I have a web site I'm converting to Codeigniter and I want to simplify and decouple. I like what I've read about the Observer pattern for things like "new survey created" (which triggers a new help ticket, which triggers an email, etc).
But how do I implement such a thing in Code Igniter? I see the Symfony component but at this point I'm not concerned about understanding the system as much as figuring out how to use it in controllers and models. I have extended both CI_Model and CI_Controller already for other reasons. Would putting Observer pattern code there be the best?
I imagine a point like this: someone hits the web site and spawns a request which gets routed to a controller/action: http://localhost/test/save_changes
// warning, pseudo-code!
class Test extends MY_Model
{
public function __construct ()
{
// do I put this here?!? - or maybe in MY_Model?
// Should it be a singleton?
$this->load->library('dispatcher');
// where do I attach what I want... here?
$this->load->library('emailer');
$this->dispatcher->attach($this->emailer);
// what if I have 50 possible things that might happen
// based on any given event, from adding a user to
// deleting a survey or document? There has got to be a
// way to attach a bunch of observers that trickle
// down to each object, right?
}
public function save_changes ()
{
$this->load->model('user');
$this->user->init($this->session->userdata('user.id'))->save();
}
}
class User extends MY_Model
{
public function __construct ()
{
parent::__construct ();
// do I put this here?!?
$this->load->library('dispatcher'); // just something to call it
}
public function init($id)
{
if($this->_loadUser ($id))
{
$this->dispatcher->notify($this, 'user.loaded');
}
}
public function save($id)
{
if(parent::save())
{
$this->dispatcher->notify($this, 'user.saved');
}
}
}
class Emailer
{
public function update ($caller,$msg)
{
switch ($msg)
{
case 'user.saved':
// send user an email
// re-cache some stuff
// other things that we might want to do, including more of these:
$this->dispatch->notify('user-saved-email-sent');
break;
}
}
}
class Dispatcher
{
public function notify ($caller, $msg) { ...foreach attached do $obj->update($caller,$msg) ...}
public function attach ($obj) { ... }
public function detach ($obj) { ... }
}
I can see how powerful that would be. But I'm not sure how to simplify the setup and attaching of all of these listener/observers.
Maybe I should have a factory to create them all? It just seems like yes, they would be decoupled from the way it currently works, but it seems managing all the different objects that I'd have to 'attached' in each controller or method would become coupled in a different way.
Thanks,
Hans
Your proposed structure would have to be something like:
$this->load->library('observer_factory', 'of'); // factory for creating observers
// Observer_factory would have knowledge/access to different classes which relate
// to the pattern.
$ync = $this->of->getNotifier( $some_variable ) );
$ync->attach( $this->of->getObserver( $some_other_variable ) );
$ync->attach( $this->of->getObserver( $some_final_variable ) );
$ync->someMethod(); // someMethod calls notify
But I wonder about it. You'd have a factory class that slowly becomes all-knowing. It starts usurping the functionality of the Loader. Why load a library when my Observer_factory can handle it by doing exactly the same thing?
I think you're better off with a library or a model that knows what it is supposed to do and is well designed, then adding this class structure. I do not see the gains outweighing the costs.
Related
good day, I am using Tucker-Eric/EloquentFilter Laravel.
I want to filter it by relationship using Models
I want to automate it, instead of using the following:
public function users($users)
{
// dd($users);
return $this->r('users', $users);
}
public function user($user)
{
// dd($user);
return $this->r('user', $user);
}
public function owner($owner)
{
// dd($owner);
return $this->r('owner', $owner);
}
I want to make it one function that based on the relationship
so that I want to add another relationship on the model I don't need anymore to add another function.
Thanks!
We specifically stayed away from the type of implicit functionality you're looking for and opted for explicit filter methods to avoid security issues if/when new relations/properties were added to a model they wouldn't implicitly be available to filter against.
With that, what you're looking for isn't recommended because of the security concerns above but it can still exist if you implement it.
It sounds like the setup method would be the best place to implement it since it would be called first every time ->filter() is called.
public function setup()
{
foreach($this->input() as $key => $val) {
if($this->getModel()->$key() instanceof \Illuminate\Database\Eloquent\Relations\Relation) {
// Your logic here
}
}
}
I'm trying to overwrite some methods in models, and I'm on a mission to avoid overwrites and rewrites of models for maximum compatibility with other modules.
I figured the best way would be to simply decorate models after they are loaded from Magento, however as far as I can tell because of the way the observer pattern in Magento is written it's impossible to accomplish this. ( As Magento always returns the reference to $this ), and the lack of interfaces might also cause trouble later down the road? See this partial of Mage/Core/Model/Abstract.php
/**
* Processing object after load data
*
* #return Mage_Core_Model_Abstract
*/
protected function _afterLoad()
{
Mage::dispatchEvent('model_load_after', array('object'=>$this));
Mage::dispatchEvent($this->_eventPrefix.'_load_after', $this->_getEventData());
return $this;
}
My question boils down to the title, is there a decent way of accomplishing this?, or am I simply stuck with rewrites :(?
