The second reason to celebrate: we finally popped the cherry and sold our first enterprise copy of Yonkly (http://about.yonkly.com).   Now, we just have to do it again a few thousand times.

Yonkly is now in private beta, but you can signup for an invitation at http://about.yonkly.com/.  It will be coming out shortly and it will be kicking some ass.  If you can’t wait, you can always buy the latest version with source code for $399, go to http://about.yonkly.com/ enter your email and you will receive instructions.

You can follow Yonkly’s progress on our new blog at http://blog.yonkly.com (it’s a work in progress).  You can also follow us on twitter at http://twitter.com/yonkly 

Back to work, we have got a lot to do and not enough time.

Also, don’t for get to check Scott’s blog is http://scottdig.com/

One approach to my marketing strategy is to use an affiliate program.  We have all seen affiliate programs for everything from selling t-shirts to selling domains.  The question I struggled with is "how much do I pay my affiliates?"  I want it to be fair, competitive and also motivational enough to encourage the affiliates to market it.

I found out that 37 Signals pays 5% to their affiliates that sounds way too low.  What would be a good percentage?  15%? 25%?  The plan I came up with is a tiered approach which gradually increases your percentage from 10% to 50% - you can read more about it at http://www.yonklyapp.com/affiliates.aspx and maybe even signup to be an affiliate.  The best part is that it is recurring. 

So if you bring in a 100 customers, you will earn:

$200/month on the first 20
$400/month on the next 20
$600/month on the next 20
$800/month on the next 20
$1000/month on the last 20

That’s a total of $3000 EVERY MONTH.  Isn’t that pretty damn good?  Or not?  I would love to hear your opinions.

I created a themes folder under the content folder

Then I referenced my css file in the master page using a helper method

<%= ThemeHelper.GetCss() %>

I also use a helper method for images

<img src=’<%=ThemeHelper.GetImageUrl("reply.png")%>’ alt="reply" class="icon" />

But most of my images are set in the stylesheet, which makes it easier to manipulate different skins

The helper methods above look at the defined theme in the config file (or database or wherever you store your settings) and then return the path to the correct resource.

Take a look at these live samples:

• www.yonkly.com
• www.isweat.com
• www.hokietalk.com

They are all using the same codebase but have a different theme defined.  The trick is in making your HTML css-friendly by naming elements and assigning them classes as well as using Divs and avoiding tables.  This allows you to create a stylesheet that radically changes the look of the site.  Think of the element ids and classes as an API to your view, that the css can manipulate.

I also added a feature that lets you upload a folder theme as a zip file and have the application unzip it into the themes folder.

It would be cool if we can define a "virtual folder" in our application, so I wouldn’t have to use helper methods.  Imagine if you could just say /content/theme/logo.gif and it would just work.  The theme folder doesn’t really exist but instead it would route to the correct folder based on a setting.  I wonder if I can do that with current routing mechanism in asp.net mvc!!!  Anyone?

Widgets

If you are logged in, you can add a widget to your sidebar even place Google ads if you wanted.  Take a look at the widget in my profile at http://yonkly.com/eibrahim.  To add your own widget, go to settings and click the widgets link then simply add your widget code and hit save

If you want your widget to fit nicely with the rest of the sidebar, click the link "insert default widget" and add your script in the correct place.

Control Panel

If you are the owner of the site i.e. an admin, you will have access to a control panel where you can customize a ton of things

Site Settings

You can modify all sort of site settings such as title, description, keywords, amazon s3 information and so on.

Email Templates

You can customize all the emails that get sent out e.g. "someone is following", "someone joined your group", "thanks for signing up", etc.

Ad Management

Now, you can add advertisements to your site by adding the ads code in the appropriate location.  You can add it as a banner in the sidebar or in the message stream.

Here is a sidebar ad, you can also skin to match the site

Here is a banner ad

And here is an embedded ad in the message stream

Theme Management

You can change the theme of the site.  For examples, check www.hokietalk.com and www.isweat.com.  Here is the theme selection screen

Content Management

See the TOS, privacy and contact us links in the footer, those are dynamically created and you can create other pages as well e.g. "FAQ", "Help" and so on…  You simply enter a page name and the content using a rich text editor

User Management

This is just a way to assign/revoke admin privileges to users.

