If you look up the definition of MVC or Model View Controller, it will hail the definition as being able to change your front end without affecting other parts of the application and vice versa. This sounds great in theory, however this claim is nothing more than a blatant lie.

A failing of the architectural pattern comes from the whole codebase being tied to a specific stack. Take ASP.NET MVC. The domain logic will most likely be in C#. Therefore your models will be in C#. Your controllers will be in C#. Your views will be a mixture of C# and some form of a templating language.

If you want to change your stack to the “next big thing” you are forced to take a big bang approach. ASP.NET MVC won’t be around forever. Being tied to a specific technology feels wrong. Therefore this coupling means your designers are forced to use the templating language that your framework supports. This should be a flexible option that should be easy to change, after all the MVC pattern states this as one of it’s benefits.

Being tied to a specific technology leads onto our most recent project. One of our biggest and most important projects is a legacy Flash application. Back in the early 2000′s it was a cutting edge application – consistent across all browsers, ajax style requests, responsive design, you name it.

That being said we all know Flash is on its way out, and there lies the problem. It took myself about two weeks to add a few text boxes to the app in my first year at Codeweavers, all because the UI code is so difficult to work with. The logic is mixed within the UI. Had the app been developed in a MVC style we would be in a position to replace the legacy UI with a modern alternative.

We make use of SOA or Service Oriented Architecture at Codeweavers, therefore it seemed a natural fit to apply this to our rewrite of our legacy application. I proposed a theory:

“for an application to be truly independent of the frontend and backend the code must be developed in different languages.”

For example, I taught myself enough PHP to make a JSON request, perform some conditional logic and loop over a collection. With this I was able to recreate one of our applications that was powered by our backend C# services. I would not want to create an application in PHP, but using PHP as a templating language was a great fit. Afterall this is one of the intentions of the language. Limiting myself to just three simple PHP constructs I was forced to put all logic on the service in question.

This complete separation of concerns is made possible due to the fact it is simply not possible for code to leak between the layers due to the different languages used in the implementation. This means I could easily spin up numerous front end views while the backend remains unchanged. Likewise we could change the back end implementation from C# to another language. Providing the endpoints and request/responses match, the front end will still be functional. This full separation of concerns is what MVC style frameworks have failed to achieve.

In ten years from now it is hard to say what the web will look like. What I can guarantee is that the web will still be here. We’ll still be making HTTP requests. We’ll still be making back end services that powers much of the apps on the internet. One thing no one can really comment on is what the web will look like. One point we all could agree on is that HTML5 should be wide spread and no doubt “the next big thing” will be on the horizon. The great thing by taking the approach discussed previously is that Codeweavers will be in the position to change either the front end or back end of our codebase at any time. Precisely what the MVC pattern has failed to deliver.

Image by lynnepet (Creative Commons)

Here is our updated reading list for 2012. We might have missed something so if you think we should be reading it, let us know.

Process

• The Toyota Way – Jeffrey Liker
• The Goal – Eli Goldratt
• XP Explained - Kent Beck
• The Pragmatic Programmer- Andrew Hunt and David Thomas
• The Clean Coder – Robert C. Martin

Code

• Clean Code – Robert C. Martin
• Test-Driven Development by Kent Beck
• Refactoring – Martin Fowler
• Refactoring Workbook – William Wake
• Patterns of Enterprise Application Architecture – Martin Fowler
• Head First Design Patterns – Freeman, et al
• Growing Object-Oriented Software, Guided by Tests (Beck Signature) – Steve Freeman and Nat Pryce
• 97 Things Every Programmer Should Know- Kevlin Henney
• Pro ASP.NET MVC x+ – Steve Sanderson
• Framework Design Guidelines – Brad Abrams and Krzysztof Cwalina

Last week we had @kevinrutherford in to run a coderetreat. It was the first retreat I’ve taken part in personally, the same applying to much of the team. The day was to focus on one challenge – Conway’s Game of Life, though each iteration would introduce new constraints.

Iteration 1

The four rules of simple code were discussed, aka “Extreme Normal Form”.

1. Passes tests
2. Communicates intent
3. No duplication
4. Nothing unncessary

Our first run through was somewhat of a disaster. We became too obsessed with how the grid was to be stored. This meant we spent a long time messing around with arrays and multidemionsal arrays in C#. Both are somewhat tedious and not something we use often, therefore we concluded a more “grid agnostic” approach would be needed. After all, there is nothing stating the grid need be a square, fixed layout.

Iteration 2

The concept of Arrange, Act, Assert or AAA was discussed.

• Arrange – set up pre conditions
• Act – do something on the subject under test
• Assert – verify or check some result

The constraint during this iteration was that all asserts were to be created first, and we would work backwards. Personally I found this difficult, both mentally and in terms of tooling. For the past four years I’ve rarely wrote an assert first, therefore this switch proved tricky.

Visual Studio and Resharper also tend to favour a top down approach. In other words, it’s easier to stub out a class and add methods, rather than the inverse.

New members to the team found the act of writing the assert first much easier however, and they’re looking to continue this in day to day development.

The point I took from here is that while I may personally know what I’m writing, my partner might not be aware of where we are heading. Starting with the assert first allows both developers to see the goal for that particular test.

Iteration 3

Our third iteration introduced the unrealistic concept of mute pairing. Only the code and tests can reveal our intent.

Using ping ponging (switching between developers who writes tests/production code) we managed to get a good chunk of functionality complete. Despite this the rest of the iteration was very difficult. Being unable to talk about design concepts was incredibly frustrating.

