Why would one use the Publish/Subscribe pattern (i

问题:

So, a colleague introduced me to the publish/subscribe pattern (in JS/jQuery), but I'm having a hard time getting to grips with why one would use this pattern over 'normal' JavaScript/jQuery.

For example, previously I had the following code...

$container.on('click', '.remove_order', function(event) {
    event.preventDefault();
    var orders = $(this).parents('form:first').find('div.order');
    if (orders.length > 2) {
        orders.last().remove();
    }
});

And I could see the merit of doing this instead, for example...

removeOrder = function(orders) {
    if (orders.length > 2) {
        orders.last().remove();
    }
}

$container.on('click', '.remove_order', function(event) {
    event.preventDefault();
    removeOrder($(this).parents('form:first').find('div.order'));
});

Because it introduces the ability to re-use the removeOrder functionality for different events etc.

But why would you decide to implement the publish/subscribe pattern and go to the following lengths, if it does the same thing? (FYI, I used jQuery tiny pub/sub)

removeOrder = function(e, orders) {
    if (orders.length > 2) {
        orders.last().remove();
    }
}

$.subscribe('iquery/action/remove-order', removeOrder);

$container.on('click', '.remove_order', function(event) {
    event.preventDefault();
    $.publish('iquery/action/remove-order', $(this).parents('form:first').find('div.order'));
});

I've read about the pattern for sure, but I just can't imagine why this would ever be necessary. The tutorials I've seen that explain how to implement this pattern only cover just as basic examples as my own.

I imagine that the pub/sub's usefulness would make itself apparent in a more complex application, but I can't imagine one. I'm afraid that I am completely missing the point; but I'd like to know the point if there is one!

Could you explain succinctly why and in what situations this pattern is advantageous? Is it worth using the pub/sub pattern for code snippets like my examples above?

回答1:

It’s all about loose coupling and single responsibility, which goes hand to hand with MV* (MVC/MVP/MVVM) patterns in JavaScript which are very modern in the last few years.

Loose coupling is an Object-oriented principle in which each component of the system knows its responsibility and doesn’t care about the other components (or at least tries to not care about them as much as possible). Loose coupling is a good thing because you can easily reuse the different modules. You’re not coupled with the interfaces of other modules. Using publish/subscribe you’re only coupled with the publish/subscribe interface which is not a big deal – just two methods. So if you decide to reuse a module in a different project you can just copy and paste it and it’ll probably work or at least you won’t need much effort to make it work.

When talking about loose coupling we should mention the separation of concerns. If you’re building an application using an MV* architectural pattern you always have a Model(s) and a View(s). The Model is the business part of the application. You can reuse it in different applications, so it’s not a good idea to couple it with the View of a single application, where you want to show it, because usually in the different applications you have different views. So it’s a good idea to use publish/subscribe for the Model-View communication. When your Model changes it publishes an event, the View catches it and updates itself. You don’t have any overhead from the publish/subscribe, it helps you for the decoupling. In the same manner you can keep your application logic in the Controller for example (MVVM, MVP it’s not exactly a Controller) and keep the View as simple as possible. When your View changes (or the user clicks on something, for example) it just publishes a new event, the Controller catches it and decides what to do. If you are familiar with the MVC pattern or with MVVM in Microsoft technologies (WPF/Silverlight) you can think of the publish/subscribe like the Observer pattern. This approach is used in frameworks like Backbone.js, Knockout.js (MVVM).

