Sunday, August 3, 2014

No Need to $Watch AngularJS “Controller As”

In a previous blog post I discussed how to use Angular’s extend to improve code quality and reusability. In my example I used the new controller as syntax. I see a lot of discussion online about this approach. I like it because it allows you to treat your controllers as pure JavaScript objects rather than glorified grab bags that do nothing but manipulate $scope. The biggest complaint I read is that you still have to take a dependency on $scope when you want to watch properties. Or, do you?

From lessons learned working on a large Angular project (by “large” I mean a team of 25+ developers distributed around the world, with a code base featuring 80,000+ lines of TypeScript code with hundreds of controllers, filters, and services) one mistake I made early on was depending too much on scope. For example, I might assume a detail page inherited the master page’s scope. This dependency on the hierarchy made it difficult to refactor pages, so I quickly learned that communication of properties should take place via components and services and not be based on implied scope.

Watches are similar. You have to ask yourself what you are watching for, then decide if using an actual $watch is worth it. A $watch introduces significant overhead and fires every digest loop. It will be called multiple times when other code mutates the data model, so you want to conserve your watches as much as possible. How can do you do this?

One example I covered in my Angular Debugging and Performance video. The scenario is a paged list. I have one million items in the list and am showing a few per page. You likely know from past experience with paging that there are several variables to calculate, such as how many total pages exist and where a “page” fits into the range of indexed items that represent that page. To handle the paging I created a controller that keeps track of the full list, then exposes a “display list” with the current page. Here is the basic definition with the code to generate the million entries:

function Controller() {
    var idx = 0;
    this.list = [];
    this.displayList = [];
    while (idx < 1000000) {
        this.list.push(Math.random());
        idx += 1;
    }
    this.refreshPages();
}

The controller keeps track of several variables, and they are all recomputed in the refreshPages function that should fire any time the current page changes. Here is the initial definition of the controller and that method:

angular.extend(Controller.prototype, {
    pageSize: 20,
    _currentPage: 1,
    nextPage: 2,
    previousPage: 0,
    currentIndex: 0,
    totalPages: 0,
    refreshPages: function () {
        var curIdx = this.currentIndex;
        this.totalPages = Math.ceil(this.list.length / this.pageSize);
        this.currentIndex = this.pageSize * (this._currentPage - 1);
        this.previousPage = this._currentPage - 1;
        this.nextPage = this._currentPage + 1;
        if (curIdx !== this.currentIndex || this.displayList.length === 0) {
            while (this.displayList.length > 0) {
                this.displayList.pop();
            }
            for (curIdx = this.currentIndex;
                 curIdx < this.list.length && curIdx < this.currentIndex + this.pageSize;
                 curIdx += 1) {
                this.displayList.push(this.list[curIdx]);
            }
        }
    }
});

You may notice the current page is defined as a private property. Why not expose it and then use a $watch? The answer is simple. I control my model, so I don’t need to watch it to know when it is mutated. Instead, I can take advantage of pure JavaScript and expose the current page as a property that updates the necessary variables whenever it changes. After that happens, I quickly clear out the exposed array for the page and repopulate it based on the newly computed variables. Here is the solution that doesn’t involve adding an unnecessary watch but instead uses Object.defineProperty:

Object.defineProperty(Controller.prototype, "currentPage", {
    enumerable: true,
    configurable: true,
    get: function () { return this._currentPage; },
    set: function (val) {
        this._currentPage = val;
        this.refreshPages();
    }
});

Now the property is defined on the controller using pure JavaScript, and can be bound like any other property. Here is markup for the button to navigate to the previous page. It simply decrements the currentPage property and is disabled when you are on the first page.

<button data-ng-click="ctrl.currentPage = ctrl.currentPage-1"
          data-ng-disabled="ctrl.currentPage == 1">
    &lt;
</
button>

If I wanted to optimize even further, I could add an additional condition to ensure the value is actually different before refreshing the pages. The result is a huge list paged efficiently using “controller as” syntax. You can see the running example here: long paged list in AngularJS. The full source is here: long paged list in AngualrJS source code.

The bottom line: I prefer the “controller as” syntax because it allows me to define controllers as pure JavaScript objects with minimal dependencies. I don’t have to explicitly concern myself with $scope and data-binding, and instead can treat the controller itself as a view model and know the exposed properties are available for binding.

Instead of using $watch I use built-in JavaScript features to manage my model. This has the added advantage of making the component easier to test (in this example, you can test the controller without any dependency on Angular whatsoever). If I am concerned about the value changing from another controller, I set up a service for communication rather than adding the watch and again maintain control over the changes without the overhead of being called every digest loop.

$watch my latest video to learn more about AngularJS Debugging and Performance.

Sunday, July 27, 2014

Using AngularJS to Extend Your Code Quality

The AngularJS API provides a function named extend that can help improve your code quality and efficiency. I always look for ways to improve quality, increase efficiency, reduce risk and eliminate ritual and ceremony when I am developing software. Perhaps the simplest way to express this is the DRY principle (Don’t Repeat Yourself). I prefer a refactoring-driven approach to this principle. Instead of trying to anticipate what might be needed in a framework, I simply evolve it and refactor when I see an opportunity to improvement. Oftentimes Angular’s extend function is a part of that refactoring.

