happens:

interface MediaAPIInterface
{
  public function getImageSrcAndDimensions($image_id, $size = null): array;
  public function getImageURL($image_id, $size = null): string;
}

The response by function getImageSrcAndDimensions can be asbtracted too, returning an array of our own design instead of simply re-using the one from the WordPress function wp_get_attachment_image_src:

class WPMediaAPI implements MediaAPIInterface
{
  public function getImageSrcAndDimensions($image_id, $size = null): array
  {
    $img_data = wp_get_attachment_image_src($image_id, $size);
    return [
      'src' => $img_data[0],
      'width' => $img_data[1],
      'height' => $img_data[2],
    ];
  }
  public function getImageURL($image_id, $size = null): string
  {
    return wp_get_attachment_image_url($image_id, $size);
  }
}

Conclusion

Setting-up a CMS-agnostic architecture for our application can be a painful endeavor. As it was demonstrated in this article, abstracting all the code was a lengthy process, taking plenty of time and energy to achieve, and it is not even finished yet. I wouldn’t be surprised if the reader is intimidated by the idea of going through this process in order to convert a WordPress application into a CMS-agnostic one. If I hadn’t done the abstraction myself, I would certainly be intimidated too.

My suggestion is for the reader is to analyze if going through this process makes sense based on a project-by-project basis. If there is no need whatsoever to port an application to a different CMS, then you will be right to stay away from this process and stick to the WordPress way. However, if you do need to migrate an application away from WordPress and want to reduce the effort required, or if you already need to maintain several codebases which would benefit from code reusability, or even if you may migrate the application sometime in the future and you have just started a new project, then this process is for you. It may be painful to implement, but well worth it. I know because I’ve been there. But I’ve survived, and I’d certainly do it again. Thanks for reading.

So far as I can tell, it is uncommon to come across educational content in the field of software development which provides an appropriate mixture of theoretical and practical information. If I was to guess why, I assume it is because individuals who focus on theory tend to get into teaching, and individuals who focus on practical information tend to get paid to solve specific problems, using specific languages and tools.

This is, of course, a broad generalization, but if we accept it briefly for arguments sake, it follows that many people (by no means all people) who take on the role of teacher, tend to be either poor at, or utterly incapable of explaining the practical knowledge relevant to a particular concept.

In this article, I will do my best to discuss three core mechanisms which you will find in most Object Oriented Programming (OOP) languages: Inheritance, interfaces (a.k.a. protocols), and abstract classes. Rather than giving you technical and complex verbal explanations of what each mechanism is, I will do my best to focus on what they do, and when to use them.

However, before addressing them individually, I would like to briefly discuss what it means to give a theoretically sound, yet practically useless explanation. My hope is that you can use this information to help you sift through different educational resources and to avoid blaming yourself when things do not make sense.

Different Degrees Of Knowing

Knowing Names

Knowing the name of something is arguably the most shallow form of knowing. In fact, a name is only generally useful to the extent that it either is commonly used by many people to refer to the same thing and/or it helps to describe the thing. Unfortunately, as anyone who has spent time in this field has discovered, many people use different names for the same thing (e.g. interfaces and protocols), the same names for different things (e.g. modules and components), or names which are esoteric to the point of being absurd (e.g. Either Monad). Ultimately, names are just pointers (or references) to mental models, and they can be of varying degrees of usefulness.

To make this field even more difficult to study, I would hazard a guess that for most individuals, writing code is (or at least was) a very unique experience. Even more complicated is understanding how that code is ultimately compiled into machine language, and represented in physical reality as a series of electrical impulses changing over time. Even if one can recall the names of the processes, concepts, and mechanisms that are employed in a program, there is no guarantee that the mental models which one creates for such things are consistent with the models another individual; let alone whether they are objectively accurate.

It is for these reasons, alongside the fact that I do not have a naturally good memory for jargon, that I consider names to be the least important aspect of knowing something. That is not to say that names are useless, but I have in the past learned and employed many design patterns in my projects, only to learn of the commonly used name months, or even years later.

Knowing Verbal Definitions And Analogies

Verbal definitions are the natural starting point for describing a new concept. However, as with names, they can be of varying degrees of usefulness and relevance; much of that depending on what the end goals of the learner are. The most common problem I see in verbal definitions is assumed knowledge typically in the form of jargon.