The Future

I am very excited about all the new features specially that the site owner and the user can display ads and make money.   I can’t wait to release the new version.  Contact me if you are interested in setting up your own microblogging platform.

Stay Tuned!!!

Take a look below and let me know what is wrong, what is right and what can be done better.

I drew the above diagram based on my understanding of TDD and by referencing other diagrams that I found online.  Specifically:

from http://www.agiledata.org/essays/tdd.html

from http://designindrive.com/?p=31

from http://tinyurl.com/498pjg

The asp.net mvc codeplex preview 5 version just came out a few days ago and it includes a lot nice enhancements.  For the best introduction read Scott’s post. 

Some of my favorite changes happened in validation.  Incidentally, I was already working on a post to talk about doing client and server side validation the easy way.  The inspiration came after reading Steve Sanderson’s post and watching his screencast on how he created a model-based client-side validation - very cool.

I want a simple way to perform client and server side validation for the application I am creating for my book and my objectives are simple (yet there is no easy way to do it):

1. Have a model-based validation mechanism
2. Have the least amount of repetition (DRY)
3. Client side validation using jQuery
4. Server side validation

First, I will show you how it works then I will explain how I got there.  Here is my model class for a user

   1: public class User

   2: {

   3:     public virtual string UserName { get; set; }

   4:     public virtual string Email { get; set; }

   5:     public virtual string Password { get; set; }

   6:     public virtual Guid ProviderId { get; set; }

   7: }

To add validation to it, I simply decorate it with the appropriate validation attributes.  It ends up looking like this:

public class User
{
    [ValidateNonEmpty("Username is required")]
    public virtual string UserName { get; set; }

    [ValidateNonEmpty("Email is required")]
    [ValidateEmail("Invalid email address")]
    public virtual string Email { get; set; }

    [ValidateNonEmpty("Password is required")]
    [ValidateLength(4,50,"Password should be at least 4 characters long")]
    public virtual string Password { get; set; }

    public virtual Guid ProviderId { get; set; }
}

I then generate the client script by adding 1 line of code to my view

<%=Html.ClientSideValidation("formCreateList", ViewData.Model) %>

On the server side, I simply make a call to a validation method from my controller

[Authorize]
[AcceptVerbs("POST"), ActionName("Create")]
public ActionResult Save(User newUser)
{
    AppHelper.ValidateModel(ViewData.ModelState, newUser);
    //the rest of the action code
}

If there are any validation errors on the server, I display the errors by using the framework helper method ValidationSummary.  All I have to do is add this 1 line to the view

<%=Html.ValidationSummary() %>

JavaScript

That’s all there is to it.  Now let’s talk about how it works.  The first challenge was generating the JavaScript to validate on the client. 

I decided to use the excellent jQuery Validation plugin.  The plugin gives me 2 ways to validate my form.  The first method uses classes on the the fields to be validated, for example a required email field would look like this

<input id="email" name="email" size="25"  class="required email" />

The second method uses rules that are defined in JavaScript and look like this

<script type="text/javascript">
$("#signupForm").validate({
        rules: {
            email: {
                required: true,
                email: true
            }
        },
        messages: {
            email: "Please enter a valid email address"
        }
    });
</script>

Check out the plugin documenation and demos to understand how it all works.

Since I was generating the script on the server and I wanted to make only one call, I went with the second option.

The script generator simply loops through all the model properties and their attributes and generate the appropriate script.  Afterthought: Now that I think about it, I could have easily generated a script to add classes to the elements using jQuery’s addClass method…  Oh well, maybe someone else will be kind enough to do it.