Assume I am writing a page that allows the user to click two buttons to produce a list of categories and products. Here is a screenshot with the categories loaded and the products waiting for the user to request them.

image

The source for data is exposed via the example API at OData.org. To encapsulate the call for categories, I create a component and register it with Angular that looks like this:

function CategoriesService($http, $q) {
     this.$http = $http;
     this.$q = $q; } 




CategoriesService.prototype.get = function () {
     var deferral = this.$q.defer();
     this.$http.get('http://services.odata.org/V4/OData/OData.svc/')
         .success(function (response) {
             deferral.resolve(response.value);
         })
         .error(function (err) {
             deferral.reject(err);
         });
     return deferral.promise; }; app.service('categories', CategoriesService);

Next, I move on to products. It turns out that the products service looks almost identical to the categories service! I don’t want to repeat myself so it’s time to refactor the code to take advantage of the principle of inheritance. I encapsulate the base functionality for dealing with the service in a base class, then inherit from that and specify what’s unique between products and categories. The base class looks like this:

var baseOData = {
     $http: {},
     $q: {},
     baseUrl: 'http://services.odata.org/V4/OData/OData.svc/',
     entity: '',
     get: function () {
         var defer = this.$q.defer();
         this.$http.get(this.baseUrl + this.entity)
             .success(function (response) {
                 defer.resolve(response.value);
             })
             .error(function (err) {
                 defer.reject(err);
             });
         return defer.promise;
     } };

Notice I’ve captured everything that is repeated: the base portion of the URL and the wrapper that handles the promise so the result is returned and the consuming class doesn’t have to understand how the collection is implemented in the API. The only difference I found between the categories and products is the name of the entity specified in the URL, so I expose that with a property on the base class that can be overridden by the service implementation. Using this base class I implement the category service like this:

function CategoriesService($http, $q) {
     this.$http = $http;
     this.$q = $q;
     this.entity = 'Categories'; } angular.extend(CategoriesService.prototype, baseOData); app.service('categories', CategoriesService);

The shell for the categories service simply sets up the dependencies and registers the entity because the base class holds all of the common functionality. The call to extend automatically applies the properties and functions from the base definition to the category service. Notice that I am extending the prototype; this will ensure that the properties and functions are part of any instance that is created. Angular will also bind the properties and functions so this refers to the instance itself.

The products service is then implemented the same way with a different entity specified. Although I could provide a service that takes in the entity as a parameter and returns the promise (even less code), I may want to have specific properties or methods that are unique to the category and/or product implementation. I really don’t know yet so I keep them as separate components and will refactor them down to a single service if the pattern doesn’t change.

You can call extend multiple times or pass a collection of objects. This enables your components to inherit from multiple “base classes.” Another way to look at it is that you can define behaviors and apply those behaviors to the class.

I prefer the “controller as” syntax for my controller definitions. In this example the controller takes a dependency on the product and category service and exposes methods to request them that are bound to buttons. The initial implementation looked like this:

function Controller(products, categories) {
     this.productService = products;
     this.categoryService = categories;
     this.products = [];
     this.categories = []; } Controller.prototype.getCategories = function () {
     var _this = this;
     this.categoryService.get().then(function (result) {
         _this.categories = result;
     }); }; Controller.prototype.getProducts = function () {
     var _this = this;
     this.productService.get().then(function (result) {
         _this.products = result;
     }); }; app.controller('exampleCtrl', Controller);

Wouldn’t it be nice if to encapsulate the controller functionality in a single definition? Actually, that is possible! Using extend I simplified the declaration for my controller by combining the functions into a single object definition. I removed the initialization of the product and categories list from the constructor and moved them into a consolidated definition that looks like this:

angular.extend(Controller.prototype, {
     products: [],
     categories: [],
     getCategories: function () {
         var _this = this;
         this.categoryService.get().then(function (result) {
             _this.categories = result;
         });
     },
     getProducts: function () {
         var _this = this;
         this.productService.get().then(function (result) {
             _this.products = result;
         });
     } });

This convention makes it easier to group related functionality together and ensure there is a consistent implementation of this. The implementation is very similar to the way that TypeScript handles inheritance.

As you can see, although the documentation for extend is quite simple, the functionality can be quite powerful. View the source code and full working example here.

Sunday, July 20, 2014

AngularJS Debugging and Performance

Just over a week ago an Angular project went live after several years of development. Our team peaked at over 25 developers distributed around the world. The source ended up at over 80,000 lines of TypeScript code (although I jokingly note this ends up as just one line of minified JavaScript) and includes hundreds of controllers, services, views, etc. You don’t get that far in an enterprise app without learning a thing or two about debugging and performance.

I just released my video “AngularJS: Debugging and Performance” as the 12th lesson in my course, Mastering AngularJS. This course covers everything you need to know to build business apps in Angular. Topics start out with fundamentals that provides an end-to-end overview, followed by a deep dive into scope and the digest loop. Learn about advanced filters, how dependency injection works, and what the built-in Angular services are. Lessons cover consuming web services from Angular (using both $http and ngResource), handling routing with ngRoute and ui-router, fundamentals of directives and advanced directives. I cover testing using Jasmine and this latest video is all about debugging and performance.

The debugging and performance lesson starts by covering some of the built-in Angular services like $exceptionHandler and $log, and how to leverage the JavaScript debugger key word. I then dive into how to troubleshoot Angular applications by inspecting the HTML DOM and using that to dive into source code and unravel the HTML nodes generated by templates. I cover the Chrome Batarang extension and then go into using Chrome Developer Tools to troubleshoot a memory leak. Learn how to use the Zone library to instrument your apps, how to avoid the overhead of extra watches when data-binding to render static data, and how to deal with extremely long (one million records or more) lists on the client. I include tips for using the right directives to condition rendering of the DOM, how to minimize overhead of watches, and how to take advantage of lazy-loading or instantiation of new components “on the fly.”

The course includes full access to working examples and source code. If you are working with Angular, I know you will want to check this lesson out.