Here is an example:

//Model
function Book(name, isbn) {
    this.name = name;
    this.isbn = isbn;
}

function BookCollection(books) {
    this.books = books;
}

BookCollection.prototype.addBook = function (book) {
    this.books.push(book);
    $.publish('book-added', book);
    return book;
}

BookCollection.prototype.removeBook = function (book) {
   var removed;
   if (typeof book === 'number') {
       removed = this.books.splice(book, 1);
   }
   for (var i = 0; i < this.books.length; i += 1) {
      if (this.books[i] === book) {
          removed = this.books.splice(i, 1);
      }
   }
   $.publish('book-removed', removed);
   return removed;
}

//View
var BookListView = (function () {

   function removeBook(book) {
      $('#' + book.isbn).remove();
   }

   function addBook(book) {
      $('#bookList').append('<div id="' + book.isbn + '">' + book.name + '</div>');
   }

   return {
      init: function () {
         $.subscribe('book-removed', removeBook);
         $.subscribe('book-aded', addBook);
      }
   }
}());

Another example. If you don’t like the MV* approach you can use something a little different (there’s an intersection between the one I’ll describe next and the last mentioned). Just structure your application in different modules. For example look at Twitter.

If you look at the interface you simply have different boxes. You can think of each box as a different module. For example you can post a tweet. This action requires the update of a few modules. Firstly it has to update your profile data (upper left box) but it also has to update your timeline. Of course, you can keep references to both modules and update them separately using their public interface but it’s easier (and better) to just publish an event. This will make the modification of your application easier because of looser coupling. If you develop new module which depends on new tweets you can just subscribe to the “publish-tweet” event and handle it. This approach is very useful and can make your application very decoupled. You can reuse your modules very easily.

Here is a basic example of the last approach (this is not original twitter code it’s just a sample by me):

var Twitter.Timeline = (function () {
   var tweets = [];
   function publishTweet(tweet) {
      tweets.push(tweet);
      //publishing the tweet
   };
   return {
      init: function () {
         $.subscribe('tweet-posted', function (data) {
             publishTweet(data);
         });
      }
   };
}());


var Twitter.TweetPoster = (function () {
   return {
       init: function () {
           $('#postTweet').bind('click', function () {
               var tweet = $('#tweetInput').val();
               $.publish('tweet-posted', tweet);
           });
       }
   };
}());

For this approach there's an excellent talk by Nicholas Zakas. For the MV* approach the best articles and books I know of are published by Addy Osmani.

Drawbacks: You have to be careful about the excessive use of publish/subscribe. If you’ve got hundreds of events it can become very confusing to manage all of them. You may also have collisions if you’re not using namespacing (or not using it in the right way). An advanced implementation of Mediator which looks much like an publish/subscribe can be found here https://github.com/ajacksified/Mediator.js. It has namespacing and features like event “bubbling” which, of course, can be interrupted. Another drawback of publish/subscribe is the hard unit testing, it may become difficult to isolate the different functions in the modules and test them independently.

回答2:

The main goal is to reduce coupling between the code. It's a somewhat event-based way of thinking, but the "events" aren't tied to a specific object.

I'll write out a big example below in some pseudo code that looks a bit like JavaScript.

Let's say we have a class Radio and a class Relay:

class Relay {
    function RelaySignal(signal) {
        //do something we don't care about right now
    }
}

class Radio {
    function ReceiveSignal(signal) {
        //how do I send this signal to other relays?
    }
}

Whenever radio receives a signal, we want a number of relays to relay the message in some way. The number and types of relays can differ. We could do it like this:

class Radio {
    var relayList = [];

    function AddRelay(relay) {
        relayList.add(relay);
    }

    function ReceiveSignal(signal) {
        for(relay in relayList) {
            relay.Relay(signal);
        }
    }

}

This works fine. But now imagine we want a different component to also take part of the signals that the Radio class receives, namely Speakers:

(sorry if the analogies aren't top notch...)

class Speakers {
    function PlaySignal(signal) {
        //do something with the signal to create sounds
    }
}

We could repeat the pattern again:

class Radio {
    var relayList = [];
    var speakerList = [];

    function AddRelay(relay) {
        relayList.add(relay);
    }

    function AddSpeaker(speaker) {
        speakerList.add(speaker)
    }

    function ReceiveSignal(signal) {

        for(relay in relayList) {
            relay.Relay(signal);
        }

        for(speaker in speakerList) {
            speaker.PlaySignal(signal);
        }

    }

}

We could make this even better by creating an interface, like "SignalListener", so that we only need one list in the Radio class, and always can call the same function on whatever object we have that wants to listen to the signal. But that still creates a coupling between whatever interface/base class/etc we decide on and the Radio class. Basically whenever you change one of the Radio, Signal or Relay class you have to think about how it could possibly affect the other two classes.