The path I would like to take is;
On event [model]_load_after
return new Decorator($event->getObject())
Where the decorator class in my case would be something like;
public function __construct(Mage_Sales_Model_Order_Invoice $model)
{
parent::__construct($model); // sets $this->model on parent class, see below
}
// overwrite the getIncrementId method
public function getIncrementId()
{
return '12345';
}
// partial of parent class
public function __call($method, array $args)
{
return call_user_func_array(array($this->model, $method), $args);
}
And just some pseudo-code for extra clarification;
$model = Mage::getModel('sales/order_invoice')->load(1);
echo get_class($model);
Namespace_Decorator **INSTEAD OF** Mage_Sales_Model_...
echo $model->getIncrementId();
'12345' **INSTEAD OF** '1000001' ( or whatever the format might be )
Thanks for your time reading / commenting, I really hope there actually is a way to accomplish this in a clean fashion without making use of code overrides or rewrites of models.
Edit: extra clarification
Basically what I would like is to return an instance of the Decorator in a few cases, the sales_invoice being one of them and customer the other. So when any load() call is made on these models, it will always return the instance of the Decorator instead of the Model. Only method calls that the decorator overrides would be returned, and any other method calls would "proxied" through __call to the decorated object.
I'm not sure if I got your question right but here goes.
I think you can use the event [model]_load_after and simply do this:
$object = $event->getObject();
$object->setIncrementId('12345');
Or if you want to use a decorator class make it look like this:
public function __construct(Mage_Sales_Model_Order_Invoice $model)
{
parent::__construct($model);
$model->setIncrementId($this->getIncrementId());
}
public function getIncrementId()
{
return '12345';
}
I know that this is not exactly a decorator pattern but it should work.
I know that when adding a new method to the 'decorator' class you need to add it to attach data to the main model.
This is just my idea. I haven't got an other.
[EDIT]
You can try to rewrite the load method on the object to make it return what you need. But I wouldn't go that way. You can end up screwing a lot of other things.
I don't think there is an other way to do it because load always returns the current object no mater what you do in the events dispatched in the method. see Mage_Core_Model_Abstract::load()
public function load($id, $field=null)
{
$this->_beforeLoad($id, $field);
$this->_getResource()->load($this, $id, $field);
$this->_afterLoad();
$this->setOrigData();
$this->_hasDataChanges = false;
return $this;
}
By making it return new Decorator($this), you might achieve what you need, but just make sure that when calling $model->doSomething() and doSomething() is not a method in your decorator you still end up calling the original method on the model.
I'm building a Zend Framework application wherein the Model layer is separated into Services and Models. Controller actions call service methods, which in turn call model methods.
For instance: LanguagesController::addAction() checks if a form is submitted and valid. If so, it passes the form data to Service_Language::add(), where some business logic is applied to the data before calling Model_Language::add(), which effectively adds the record to the database.
This means that most controller actions will need an instance of a service class, and most methods in the service class will need an instance of a model class.
I used to do it like this (example of a Service class)
class Service_Language
{
public function add()
{
$languageModel = new Model_Language;
// perform some business logic and add record
}
public function edit()
{
$languageModel = new Model_Language;
// perform some business logic and edit record
}
public function delete()
{
$languageModel = new Model_Language;
// perform some business logic and delete record
}
}
Not only does it become cumbersome, in more complex applications where your controller actions call multiple Service methods, there's going to be multiple instances of the same Model class, which is just unnecessary.
A colleague told me to look into two options:
keep a Model instance in a property of the Service
keep a Model instance in Zend_Registry
I think the best solution would be the first option. The reason being that Zend_Registry acts as a global container. We don't want our Model instances to be available in our Controller actions, it's bad architecture. What are your opinions on this?
The first option could be implemented as follows:
class Service_Language
{
protected $_model = null;
function setModel()
{
$this->_model = new Model_Language();
}
function getModel()
{
if($this->_model == null)
{
$this->setModel();
}
return $this->_model;
}
public function add()
{
$languageModel = $this->getModel();
// perform some business logic and add
}
}
From your explanation it sounds like your services classes require tightly coupled models.
In which case I don't think public a public getter/setter for your model is necessary - would there ever realistically be a situation where you would need to set another model for the service?
In which case, assigning the model to a property makes sense - why not do this in the constructor?
class Service_Language
{
protected $_model = null;
public function __construct()
{
$this->_model = new Model_Language();
}
public function add()
{
// perform some business logic and add
$this->_model->add($data);
}
public function edit()
{
// perform some business logic and add
$this->_model->edit($data);
}
}
The constructor would have been a good option, but not every method in the service layer needs to have a model instance to do its job, so I ended up doing it like this. I'm relatively new to OOP programming, so I'm wondering if this is a good solution. Any thoughts are more than welcome.
class Service_Language
{
protected $_model = null;
protected function setModel()
{
$this->_model = new Model_Language();
}
protected function getModel()
{
if($this->_model == null)
{
$this->setModel();
}
return $this->_model;
}
// Function where model is needed
public function add($data)
{
// Perform some business logic
$this->getModel()->add($data);
// return something
}
// Function where no model is needed
public function add($data)
{
// Perform some business logic
// return something
}
}
I'm having Codeigniter object scope confusion.