Suppose for example, that I was to explain that a thread is very much like a process, except that threads occupy the same address space of a given process. To someone who is already familiar with processes and address spaces, I have essentially stated that threads can be associated with their understanding of a process (i.e. they possess many of the same characteristics), but they can be differentiated based on a distinct characteristic.

To someone who does not possess that knowledge, I have at best not made any sense, and at worst caused the learner to feel inadequate in some way for not knowing the things I have assumed they should know. In fairness, this is acceptable if your learners really ought to possess such knowledge (such as teaching graduate students or experienced developers), but I consider it to be a monumental failure to do so in any introductory level material.

Often it is very difficult to provide a good verbal definition of a concept when it is unlike anything else the learner has seen before. In this case, it is very important for the teacher to select an analogy which is likely to be familiar to the average person, and also relevant insofar as it conveys many of the same qualities of the concept.

For example, it is critically important for a software developer to understand what it means when software entities (different parts of a program) are tightly-coupled or loosely-coupled. When building a garden shed, a junior carpenter may think that it is faster and easier to put it together using nails instead of screws. This is true up until the point at which a mistake is made, or a change in the design of the garden shed necessitates rebuilding part of the shed.

At this point, the decision to use nails to tightly-couple the parts of the garden shed together, has made the construction process as a whole more difficult, likely slower, and extracting nails with a hammer runs the risk of damaging the structure. Conversely, screws can take a bit of extra time to assemble, but they are easy to remove and pose little risk of damaging nearby parts of the shed. This is what I mean by loosely-coupled. Naturally, there are cases where you really just need a nail, but that decision ought to be guided by critical thinking and experience.

As I will discuss in detail later on, there are different mechanisms for connecting parts of a program together which provide varying degrees of coupling; just like nails and screws. While my analogy may have helped you understand what this critically important term means, I have not given you any idea of how to apply it outside of the context of building a garden shed. This leads me to the most important kind of knowing, and the key to deeply understanding vague and difficult concepts in any field of inquiry; although we will stick to writing code in this article.

Knowing In Code

In my opinion, strictly regarding software development, the most import form of knowing a concept comes from being able to use it in the working application code. This form of knowing can be attained simply by writing lots of code and solving many different problems; jargon names and verbal definitions need not be included.

In my own experience, I recall solving the problem of communicating with a remote database, and a local database through a single interface (you will know what that means soon if you do not already); rather than the client (whatever class which talks to the interface) needing to call the remote and local (or even a test database) explicitly. In fact, the client had no idea what was behind the interface, so I did not need to change it regardless of if it was running in a production app or a test environment. About a year after I solved this problem, I came across the term “Facade Pattern”, and not long after the term “Repository Pattern”, which are both names that people use for the solution described prior.

All of this preamble is to hopefully illuminate some of the flaws which are most often made in explaining topics such as inheritance, interfaces, and abstract classes. Of the three, inheritance is likely the simplest one to both use and understand. In my experience both as a student of programming, and a teacher, the other two are almost invariably an issue to learners unless very special attention is paid to avoiding the mistakes discussed earlier. From this point on, I will do my best to make these topics as simple as they should be, but no simpler.

A Note On The Examples

Being most fluent in Android mobile application development myself, I will use examples taken from that platform so that I may teach you about building GUI applications at the same time as introducing language features of Java. However, I will not be going into so much detail that the examples ought to be unintelligible by someone with a cursory understanding of Java EE, Swing or JavaFX. My ultimate goal in discussing these topics is to help you to understand what they mean in the context of solving a problem in just about any sort of application.

I would also like to warn you, dear reader, that at times it may seem like I am being needlessly philosophical and pedantic about specific words and their definitions. The reason for this is that there truly is a deep philosophical underpinning required to understand the difference between something which is concrete (real), and something which is abstract (less detailed than a real thing). This understanding applies to many things outside of the field of computing, but it is of particularly high importance for any software developer to grasp the nature of abstractions. In any case, if my words fail you, the examples in code will hopefully not.

Inheritance And Implementation

When it comes to building applications with a graphical user interface (GUI), inheritance is arguably the most important mechanism for making it possible to quickly build an application.

Although there is a lesser understood benefit to using inheritance to be discussed later, the primary benefit is to share implementation between classes. This word “implementation”, at least for the purposes of this article, has a distinct meaning. To give a general definition of the word in English, I might say that to implement something, is to make it real.

