technical musings of a caffeine converter


Best Approach? EF5, Design Patterns, MVC and Knockout

One of the things I particularly like about Silverlight is that its rich client side model enables easy manipulation of data in a decoupled, data-bound fashion without the need to perform projection and/ or switching between to two different languages such as C# and JavaScript. What I don’t enjoy is the fact that in any distributed system that involves Silverlight, you will have to use some sort of asynchronous data-grab architecture so any lazy-loading or change tracking that may have been available on the server side gets lost. When working with MVC however, I find that even though I may have the full capabilities of EF at my disposal within the controller, the likelihood that I will push the same EF domain model classes into my view and few and far between. The main reasons:

1. Views usually only require a subset of the data so I don’t want to bulk up the response with more than I need.

2. JSON serialisation does not handle proxy version of classes designed to handle change tracking.

With point 2, you may come across this if you use Web API controllers and hit the error,

The type System.Data.Entity.DynamicProxies.SomeClass_SomeGUID was not expected…

The reason why point 2 is so important for me is because, coming back to Asp.Net from Silverlight, the desire to mimimise not just the amount of return trips to the server but to minimise the amount of data sent as well is too strong to ignore. With SL, the view is already on the client and all that happens is the data that is required is asked for and given. With vanilla MVC, you can make a request that returns the same view just with different data and sometimes, the data can be a subset of what you already had. So I wanted to get to grasps with an MVC approach that would satisfy these wishes and give me a chance to get to know MVC as it is. To this end, I opted to take a look at KnockoutJs as a) it offers what I am looking for and b) it seems to be the most prevalent of MVVM js libraries out there right now. I’m not saying it’s the best or that there aren’t other approaches but as contractor I need to think not just about keeping ahead of the curve, but, in this instance, getting myself up-to-date with it.

I have spent many an hour looking at various approaches covering all aspects of MVC. From the vanilla flavour to a full blown SPA (courtesy of Mr John Papa), I have settled on something in between – I don’t want a clunky ‘return a view every time’ approach, but I don’t feel ready for the single page approach where both the view and data is dynamically grabbed using JavaScript. I love JavaScript for its flexibility but struggle with it because of its flexibility! We C# developers do enjoy the seat-belt constraints of our strongly typed language :)

Before I had even settled on a web approach, I knew that I would want to be able to target more than one front-end and not just via the web so with that in mind, I wanted to abstract my DAL and not just by using EF. I opted to use a POCO domain model with a Code First approach and encapsulate that within a set of Repositories that themselves were house within a UnitOfWork class. I then created a set of services to manage specific areas of logic (but could still overlap), i.e. a CompanyService responsible for basic CRUD ops as well as more specific methods. It is these services that are inject into my controllers via Ninject. Whilst this abstraction provides me with more testable code and allows me to control what operations are performed and how they are performed, it does mean that I lose the more finer points of some of the EF functionality such as Eager Loading etc. However, I have found in development, if something is available, sooner or later, someone will come along and decide to take it, regardless of whether or not it is always needed. Forcing developers through the route of declaring specific business functionality in the service for their needs or simply allowing them to delay load the required resources as and when required minimises the risk of data bloat where each developer adds on their own specific data needs until everyone is complaining of performance issues.

So my layers up to the UI look a little like this:

Domain Model Entities + other common elements (Interfaces, enums etc.)
EF DbContext DbSet<Company>, Fluent API Configurations
Repositories Using generic DataEntityRepository<T>
Unit Of Work Combines all repositories and provides Commit () method
Services CompanyService – i.e. GetAllCompanies()

note: Services is analogous to BLL, not a web or windows service

So, on to the MVC app. I have taken an empty application and applied the Twitter Bootstrap templating to it. I created my own authentication mechanism through the services, based on Forms Authentication. I added KnockoutJs through Nuget and set about creating two initial views – one to list all companies, and one to view details.

Standard controller approach in MVC would be to call the company service, grab all companies, then either inject the result as the model into the View to return or project the results into a DTO first. But as this means that my model is now static on the client side, if I want to manipulate it, I need to return trip to the server.