Castle Validator

I used the Castle Validator Component as my validation framework.  That’s where all the validation attributes you see above come from, but you can also create your own validators.  I then created a class called JQueryValidationProvider that implements IBrowserValidationGenerator interface in the Castle.Components.Validator namespace.  I haven’t implemented the entire interface and only have a couple of validations working (I will update the source code when it is all done).  Here is a couple of methods that I implemented.

public void SetEmail(string target, string violationMessage)
{
     Rules.Add(new ValidationRule(Validate.Email, violationMessage));
}
public void SetAsRequired(string target, string violationMessage)
{
    Rules.Add(new ValidationRule(Validate.Required, violationMessage));
}

As you can see, all I am really doing is just adding the validation rule to List of ValidationRules.  ValidationRule is a straight forward class and looks like this:

internal class ValidationRule
{
    public ValidationRule(Validate validate) : this(validate, null) { }
    public ValidationRule(Validate validate, string failureMessage)
        : this(validate, failureMessage, null) { }
    public ValidationRule(Validate validate, string failureMessage, object value)
    {
        Validate = validate;
        Value = value;
        if (failureMessage != null)
        {
            ErrorMessage = failureMessage;
        }
        else
        {
            //setup default error messages
            switch (Validate)
            {
                case Validate.Required:
                    ErrorMessage = "This field is required";
                    break;
                default:
                    ErrorMessage = "Field error";
                    break;
            }
        }

    }
    public readonly object Value;
    public readonly Validate Validate;
    public readonly string ErrorMessage;
}

The helper ClientSideValidation method simply loops through the collection of rules and generate the appropriate JavaScript.  Here it is

public static string ClientSideValidation(this HtmlHelper htmlHelper,
    string formName,
    object modelToValidate)
{
    var results = new StringBuilder();
    var rules = new StringBuilder();
    var messages = new StringBuilder();
    results.AppendFormat("$(\"#{0}\").validate({1}", formName, "{");
    results.AppendLine();
    rules.AppendFormat("rules: {0}", "{");
    messages.AppendFormat("messages: {0}", "{");

    var props = TypeDescriptor.GetProperties(modelToValidate.GetType());

    var propCounter = 0;
    foreach (PropertyDescriptor prop in props)
    {
        var generator = new JQueryValidationProvider();
        foreach (var attrib in prop.Attributes.OfType<AbstractValidationAttribute>())
        {
            var v = attrib.Build();
            v.ErrorMessage = v.ErrorMessage ?? "*";
            v.Initialize(new CachedValidationRegistry(), null);
            v.ApplyBrowserValidation(null, InputElementType.Undefined,
                                     generator, null, null);
        }

        if (generator.Rules.Count > 0)
        {
            if (propCounter > 0)
            {
                rules.Append(",");
                messages.Append(",");
            }
            rules.AppendLine();
            messages.AppendLine();
            rules.AppendFormat("{0}: {1}", prop.Name.ToLower(), "{");
            messages.AppendFormat("{0}: {1}", prop.Name.ToLower(), "{");
            rules.AppendLine();
            messages.AppendLine();

            for (var i = 0; i < generator.Rules.Count; i++)
            {
                var rule = generator.Rules[i];
                rules.AppendFormat("{0}", rule.GetRuleString());
                messages.AppendFormat("{0}", rule.GetMessageString());

                if (i < generator.Rules.Count - 1)
                {
                    rules.Append(",");
                    messages.Append(",");
                }
                else
                {
                    rules.Append("}");
                    messages.Append("}");
                }
                rules.AppendLine();
                messages.AppendLine();
            }

            propCounter++;
        }
    }

    rules.Append("},");
    messages.Append("}");
    rules.AppendLine();
    messages.AppendLine();
    results.Append(rules.ToString());
    results.Append(messages.ToString());
    results.Append("});");

    return string.Format("<script type='text/javascript'><!--{1}{0}{1}--></script>",
                            results.ToString(), Environment.NewLine);
}

Changes

Halfway through this post, I read Steve Sanderson’s new post and Stephen Walther’s new post on validation, I made some changes to the code.  I agree with their point that the controller shouldn’t be responsible for initiating the validation and it should be done at a lower layer.  I refactored the code and my controller now looks like this

[AcceptVerbs("POST")]
public ActionResult Create(User user)
{
    try
    {
        user = new User();
        UpdateModel(user, new[] { "name", "email" });
        var service = new UserService();
        service.Save(user);
    }
    catch (RuleViolationException ex)
    {
        MvcValidator.PopulateModelStateWithErrors(user, ViewData.ModelState, ex);
    }
    return View(user);
}

The validation gets initiated at the service (business logic) layer and if there is a rule violation an exception is thrown.  I then take the exception and pass it to a helper method that populates the model state. Note: I could probably rewrite this so that the ModelState population happens before the exception throwing which will eliminate the call to the population method.