To give a technical definition specific to software development, I might say that to implement a piece of software, is to write concrete lines of code which satisfy the requirements of said piece of software. For example, suppose I am writing a sum method:
private double sum(double first, double second){

private double sum(double first, double second){
        //TODO: implement
}

The snippet above, even though I have made it as far as writing a return type (double) and a method declaration which specifies the arguments (first, second) and the name which can be used to call said method (sum), it has not been implemented. In order to implement it, we must complete the method body like so:

private double sum(double first, double second){
        return first + second;
}

Naturally, the first example would not compile, but we will see momentarily that interfaces are a way in which we can write these sorts of unimplemented functions without errors.

Inheritance In Java

Presumably, if you are reading this article, you have used the extends Java keyword at least once. The mechanics of this keyword are simple and most often described using examples to do with different kinds of animals or geometric shapes; Dogand Catextend Animal, and so forth. I will assume that I do not need to explain rudimentary type theory to you, so let us get right into the primary benefit of inheritance in Java, via the extendskeyword.

Building a console-based “Hello World” application in Java is very simple. Assuming you possess a Java Compiler (javac) and runtime environment (jre), you can write a class which contains a main function like so:

public class JavaApp{
     public static void main(String []args){
        System.out.println("Hello World");
     }
}

Building a GUI application in Java on almost any of its main platforms (Android, Enterprise/Web, Desktop), with a bit of help from an IDE to generate the skeleton/boilerplate code of a new app, is also relatively easy thanks to the extendskeyword.

Suppose that we have an XML Layout called activity_main.xml (we typically build user interfaces declaratively in Android, via Layout files) containing a TextView(like a text label) called tvDisplay:

<?xml version="1.0" encoding="utf-8"?>
<FrameLayout xmlns:android="http://schemas.android.com/apk/res/android"
    android:layout_width="match_parent" android:layout_height="match_parent">

    <TextView
        android:id="@+id/tvDisplay"
        android:layout_width="wrap_content"
        android:layout_height="wrap_content"
        android:layout_gravity="center"
        />


</FrameLayout>

Also, suppose that we would like tvDisplayto say “Hello World!” To do so, we simply need to write a class which uses the extendskeyword to inherit from the Activityclass:

import android.app.Activity;
import android.os.Bundle;
import android.widget.TextView;

public class MainActivity extends Activity {
    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        setContentView(R.layout.activity_main);


        ((TextView)findViewById(R.id.tvDisplay)).setText("Hello World");
    }

The effect of inheriting the implementation of the Activityclass can be best appreciated by taking a quick look at its source code. I highly doubt that Android would have become the dominant mobile platform if one needed to implement even a small portion of the 8000+ lines necessary to interact with the system, just to generate a simple window with some text. Inheritance is what allows us to not have to rebuild the Android framework, or whatever platform you happen to be working with, from scratch.

Inheritance Can Be Used For Abstraction

Insofar as it can be used to share implementation across classes, inheritance is relatively simple to understand. However, there is another important way in which inheritance can be used, which is conceptually related to the interfaces and abstract classes which we will be discussing soon.

If you please, suppose for the next little while that an abstraction, used in the most general sense, is a less detailed representation of a thing. Instead of qualifying that with a lengthy philosophical definition, I will try to point out how abstractions work in daily life, and shortly thereafter discuss them expressly in terms of software development.

Suppose you are traveling to Australia, and you are aware that the region you are visiting is host to a particularly high density of inland taipan snakes (they are apparently quite poisonous). You decide to consult Wikipedia to learn more about them by looking at images and other information. By doing so, you are now acutely aware of a particular kind of snake which you have never seen before.

Abstractions, ideas, models, or whatever else you want to call them, are less detailed representations of a thing. It is important that they are less detailed than the real thing because a real snake can bite you; images on Wikipedia pages typically do not. Abstractions are also important because both computers and human brains have a limited capacity to store, communicate, and process information. Having enough detail to use this information in a practical way, without taking up too much space in memory, is what makes it possible for computers and human brains alike to solve problems.