The most important concept we took away from here was that more than likely the developer working with your code tomorrow will not have worked with you previously. That means any conversations you’ve had to explain the code will be lost. The only thing left to communicate will be your tests or production code. Due to this observation, it is cruicly we name our tests and code in a manner that is cleary and understandable, whether six minutes or six months have passed.

Iteration 4

Our fourth iteration had a focus on OO concepts. Up until now, our code had exposed much of the state in various ways. The challenge was to focus on the messages that are exchanged between objects, rather the change in state.

This introduced a new problem. How do we test an object without exposing state? We talked about a couple of methods; mock objects or the self shunt pattern.

Using the shelf shunt pattern was an enjoyable experience – we identified numerous violations of the interface segregation principle as it was actually painful to add to the interface in question. With mocking frameworks this pain is often lessened due to the dynamic and often “magic” constructs they employ.

Personally I found code writting using this pattern easier to understand. The plan will be to adopt this approach for a month or so and see how it effects day to day development.

Our code looked pretty good at this point, though we were struggling to replace a conditional statement. After a few minutes the idea seemed obvious once we recieved a hint – polymorphism. This sparked a nice discussion about the “Anti If” campaign – another point I’m looking to take further. This iteration proved the most challenging for the whole team. A second run through allowed further progress.

Iteration 5

The final round had us regroup with our original partners. Here we could use any approach and see how far we could progress. Me and my partner opted for a failing acceptance test, from here we would implement the functionality top down until we had the components wired together. This worked well until we struggled to test how the cells were stored without exposing any state needlessly. The solution came from some discussion around object calesethenics – first class collections.

One thing to note was how drastically different our code looked, when compared to the first iteration. During the course of the day we all noticed a difference in our overall approaches.

Conclusion

Overall I believe I speak for the whole team when I say how much we enjoyed the day. It was refreshing to spend a whole day coding, rather than being interuppted with other issues. Naturally this was quite tiring – though left lots energy for discussion afterwards. As a conclusion it would appear that as a collective we need to spend more time each week carrying out deliberate practice. For new starters this is essential training, while for more seasoned developers this is a useful way to keep our skillset up to date.

At the time of writing the latest version of Firefox (version 13) has just been released. Bare in mind that a week ago I updated our Selenium bindings so that we could use Firefox 9+ for running our browser tests.

The latest release is another great release for the Firefox team, except there is software out there will be broken. The software in question I’m talking about is any code that uses Selenium 2.22.0 that was released 2012-05-29. It turns out the bindings only work for Firefox 12 or less.

For whatever reason any tests that used Selenium this morning just stopped working for us – and others. The tests in question caused the runner to hang as no window could be opened. I’m not sure what causes this, as the browser is essentially the same to the end user, bar some new features. Not being a Selenium developer I cannot comment how or why this has happened, nor can I suggest the Selenium team should be version agnositc.

Our solution in the end was simple. Turn off the auto updating and downgrade the browser. I’ve blogged about this in the past, but since Firefox 10 – the team are adopting a “silent” update process. This is great for end users. Imagine the countless man hours saved if IE6 had shipped with an auto update feature? The problem now seems to be in the hands of developers.

Another attempt to make this problem more obvious has been to add a check prior to our tests running to ensure that it can open a window. If this fails or hangs, we display a useful error message indicating that the browser in question is not compatible. This is due to the fact that it is not immediatley obvious what the problem is. More confusion occurs when some machines will execute the tests with no problems at all.

Tools -> Options -> Advanced -> Update Tab

So if you use Selenium and Firefox – ditch the auto updating. Manually update your bindings and check compatability for now…

Good unit tests should be:

• fast
• independent
• well focused
• isolated

If your unit tests are slow, you’re not gonna run them as often as you should. Therefore one of the main benefits of unit testing is lost – the lack of instant feedback.

Each of your unit tests should be independent. The order in which you run your tests should not matter. By keeping your tests focused you should be able to refactor, add new code and not have the majority of your tests fail. If you change class A, you would expect class A’s tests to fail at worst. If other tests outside of this scope fail, your tests are not focused enough. This lack of focus leads on to isolation.

Tests should be isolated from other dependencies. Dependencies such as other classes should not affect each other, providing the contract between the code is maintained. Likewise the file system, the web and databases should not be involved anywhere with your unit tests. If any of these dependencies come into play, you’re not unit testing.

At Codeweavers we have around ten thousand tests, with unit tests accounting for the majority of these tests. Naturally this means every now and then we take time to do a bit of house keeping regarding our tests.

One thing we noticed was that some of our tests were taking longer to run than other tests. They were taking anywhere from one to ten seconds. Ten seconds for a unit test is a huge time. During this period we could have run hundreds of other tests! As for why these tests took so long to run? Easy. They were not unit tests. Code had been added that broke that layer of isolation. Some tests were hitting real web services for example.

In order to be fully isolated I proposed a simply solution. Unplug the network cable. Any tests that failed would not be unit tests. This gave us one of two options:

• Refactor the code – remove or stub the dependencies
• Promote the tests to integration/regression tests (only run prior to check in)

You can take this idea one step further. Next time you run your tests try running them from a different location. Any tests that fail are relying on relative/hardcoded paths and will need attention.

After performing this task on our codebase we had some failures. The nice thing about solving these failures is that our tests now run a lot faster. Our slowest tests are now end to end regression tests which are only run prior to check in or by our CI server. We’ve also made sure that from a disaster recover point of view, we can continue developing locally even if our CI server is not present.

So take the experiment. Unplug your computer from the network. How many of your “unit tests” fail?