My view looks like this

<%=Html.ValidationSummary() %>
<form id="formCreate" name="formCreate"
        action='<%=Url.Action("Create")%>' method="post">
    <label for="name">Name</label>
    <input type="text" id="name" name="name" />
    <br />
    <label for="email">Email</label>
    <input type="text" id="email" name="email" />
    <br />
    <input type="submit" value="Submit" />
</form>
<%= Html.ClientSideValidation("formCreate", ViewData.Model) %>

Now if I try to submit the form, the client script validation runs and I get these errors

If you get rid of client validation and only use server validation

to submit the form, the form would look like this

Finally, here is the client JavaScript generated in the view:

<script type='text/javascript'><!--
    $("#formCreate").validate({
        rules: {
            name: {
                required: true,
                minlength: "3",
                maxlength: "25"
            },
            email: {
                email: true,
                required: true
            }
        },
        messages: {
            name: {
                required: "Username is required",
                minlength: "Username should be between 3 and 25 characters long",
                maxlength: "Username should be between 3 and 25 characters long"
            },
            email: {
                email: "Invalid email address",
                required: "Email is required"
            }
        }
    });
--></script> 

Pros & Cons

Here is what I think the pros of this design are:

1. I love using model attributes to define validation rules
2. I like that I only need 2 lines in my view
3. I like that I don’t have to do anything in the controller other than catch an exception and call a helper method

The one thing I don’t like about this design is that it feels like there is a lot of dependencies.  I might need to refactor the code and move things around.  Here is my dependency diagram

Credit

This code was possible because of other people’s work:

Stephen Walther - link

Steve Sanderson - link and link 2

Hamilton Verissimo - link

Rick Strahl - link

Scott Guthrie - link

Source Code

Click here to download the rough-quality source code.

Let me hear what you think of this design.  What’s wrong with it?  What’s right?  Should I put this up on codeplex?

Things to Note

I have tried to use the general consensus of the community for each decision.  For example, some might argue that MVC is better for prototyping but I would say the majority would disagree.

Keep in mind that these choices are not set in stone and are not mutually exclusive but are more like a guide to which platform would be better suited to satisfy a particular requirement.  For example you might want to use TDD but still end up choosing on WebForms.  (I should probably assign a weight to each question)

The "No" branch doesn’t mean exactly NO.  I mean, answering No to "Want cleaner HTML?" doesn’t mean you have something against clean HTML or you want crappy HTML - no one wants that.  Think of the question in terms of "is clean html important to you?" 

Let me hear your feedback below and I will incorporate it in version of 2 of the diagram and post it back on the blog.  Hopefully, this will help someone choose between ASP.NET MVC and WebForms on a new project.

Also, I don’t think a flowchart diagram is the best type of diagram for this type of decision tree.  Can you suggest a better format?

ASP.NET MVC Preview 5 introduce the ModelBinder attribute that can be used to decorate a complex type in an Action.  This allows us to have actions that look like this

public ActionResult Create([ModelBinder(typeof(GenericBinder))] ContactList myList)

Instead of this:

public ActionResult Create(string name, string description)

The problem is that you have to build a Binder for every complex type you want to use as a parameter.  For example, Maarten Balliauw created a model binder on his blog and it looks like this:

public class ContactBinder : IModelBinder
{
    #region IModelBinder Members
    public object GetValue(ControllerContext controllerContext,
                 string modelName, Type modelType,
                 ModelStateDictionary modelState)
    {
        if (modelType == typeof(Contact))
        {
            return new Contact
            {
                Name = controllerContext.HttpContext.Request.Form["name"] ?? "",
                Email = controllerContext.HttpContext.Request.Form["email"] ?? "",
                Message = controllerContext.HttpContext.Request.Form["message"] ?? ""
            };
        }
        return null;
    }
    #endregion
}

Now that is a lot of typing and because I am lazy, I decided to create a generic binder that uses reflection and can work with all my complex types. 