Now let's try something different. Let's create a fourth class named RadioMast:

class RadioMast {

    var receivers = [];

    //this is the "subscribe"
    function RegisterReceivers(signaltype, receiverMethod) {
        //if no list for this type of signal exits, create it
        if(receivers[signaltype] == null) {
            receivers[signaltype] = [];
        }
        //add a subscriber to this signal type
        receivers[signaltype].add(receiverMethod);
    }

    //this is the "publish"
    function Broadcast(signaltype, signal) {
        //loop through all receivers for this type of signal
        //and call them with the signal
        for(receiverMethod in receivers[signaltype]) {
            receiverMethod(signal);
        }
    }
}

Now we have a pattern that we are aware of and we can use it for any number and types of classes as long as they:

• are aware of the RadioMast (the class handling all the message passing)
• are aware of the method signature for sending/receiving messages

So we change the Radio class to its final, simple form:

class Radio {
    function ReceiveSignal(signal) {
        RadioMast.Broadcast("specialradiosignal", signal);
    }
}

And we add the speakers and the relay to the RadioMast's receiver list for this type of signal:

RadioMast.RegisterReceivers("specialradiosignal", speakers.PlaySignal);
RadioMast.RegisterReceivers("specialradiosignal", relay.RelaySignal);

Now the Speakers and Relay class has zero knowledge of anything except that they have a method that can receive a signal, and the Radio class, being the publisher, is aware of the RadioMast that it publishes signals to. This is the point of using a message-passing system like publish/subscribe.

回答3:

The other answers have done a great job in showing how the pattern works. I wanted to address the implied question "what is wrong with the old way?" as I've been working with this pattern recently, and I find it involves a shift in my thinking.

Imagine we have subscribed to an economic bulletin. The bulletin publishes a headline: "Lower the Dow Jones by 200 points". That would be an odd and somewhat irresponsible message to send. If however, it published: "Enron filed for chapter 11 bankrupcy protection this morning", then this is a more useful message. Note that the message may cause the Dow Jones to fall 200 points, but that is another matter.

There is a difference between sending a command, and advising of something that has just happened. With this in mind, take your original version of the pub/sub pattern, ignoring the handler for now:

$.subscribe('iquery/action/remove-order', removeOrder);

$container.on('click', '.remove_order', function(event) {
    event.preventDefault();
    $.publish('iquery/action/remove-order', $(this).parents('form:first').find('div.order'));
});

There is already an implied strong coupling here, between the user-action (a click) and the system-response (an order being removed). Effeectively in your example, the action is giving a command. Consider this version:

$.subscribe('iquery/action/remove-order-requested', handleRemoveOrderRequest);

$container.on('click', '.remove_order', function(event) {
    event.preventDefault();
    $.publish('iquery/action/remove-order-requested', $(this).parents('form:first').find('div.order'));
});

Now the handler is responding to something of interest that has happened, but is under no obligation to remove an order. In fact, the handler can do all sorts of things not directly related to removing an order, but still maybe relevant to the calling action. For example:

handleRemoveOrderRequest = function(e, orders) {
    logAction(e, "remove order requested");
    if( !isUserLoggedIn()) {
        adviseUser("You need to be logged in to remove orders");
    } else if (isOkToRemoveOrders(orders)) {
        orders.last().remove();
        adviseUser("Your last order has been removed");
        logAction(e, "order removed OK");
    } else {
        adviseUser("Your order was not removed");
        logAction(e, "order not removed");
    }
    remindUserToFloss();
    increaseProgrammerBrowniePoints();
    //etc...
}

The distinction between a command and a notification is a useful distinction to make with this pattern, IMO.