To tie this back into inheritance, all of the three main topics I am discussing here can be used as abstractions, or mechanisms of abstraction. Suppose that in our “Hello World” app’s layout file, we decide to add an ImageView, Button, and ImageButton:

<?xml version="1.0" encoding="utf-8"?>
<LinearLayout xmlns:android="http://schemas.android.com/apk/res/android"
    android:layout_width="match_parent" android:layout_height="match_parent">
    <Button
       android:id="@+id/btnDisplay"
       android:layout_width="wrap_content"
       android:layout_height="wrap_content"/>

    <ImageButton
        android:id="@+id/imbDisplay"
        android:layout_width="wrap_content"
        android:layout_height="wrap_content"/>

    <ImageView
        android:id="@+id/imvDisplay"
        android:layout_width="wrap_content"
        android:layout_height="wrap_content"/>
</LinearLayout>

Also suppose that our Activity has implemented View.OnClickListener to handle clicks:

public class MainActivity extends Activity implements View.OnClickListener {
    private Button b;
    private ImageButton ib;
    private ImageView iv;    


    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        setContentView(R.layout.activity_main);

        //...
        b = findViewById(R.id.imvDisplay).setOnClickListener(this);
        ib = findViewById(R.id.btnDisplay).setOnClickListener(this);
        iv = findViewById(R.id.imbDisplay).setOnClickListener(this);
    }

    @Override
    public void onClick(View view) {
        final int id = view.getId();
        //handle click based on id...
    }
}

The key principle here is that Button, ImageButton, and ImageViewinherit from the Viewclass. The result is that this function onClickcan receive click events from disparate (though hierarchically related) UI elements by referencing them as their less detailed parent class. This is far more convenient than having to write a distinct method for handling every kind of widget on the Android platform (not to mention custom widgets).

Interfaces And Abstraction

You may have found the previous code example to be a bit uninspiring, even if you understood why I chose it. Being able to share implementation across a hierarchy of classes is incredibly useful, and I would argue that to be the primary utility of inheritance. As for allowing us to treat a set of classes that have a common parent class as equal in type (i.e. as the parent class), that feature of inheritance has limited use.

By limited, I am speaking of the requirement of child classes to be within the same class hierarchy in order to be referenced via, or known as the parent class. In other words, inheritance is a very restrictive mechanism for abstraction. In fact, if I suppose that abstraction is a spectrum that moves between different levels of detail (or information), I might say that inheritance is the least abstract mechanism for abstraction in Java.

Before I proceed to discussing interfaces, I would like to mention that as of Java 8, two features called Default Methods and Static Methods have been added to interfaces. I will discuss them eventually, but for the moment I would like us to pretend that they do not exist. This is in order for me to make it easier to explain the primary purpose of using an interface, which was initially, and arguably still is, the most abstract mechanism for abstraction in Java.

Less Detail Means More Freedom

In the section on inheritance, I gave a definition of the word implementation, which was meant to contrast with another term we will now go into. To be clear, I do not care about the words themselves, or whether you agree with their usage; only that you understand what they conceptually point to.

Whereas inheritance is primarily a tool to share implementation across a set of classes, we might say that interfaces are primarily a mechanism to share behavior across a set of classes. Behavior used in this sense is truly just a non-technical word for abstract methods. An abstract method is a method which does not, in fact cannot, contain a method body:

public interface OnClickListener {
        void onClick(View v);
}

The natural reaction for me, and a number of individuals whom I have tutored, after first looking at an interface, was to wonder what the utility of sharing only a return type, method name, and parameter list might be. On the surface, it looks like a great way to create extra work for yourself, or whoever else might be writing the class which implementsthe interface. The answer, is that interfaces are perfect for situations where you want a set of classes to behave in the same manner (i.e. they possess the same public abstract methods), but you expect them to implement that behavior in different ways.

To take a simple but relevant example, the Android platform possesses two classes which are primarily in the business of creating and managing part of the user interface: Activity and Fragment. It follows that these classes will very often have the requirement of listening to events which pop up when a widget is clicked (or otherwise interacted with by a user). For argument’s sake, let us take a moment to appreciate why inheritance will almost never solve such a problem:

public class OnClickManager {
    public void onClick(View view){
        //Wait a minute... Activities and Fragments almost never 
        //handle click events exactly the same way...
    }
}

Not only would making our Activities and Fragments inherit from OnClickManagermake it impossible to handle events in a different manner, but the kicker is that we could not even do that if we wanted to. Both Activity and Fragment already extend a parent class, and Java does not allow multiple parent classes. So our problem is that we want a set of classes to behave the same way, but we must have flexibility on how the class implements that behavior. This brings us back to the earlier example of the View.OnClickListener:

public interface OnClickListener {
        void onClick(View v);
}

This is the actual source code (which is nested in the Viewclass), and these few lines allow us to ensure consistent behavior across different widgets (Views) and UI controllers (Activities, Fragments, etc.).