Note: By generic I mean common - it has nothing to do with .net Generics

Also note that this will only work if you follow these conventions:

1. The html field name must match the property name
2. User lower case names for the html fields
3. You don’t have to user lower case on your model properties

Here is the very rough and untested Generic Binder:

class GenericBinder : IModelBinder
{
    public object GetValue(ControllerContext controllerContext,
                            string modelName, Type modelType,
                            ModelStateDictionary modelState)
    {
        var instance = Activator.CreateInstance(modelType);
        foreach (var prop in modelType.GetProperties())
        {
            prop.SetValue(instance,
                    controllerContext.HttpContext.Request
                                    .Form[prop.Name.ToLower()],
                    null);
        }
        return instance;
    }
}

If you find any bugs or have a better implementation, please share.

I ran these tests http://mootools.net/slickspeed/ on IE, Firefox and Chrome.

Here are the results:

It looks like Chrome is about 13% faster than Firefox and 82% faster than IE 7.

All browsers are running on the same computer, so the hardware/os settings are irrelevant.  But if you are curious, it’s a Windows Vista 64-bit machine with 6 Gigs of RAM and a quad core chip.

[Update] If you are reading this in Chrome, the embedded document is broken.  I guess Google didn’t test there own sites.  It should look like this:

Instead it looks like this:

public IAccountService Service { get; set; }

[Inject]
public AccountController(IAccountService service)
{
    Service = service;
}

I could just have a property Service do the work like this:

public IAccountService Service
{
    get
    {
        return (IAccountService) Kernel.Get(typeof (IAccountService));
    }
}

If I get rid of all my constructor parameters then I won’t need to change my ControllerFactory to use the Ninject Factory.  This keeps looking better and better.

What do you think?  I keep reading everywhere that construction injection is the way to go.  But why complicate my constructor when I can have the properties do the heavy lifting?  I would love to hear what you think.

The CurrentUser Property needs to look something like this:

public User CurrentUser
{
    get
    {
        var key = "currentuser";
        if (Session[key] == null)
        {
            Session[key] = /*get user from database some how*/;
        }
        return (User) Session[key] ;
    }
}

This looks simple enough but it is not.  The reason it is not simple is because to get the current user I have to call Membership.Provider.GetUser from the BaseController.  The problem with that is that it creates a dependency on the MembershipProvider class which I don’t want to have, because it will make testing very hard.

One obvious solution was to add the MembershipProvider (which is an abstract class) to the BaseController’s construct and then pass a mocked instance during testing…  The problem with this design is that now my BaseController will be forced to have a parameterized constructor which means that I have to change the code in all the derived controllers to handle that and pass the appropriate instance of MembershipProvider.  Sounds like a code smell.

My solution was to create the MembershipProvider class using my IoC container - in this case, my Ninject Kernel.  This allows me to inject a SqlMembershipProvider in development and runtime and inject a mocked provider in testing.  So the final CurrentUser property looks like this:

public User CurrentUser
{
    get
    {
        var key = "currentuser";
        if (Session[key] == null)
        {
            var Provider
                = (MembershipProvider)
                    Kernel.Get(typeof(MembershipProvider));

            Session[key] = AppHelper.CreateUserFromMembershipUser
                            (Provider.GetUser(User.Identity.Name, true));
        }
        return (User) Session[key] ;
    }
}

If you have been paying attention, you are probably wondering  what is this “Kernel” thing.  Well Kernel is an instance of the Ninject Kernel which itself was injected into the BaseController class like this:

[Inject]
public IKernel Kernel { get; set; }

I could have done this differently.  For example, I could have injected the provider itself using property injection like this:

[Inject]
public MembershipProvider Provider { get; set; }

The only problem is that the provider isn’t needed by all the derived classed and I didn’t want to have a public property in the base class that would hardly be used anywhere else.   On the other hand Kernel could be globally used to instantiate an object.