Say I load a model in a controller:
$this->load->model('A');
$this->A->loadUser(123); // loads user with ID 123
// output of $this->A now shows user 123
$this->load->model('B');
$this->B->examineUser ();
// output of $this->A now shows user 345
class B extends Model
{
public function examineUser ()
{
$this->load->model('A');
$this->A->loadUser(345); // loads user with ID 345
}
}
I would have thought that $this->A would be different from $this->B->A but they are not. What is the best solution to this issue? It appears the ->load->model('A') in the examineUser () method does nothing because it was loaded in the controller. Then the call to loadUser () inside that method overwrites the stored properties of $this->A. This seems like a bugfest waiting to happen. If I needed global models, I would have use static classes. What I wanted was something scoped pretty much locally to the model object I was in.
Is there a way I can accomplish this but not go way outside of CI's normal way of operating?
Followup/related:
Where do most people put there "->load->model" calls? All at the beginning of a controller action? I figured it would be easier -- though perhaps not excellent programming from a dependency injection perspective -- to load them in the model itself (construct or each method).
Whenever you use the Loader Class ($this->load->), it will load the object into the main CI object. The CI object is the one you keep referring to as $this->. What you've done is load model A twice into the CI object.
Essentially, all object loaded using the Loader class goes into a single global scope. If you need two of the same type, give them different names, as per $this->load->model('A','C'). I don't know of any way around it unless you revert to using bog-standard PHP.
In my team's code, we generally load the models in the controller's constructor, then load the data to send to the view in the function, often _remap().
This is not how the loader works sadly. CodeIgniter implements a singleton pattern, which will check to see if the class is included, instantiated and set to $this->A then will be ignored if loaded again. Even if you are inside a model, $this->A will be referenced to the super-instance via the __get() in class Model. Alis it, or just do:
class B extends Model
{
public function examineUser ()
{
$user = new A;
$user->loadUser(345); // loads user with ID 345
}
}
Here's what I've decided to do, please comment if you have advice:
I've extended the CI Loader class:
<?php
class SSR_Loader extends CI_Loader
{
function __construct()
{
parent::__construct ();
}
/**
* Model Retriever
*
* Written by handerson#executiveboard.com to create and return a model instead of putting it into global $this
*
* Based on original 2.0.2 CI_Loader::model ()
*
*/
function get_model($model)
{
if (empty ($model))
{
return;
}
$name = basename ($model);
if (!in_array($name, $this->_ci_models, TRUE))
{
$this->model ($model);
}
$name = ucfirst($name);
return new $name ();
}
}
Do any CI guru's see a problem with that before I invest time in changing my code a bit to accept the return obj, ala:
// in a controller:
public function test ($user_id=null)
{
$this->_logged_in_user = $this->load->get_model ('/db/users');
$this->_viewed_user = $this->load->get_model ('/db/users');
$this->_logged_in_user->load($this->session->userdata ('user.id'));
$this->_viewed_user->load($user_id);
}
I could also do private $_logged_in_user to make it available in the controller but positively force it to be limited to just the current controller and not spill anywhere else, or I could just do $_logged_in_user = $this->load->get_model ('/db/users'); and limit it to just the current method, which is probably what I'll do more often.
This seems like a pretty straightforward way to "fix" this issue (I say "fix" b/c it's not really a bug, just a way of doing things that I think is a bad idea). Anyone see any flaws?
Exactly how skinny should a controller be? I understand putting all of the business logic inside the models, but what about other things.
For example, say I was writing a blog site where each user can have multiple posts. Currently, the user would create posts by visiting the posts controller and running the create action. Here is a little sample of what would happen currently.
class Controller_Post extends Controller {
function action_create() {
if ( ! empty($_POST)) {
$post = new Model_Post;
$post->user_id = $this->logged_in_user->id;
$post->values($_POST);
if ( ! $post->create()) {
echo 'Error';
}
else
{
echo 'Saved';
}
}
}
}
My question is, what would stop me from putting the above logic in the user model, like so.
class Model_User extends Model {
function create_post($post) {
$post = Model::factory('post')->values($post);
$post->user_id = $this->id;
if ( ! $post->create()) {
return FALSE;
}
else
{
return TRUE;
}
}
}
If it were done this way, the controller would be even smaller than what I put. It makes more sense to me because the user is the one creating the post, so I think it should be in the user model as opposed to the controller.
If it helps, Im using the Kohana framework.
Thanks
Controllers should be directing traffic. Models are for where your business logic goes, so in general your second example would be "correct mvc".
Basically what a controller should be doing is requesting input, telling models to change state (they do the actual state change themselves), and determining which view to display (if any).
I have many controllers which simply are like so:
class Controller_Foobar extends Controller
{
public function action_index() {}
}
And if they need to process $_POST input, they grab that data, and send it off to the model, then the view.
Keeping all that logic inside your models lets you easily reuse it, and it's more maintainable and testable.