Abstraction Promotes Loose-Coupling

I have hopefully answered the general question about why interfaces exist in Java; among many other languages. From one perspective, they are just a means of sharing code between classes, but they are deliberately less detailed in order to allow for different implementations. But just as inheritance can be used both as a mechanism for sharing code and abstraction (albeit with restrictions on class hierarchy), it follows that interfaces provide a more flexible mechanism for abstraction.

In an earlier section of this article, I introduced the topic of loose/tight-coupling by analogy of the difference between using nails and screws to build some kind of structure. To recap, the basic idea is that you will want to use screws in situations where changing the existing structure (which can be a result of fixing mistakes, design changes, and so forth) is likely to happen. Nails are fine to use when you just need to fasten parts of the structure together and are not particularly worried about taking them apart in the near future.

Nails and screws are meant to be analogous to concrete and abstract references (the term dependencies also applies) between classes. Just so there is no confusion, the following sample will demonstrate what I mean:

class Client {
    private Validator validator;
    private INetworkAdapter networkAdapter;

    void sendNetworkRequest(String input){
        if (validator.validateInput(input)) {
            try {
                networkAdapter.sendRequest(input);
            } catch (IOException e){
                //handle exception
            }
        }
    }
}

class Validator {
    //...validation logic
    boolean validateInput(String input){
        boolean isValid = true;
        //...change isValid to false based on validation logic
        return isValid;
    }
}

interface INetworkAdapter {
    //...
    void sendRequest(String input) throws IOException;
}

Here, we have a class called Client which possesses two kinds of references. Notice that, assuming Clientdoes not have anything to do with creating its references (it really should not), it is decoupled from the implementation details of any particular network adapter.

There are a few important implications of this loose coupling. For starters, I can build Clientin absolute isolation of any implementation of INetworkAdapter. Imagine for a moment that you are working in a team of two developers; one to build the front end, one to build the back end. As long as both developers are kept aware of the interfaces which couple their respective classes together, they can carry on with the work virtually independently of one another.

Secondly, what if I were to tell you that both developers could verify that their respective implementations functioned properly, also independently of each other’s progress? This is very easy with interfaces; just build a Test Double which implementsthe appropriate interface:

class FakeNetworkAdapter implements INetworkAdapter {
    public boolean throwError = false;


    @Override
    public void sendRequest(String input) throws IOException {
        if (throwError) throw new IOException("Test Exception");
    }
}

In principle, what can be observed is that working with abstract references opens the door to increased modularity, testability, and some very powerful design patterns such as the Facade Pattern, Observer Pattern, and more. They can also allow developers to find a happy balance of designing different parts of a system based on behavior (Program To An Interface), without getting bogged down in implementation details.

A Final Point On Abstractions

Abstractions do not exist in the same way as a concrete thing. This is reflected in the Java Programming language by the fact that abstract classes and interfaces may not be instantiated.

For example, this would definitely not compile:

public class Main extends Application {
    public static void main(String[] args) {
        launch(args);
    }

    @Override
    public void start(Stage primaryStage) {
      //ERROR x2:
        Foo f = new Foo();
        Bar b = new Bar()

    }


    private abstract class Foo{}
    private interface Bar{}


}

In fact, the idea of expecting an unimplemented interface or abstract class to function at runtime makes as much sense as expecting a UPS uniform to float around delivering packages. Something concrete must be behind the abstraction for it to be of utility; even if the calling class does not need to know what is actually behind abstract references.

Abstract Classes: Putting It All Together

If you have made it this far, then I am happy to tell you that I have no more philosophical tangents or jargon phrases to translate. Simply put, abstract classes are a mechanism for sharing implementation and behavior across a set of classes. Now, I will admit straight away that I do not find myself using abstract classes all that often. Even so, my hope is that by the end of this section you will know exactly when they are called for.