回答4:

So that you don't have to hardcode method / function calls, you just publish the event without caring who listens. This makes the publisher independent from subscriber, reducing dependency (or coupling, whatever term you prefer) between 2 different parts of the application.

Here are some disadvantages of coupling as mentioned by wikipedia

Tightly coupled systems tend to exhibit the following developmental characteristics, which are often seen as disadvantages:

1. A change in one module usually forces a ripple effect of changes in other modules.
2. Assembly of modules might require more effort and/or time due to the increased inter-module dependency.
3. A particular module might be harder to reuse and/or test because dependent modules must be included.

Consider something like an object encapsulating business data. It has hard coded method call to update the page whenever the age is set:

var person = {
    name: "John",
    age: 23,

    setAge: function( age ) {
        this.age = age;
        showAge( age );
    }
};

//Different module

function showAge( age ) {
    $("#age").text( age );
}

Now I cannot test the person object without also including the showAge function. Also, if I need to show the age in some other GUI module as well, I need to hardcode that method call in .setAge, and now there is dependencies for 2 unrelated modules in the person object. It's also just hard to maintain when you see those calls being made and they are not even in the same file.

Note that inside the same module, you can of course have direct method calls. But business data and superficial gui behavior should not reside in the same module by any reasonable standards.

回答5:

PubSub implementation is commonly seen in where there is -

1. There is a portlet like implementation where there are multiple portlets that communicate with the help of an event bus. This helps in creating in aync architecture.
2. In a system marred by tight coupling, pubsub is a mechanism which helps in communicating between various modules.

Example code -

var pubSub = {};
(function(q) {

  var messages = [];

  q.subscribe = function(message, fn) {
    if (!messages[message]) {
      messages[message] = [];
    }
    messages[message].push(fn);
  }

  q.publish = function(message) {
    /* fetch all the subscribers and execute*/
    if (!messages[message]) {
      return false;
    } else {
      for (var message in messages) {
        for (var idx = 0; idx < messages[message].length; idx++) {
          if (messages[message][idx])
            messages[message][idx]();
        }
      }
    }
  }
})(pubSub);

pubSub.subscribe("event-A", function() {
  console.log('this is A');
});

pubSub.subscribe("event-A", function() {
  console.log('booyeah A');
});

pubSub.publish("A"); //executes the methods.

回答6:

The paper "The Many Faces of Publish/Subscribe" is a good read and one thing that they emphasize is the decoubling in three "dimensions". Here is my crude summary but please see the paper instead.

1. Space decoubling. The interacting parties do not need to know eachother. The publisher do not know who is listening, how many or what they are doing with the event. The subscribers do not know who is producing these events or how many producers there are etc.
2. Time decoubling. The interacting parties do not need to be active at the same time during the interaction. E.g. a subscriber might be disconnected while a publisher is publishing some events, it can react to it when it becomes online.
3. Synchronization decoupling. Publishers are not blocked while producing events and subscribers can be asynchronously notified through callbacks whenever an event they have subcribed to arrives.

回答7:

Simple answer The original question was looking for a simple answer. Here's my attempt.

Javascript doesn't provide any mechanism for code objects to create their own events. So you need a kind of event mechanism. the Publish / subscribe pattern will answer this need, and it's up to you to choose a mechanism that best suits your own needs.

Now we can see a need for the pub/sub pattern, then would you rather have to handle DOM events differently from how you handle your pub/sub events? For the sake of reducing complexity, and other concepts such as separation of concerns (SoC) you might see the benefit of everything being uniform.

So paradoxically, more code creates a better separation of concerns, which scales well up to very complex web pages.

I hope someone finds this a good enough discussion without going into detail.