What do you think?  Is this the way to do it?  Is there a better way?

For example, my AccountController takes IEmailService, IFormsAuthentication and MembershipProvider (abstract class) as parameters.

During my testing, I want to mock the email, authentication and membership calls.  For example when the user calls FormsAuthentication.Login, I don’t really care if actual call succeeded but rather that my login action works appropriately in the case FormstAuthentication.Login succeeds (or fails).  I just want to mock that call.

I started off creating a few tests and slowly they have grown to several.  There was a lot of repeated code in my unit tests and to be a good citizen of the DRY universe, I needed to refactor the code.

For IoC, I initially started with StructureMap but now I am using Ninject. 

I created this module to bind my interfaces to mocked instances.  It looks like this:

internal class TestModule : StandardModule
{
    public override void Load()
    {
        Bind<IEmailService>()
            .ToConstant(MyMocks.MockEmailService.Object);

        Bind<IFormsAuthentication>()
            .ToConstant(MyMocks.MockFormsAuthentication.Object);

        Bind<MembershipProvider>()
            .ToConstant(MyMocks.MockMembershipProvider.Object);

        Bind<IContactListService>()
            .ToConstant(MyMocks.MockContactListService.Object);
    }
}

Notice that I bind the interfaces to actual instances and not classes.  These instances are declared in a global static class that will be accessed from my unit tests.  As you can tell from the name, they are all mocked objects (I am using Moq).  Here is how the MockEmailService looks (all the others are declared the same way):

internal static class MyMocks
{
    private static Mock<IEmailService> _mockEmailService;
    public static Mock<IEmailService> MockEmailService
    {
        get
        {
            _mockEmailService = _mockEmailService ?? new Mock<IEmailService>();
            return _mockEmailService;
        }
    }

So all this is good to setup Ninject and create my mocks.  Now I want to easily and generically create a controller, so I can quickly create unit tests.  In order to do that, I created a TestControllerFactory class that basically creates a controller with all the appropriate dependencies injected.

   1: internal static class TestControllerFactory

   2: {

   3:     private static IKernel _kernel;

   4:     public static IKernel Kernel

   5:     {

   6:         get

   7:         {

   8:             if (_kernel == null)

   9:             {

  10:                 var modules = new IModule[] { new TestModule() };

  11:                 _kernel = new StandardKernel(modules);

  12:             }

  13:             return _kernel;

  14:         }

  15:         private set

  16:         {

  17:             _kernel = value;

  18:         }

  19:     }

  20:  

  21:     public static T GetControllerWithFakeContext<T>(string httpMethod) 

  22:         where T : Controller

  23:     {

  24:         var con = Kernel.Get<T>();

  25:         con.SetFakeControllerContext();

  26:         if (con != null) con.Request.SetHttpMethodResult(httpMethod);

  27:         return con;

  28:     }

  29:  

  30: }

In line #10, I use the TestModule class mentioned above to setup the Ninject Kernel.  In lines #21 to #28, I create an instance of T which must be of type Controller from the Kernel which will automatically create the Controller with all the mocked objects.  In line #25 and #26, I just set a fake/mocked context and the Http Method for the request (more info here).

Now my unit tests are very clean and easy to setup.    Using MbUnit as my unit test framework, here is a unit tests that tests the reset password functionality.

   1: [Test]

   2: public void ResetPasswordQuestion_Should_Send_Email_On_Success()

   3: {

   4:     var newpassword = "newpassword";

   5:     MyMocks.MockMembershipProvider

   6:          .Expect(p => p.ResetPassword(username, answer))

   7:          .Returns(newpassword);

   8:     MyMocks.MockEmailService

   9:          .Expect(m => m.SendPasswordReset(username, newpassword));

  10:  

  11:     var ac = TestControllerFactory