Workout Log Case Study

Roughly a year into building Android apps in Java, I was rebuilding my first Android app from scratch. The first version was the kind of horrendous mass of code that you would expect from a self-taught developer with little guidance. By the time I wanted to add new functionality, it became clear that the tightly coupled structure I had built exclusively with nails, was so impossible to maintain that I must rebuild it entirely.

The app was a workout log that was designed to allow easy recording of your workouts, and the ability to output the data of a past workout as a text or image file. Without getting into too much detail, I structured the data models of the app such that there was a Workoutobject, which comprised of a collection of Exerciseobjects (among other fields which are irrelevant to this discussion).

As I was implementing the feature for outputting workout data to some kind of visual medium, I realized that I had to deal with a problem: Different kinds of exercises would require different kinds of text outputs.

To give you a rough idea, I wanted to change the outputs depending on the type of exercise like so:

• Barbell: 10 REPS @ 100 LBS
• Dumbbell: 10 REPS @ 50 LBS x2
• Bodyweight: 10 REPS @ Bodyweight
• Bodyweight +: 10 REPS @ Bodyweight + 45 LBS
• Timed: 60 SEC @ 100 LBS

Before I proceed, note that there were other types (working out can become complicated) and that the code I will be showing has been trimmed down and changed to fit nicely into an article.

In keeping with my definition from before, the goal of writing an abstract class, is to implement everything (even state such as variables and constants) which is shared across all child classes in the abstract class. Then, for anything which changes across said child classes, create an abstract method:

abstract class Exercise {
    private final String type;
    protected final String name;
    protected final int[] repetitionsOrTime;
    protected final double[] weight;

    protected static final String POUNDS = "LBS";
    protected static final String SECONDS = "SEC ";
    protected static final String REPETITIONS = "REPS ";

    public Exercise(String type, String name, int[] repetitionsOrTime, double[] weight) {
        this.type = type;
        this.name = name;
        this.repetitionsOrTime = repetitionsOrTime;
        this.weight = weight;
    }

    public String getFormattedOutput(){
        StringBuilder sb = new StringBuilder();
        sb.append(name);
        sb.append("n");
        getSetData(sb);
        sb.append("n");
        return sb.toString();
    }

    /**
     * Append data appropriately based on Exercise type
     * @param sb - StringBuilder to Append data to
     */
    protected abstract void getSetData(StringBuilder sb);
    
    //...Getters
}

I may be stating the obvious, but if you have any questions about what should or should not be implemented in the abstract class, the key is to look at any part of the implementation which has been repeated in all child classes.

Now that we have established what is common amongst all exercises, we can begin to create child classes with specializations for each kind of String output:

Barbell Exercise:

class BarbellExercise extends Exercise {
    public BarbellExercise(String type, String name, int[] repetitionsOrTime, double[] weight) {
        super(type, name, repetitionsOrTime, weight);
    }

    @Override
    protected void getSetData(StringBuilder sb) {
        for (int i = 0; i 

Dumbbell Exercise:

class DumbbellExercise extends Exercise {
    private static final String TIMES_TWO = "x2";

    public DumbbellExercise(String type, String name, int[] repetitionsOrTime, double[] weight) {
        super(type, name, repetitionsOrTime, weight);
    }

    @Override
    protected void getSetData(StringBuilder sb) {

        for (int i = 0; i 

Bodyweight Exercise:

class BodyweightExercise extends Exercise {
    private static final String BODYWEIGHT = "Bodyweight";

    public BodyweightExercise(String type, String name, int[] repetitionsOrTime, double[] weight) {
        super(type, name, repetitionsOrTime, weight);
    }

    @Override
    protected void getSetData(StringBuilder sb) {

        for (int i = 0; i 

I am certain that some astute readers will find things which could have been abstracted out in a more efficient manner, but the purpose of this example (which has been simplified from the original source) is to demonstrate the general approach. Of course, no programming article would be complete without something which can be executed. There are several online Java compilers which you may use to run this code if you want to test it out (unless you already have an IDE):

public class Main {
    public static void main(String[] args) {
        //Note: I actually used another nested class called a "Set" instead of an Array
        //to represent each Set of an Exercise.
        int[] reps = {10, 10, 8};
        double[] weight = {70.0, 70.0, 70.0};