Take for instance the following scenario:

1. My Index() method in the home controller returns a list of all companies represented as DTOs.

2. My index view has a search text box to allow the user to filter the companies based on a name value.

3. To filter the companies, the user must enter a value, then hit the filter/ search button.

4. The home controller has an Index (string q) method that gets all companies where the name contains the string represented by q. It then returns the Index view again but with a subset of the original data as the model.

It’s a little clunky, isn’t it?

What I really want is for the filter text box to immediately filter the current list as the user types. And for this, I need my model to be available client side.

There are a couple of approaches to do this with Knockout but first off, I went down a web API with an Ajax call approach:

1. HomeController.Index() returns nothing but the view.

2. A new ApiController is created, CompanyController which contains the method, GetAllCompanies and returns an IEnumerable of the DTO representation of a company

3. Index view references a JavaScript file: vm.index.js, that contains the code to call the web API and set up the view model:

   1: $(function () { 

   2:  $.getJSON("/API/company", function (data) {

   3:     var viewModel = 

   4:     {

   5:          //data

   6:          companies: ko.observableArray(ko.toProtectedObservableItemArray(data)),

   7:          filterText: ko.observable(""),

   8:     }; 


  10:     viewModel.companyCount = ko.computed(function () {

  11:         if (this.companies() == null) {

  12:             return 0;

  13:         }

  14:         return this.companies().length;

  15:     }, viewModel); 


  17:     viewModel.filteredCompanies = ko.computed(function () {

  18:         var filter = this.filterText().toLowerCase();

  19:          if (!filter) {

  20:              return this.companies();

  21:          } else {

  22:              return ko.utils.arrayFilter(this.companies(), function (company) {

  23:                 return >= 0;

  24:              };

  25:          }

  26:      }, viewModel); 

  27:      ko.applyBindings(viewModel); 

  28:  }); 

  29: });

4. Set up Index View to use the Knockout databinding for both the company list:

   1: <table class="table table-striped">

   2:     <thead>

   3:         <tr>

   4:            <th>Name</th>

   5:             <th>Date Created</th>

   6:             <th>No. of Teams</th>

   7:             <th>Principal Contact</th>

   8:         </tr>

   9:     </thead>

  10:     <tbody data-bind="foreach: filteredCompanies">

  11:         <tr>

  12:             <td data-bind="text: name"></td>                                

  13:             <td data-bind="date: dateCreated, dateFormat: 'DD/MM/YY'"></td>                    

  14:             <td data-bind="text: teamCount"></td>                    

  15:             <td data-bind="text: principalContact"></td>                    

  16:         </tr>

  17:     </tbody>

  18: </table>

5. And the filter text:

   1: <input type="text" data-bind="value: filterText, valueUpdate: 'afterkeydown'" placeholder="filter by name" />


Nothing more is needed – the JavaScript grabs the initial data and populates the companies array which in turn populates the filteredCompanies array. The filter text then instigates the refresh of the filteredCompanies array.

This is a much better user experience, not to say more efficient. Granted, we must be careful about how much data we load on to the client side, but this then gives even more weight to using DTOs to minimise the overload on data that is not required.

An area I am not so certain of though is the best approach for getting the data to the client in the first place – here we are using an Ajax call to a web API but firstly, that involves a second trip to the server on top of the initial page request and what about actions where parameters are passed in? Take the details view for instance – it might look something like: /Home/Index/3. The id of 3 would be picked up by the method in the controller, i.e. Index(int id) but how do we get hold of this id were we to make a call through to the corresponding web API method (API/companies/id)?

We could scrape it out of the url I suppose but that seems a little nasty. So perhaps it would be better to allow controller methods to do the necessary work and return the model in the first call and then convert the model into JSON to be used within the client-side JSON:

   1: (function () {

   2:    var viewModel = ko.mapping.fromJS(@Html.Raw(Json.Encode(Model)));

   3:    ko.applyBindings(viewModel);

   4: })();

This allows us to have our client-side model to manipulate, minimises the trips to the server for the initial page load and handles the issue of controller action parameters.

The one downside I have found is that where previously I liked to house my view model code in a separate script file, this doesn’t work well with the above conversion code as it cannot process the server-side directive ‘@Html.Raw(….’ outside of the cshtml file itself.

I would love to hear from anyone else who has a preference on how best to approach this.