Why outside in Webdevelopment is Better; why Designs should be Leading.

Woodcut from Doré. Purely illustrative
Doré Woodcut. Its only function is to make the layout look better. And these images are really nice themselves

Designs should be leading when building a website or -application. There! I said it.

Sure, form follows function; content is king and so on: a design should be a Good Design to begin with: it should carry the function, content and user-experience to a higher level. So, for the sake of the argument, let us assume you have the Perfect Design For Your Project, Audience and Branding. It is UX-tested. It is delivered as valid, clean HTML and CSS, backed with PSDs, Wireframes and looks stunning. It has passed numerous PR- and marketing sessions, and even the boss wants it implemented like this.

Enter "the technical guys". People like me, a backend-guy. Sorry, but the CMS we are going to build it with, makes it really hard to implement this design., sometimes followed with the really pragmatic solution to Have the CMS dicate the way the application works. To redo the wireframes according to how the CMS works, because that will save lots of time and budget. That last word, being the most important one: Budget.

People think visual. People talk visual. And our delivery is mostly visual: people interact with our -technical- solutions through the "interface", that which was in "the Design".

But most important: the design, that interface, is most prone to change; from small changes in wording in a footer, to a grand rebranding-overhaul: mostly interface changes. When I calculate the tickets in the projects I am now involved in, 80% of the tickets are "interface"- or "interface-related". First because this interface is where any problems below surface and second because this is what 80% of your and my clients care about: the interface.

Yet, Even though most of the communications and deliveries are outside-in; we work inside out; starting with database-designs, CMS-configurations from which we work towards the interfaces.

But how?

You'll be developing the innards of the application (its models, controllers) and then, in the end, need a giant hammer to beat the views, glued to it, into shape to match as closely as possible to your designs.

It gets worse when these innards are fixed, because you use a CMS that has them built in, for example. You'll be writing extensions, hacks and overrides all trough that CMS, just to make it output its interfaces as close as possible to designs.

The solution lies in outside-in development. I encountered that term first in The Rspec Book — A Comprehensive Handbook for Behaviour Driven Webdevelopment. And I am loving it.

The idea is as simple as pragmatic: you develop by making the views, the interface first. Then fixate them in tests (specs), so that the interface, the HTML is fixed in your project. From there on, you develop the underlying application, whether those are modules for a CMS, models and controllers for an MVC-framework or configurations of a point-and-click-tool.

Once the HTML (and thus the CSS and interaction) is fixed, a test will start to fail the moment someone touches that HTML. Developers can therefore develop, refactor, and rewrite the underlying application as often as they wish, without breaking the Designs. Sounds cool, not?

Flexibility.

Well, maybe not if that underlying application lacks the flexibility to be rewritten, altered and changed without making these views-tests fail. If, the moment you install an extension, all your tests turn Red (they fail), your tests, the fixed interfaces, become a burden. It is then near impossible to keep them passing, without all the hacks, extensions and configurations. That is probably the moment that you decide to rather change the view-specs (i.e. Divert from the original Designs) just to get forward.

It means you have the wrong tool. It means you have a tool at hand where you'll simply never get 100% according to design. Where the tool dictates the designs and workflows of the application. If that works for you, then outside-in development makes little sense; but in that case, be very aware of this all trough the process; It really makes no sense to have a UX-lab test all the wireframes and mockups, if two weeks later, you find you cannot implement these wireframes and mockups anyway, due to the underlying inflexible tools.

Most RAD frameworks like Rails, Django, and various or their PHP-clones, are flexible enough in this, because they don't (ever) assume anything about your wireframes and designs. Some, very few actually, CMSes allow this too.

An example?

In Rails many people test with Rspec; Views are (arguably) best described in cucumber features. The exact syntax and setup goes way beyond this post, so let's look at some pseudo-code:

Scenario: I want to log in
  Given I am signed in
  When I click on my name
  Then I should see my profile-page

A simple feature, described in Gherkin and fixates the actual views with so-called step-definitions:

Given /I am signed in/ do
  @me ||= Factory(:valid_user)
  When %(I go to the new user session page)
  When %(I fill in "Username" with "#{@me.username}")
  When %(I fill in "Password" with "#{@me.password}")
  When %(I press "Sign in")
end
When /^I click on my name$/ do
  click_link("#{@me.first_name} #{@me.last_name}")
end
Then /^I should see my profile-page$/ do
  page.should have_xpath('//*/h1.name', :text => "#{@me.first_name} #{@me.last_name})
end

This is a really simple test to ensure that a logged-in user, who clicks on a link with her name, gets to see the profile-edit page, containing an H1-tag, with class "name" and contents of the users' name. Real world tests, with more detail over at Jared Carroll's blogpost on this.

Downsides?

Certainly. As you can see, just to describe a login-form, a link with my username, and a generic H1-tag, I already need one entire story. Depending on your test-suite and your application, you may have to write far, far more test-code then "actual" code.

But outside-in development is just one of many reasons why one might choose for test-driven development. Writing tests requires effort, time and dedication. Without tests you have no way to describe and fixate the interfaces. And without that, there is no outside-in development (And you'd best tell your designers and clients right-away that you are not going to get 100% pixel-perfect interfaces, within agreed budget and time... :) ).

The upside, is that your application is tested, and can easily remain so. With each release, a few commands ensure you that your (ever growing) application works. That all the work by all your colleagues or contributors remains working. And that the interfaces, the one thing your users care most about, work and look the way they expect them to look.

About the author: Bèr Kessels is an experienced webdeveloper with a great passion for technology and Open Source. A golden combination to implement that technology in a good and efficient way. Follow @berkes on Twitter. Or read more about Bèr.

blog comments powered by Disqus