        Exercise e1 = new BarbellExercise(
                "Barbell",
                "Barbell Bench Press",
                reps,
                weight
        );

        Exercise e2 = new DumbbellExercise(
                "Dumbbell",
                "Dumbbell Bench Press",
                reps,
                weight
        );

        Exercise e3 = new BodyweightExercise(
                "Bodyweight",
                "Push Up",
                reps,
                weight
        );

        System.out.println(
                e1.getFormattedOutput()
                + e2.getFormattedOutput()
                + e3.getFormattedOutput()
        );
    }
}

Executing this toy application yields the following output:
Barbell Bench Press

10 REPS  @ 70.0LBS
10 REPS  @ 70.0LBS
8 REPS  @ 70.0LBS

Dumbbell Bench Press
10 REPS  @ 70.0LBSx2
10 REPS  @ 70.0LBSx2
8 REPS  @ 70.0LBSx2

Push Up
10 REPS  @ Bodyweight
10 REPS  @ Bodyweight
8 REPS  @ Bodyweight

Further Considerations

Earlier, I mentioned that there are two features of Java interfaces (as of Java 8) which are decidedly geared towards sharing implementation, as opposed to behavior. These features are known as Default Methods and Static Methods.

I have decided not to go into detail on these features for the reason that they are most typically used in mature and/or large code bases where a given interface has many inheritors. Despite the fact that this is meant to be an introductory article, and I still encourage you to take a look at these features eventually, even though I am confident that you will not need to worry about them just yet.

I would also like to mention that there are other ways to share implementation across a set of classes (or even static methods) in a Java application that does not require inheritance or abstraction at all. For example, suppose you have some implementation which you expect to use in a variety of different classes, but does not necessarily make sense to share via inheritance. A common pattern in Java is to write what is known as a Utility class, which is a simple classcontaining the requisite implementation in a static method:

public class TimeConverterUtil {

    /**
     * Accepts an hour (0-23) and minute (0-59), then attempts to format them into an appropriate
     * format such as 12, 30 -> 12:30 pm
     */    
    public static String convertTime (int hour, int minute){
        String unformattedTime = Integer.toString(hour) + ":" + Integer.toString(minute);
        DateFormat f1 = new SimpleDateFormat("HH:mm");

        Date d = null;
        try { d = f1.parse(unformattedTime); } 
        catch (ParseException e) { e.printStackTrace(); }
        DateFormat f2 = new SimpleDateFormat("h:mm a");
        return f2.format(d).toLowerCase();
    }
}

Using this static method in an external class (or another static method) looks like this:

public class Main {
    public static void main(String[] args){
        //...
        String time = TimeConverterUtil.convertTime(12, 30);
        //...
    }
}

Cheat Sheet

We have covered a lot of ground in this article, so I would like to spend a moment summarizing the three main mechanisms based on what problems they solve. Since you should possess a sufficient understanding of the terms and ideas I have either introduced or redefined for the purposes of this article, I will keep the summaries brief.

I Want A Set Of Child Classes To Share Implementation

Classic inheritance, which requires a child class to inherit from a parent class, is a very simple mechanism for sharing implementation across a set of classes. An easy way to decide if some implementation should be pulled into a parent class, is to see whether it is repeated in a number of different classes line for line. The acronym DRY (Don’t Repeat Yourself) is a good mnemonic device to watch out for this situation.

While coupling child classes together with a common parent class can present some limitations, a side benefit is that they can all be referenced as the parent class, which provides a limited degree of abstraction.

I Want A Set Of Classes To Share Behavior

Sometimes, you want a set of classes to be capable of possessing certain abstract methods (referred to as behavior), but you do not expect the implementation of that behavior to be repeated across inheritors.

By definition, Java interfaces may not contain any implementation (except for Default and Static Methods), but any class which implements an interface, must supply an implementation for all abstract methods, otherwise, the code will not compile. This provides a healthy measure of flexibility and restriction on what is actually shared and does not require the inheritors to be of the same class hierarchy.

I Want A Set Of Child Classes To Share Behavior And Implementation

Although I do not find myself using abstract classes all over the place, they are perfect for situations when you require a mechanism for sharing both behavior and implementation across a set of classes. Anything which will be repeated across inheritors may be implemented directly in the abstract class, and anything which requires flexibility may be specified as an abstract method.