I promised a bunch of people from the BBC that I would write about this, so here it is! This is my first Software Engineering-related post, bear with me :)

We recently released a new iteration of /programmes, built on top of a completely new technology stack. As part of that move, we decided we wanted to review our template testing strategy. In our old application, the process for a new feature would basically be:

  1. Software Engineer writes models, controllers and data views;
  2. Web Developer writes XHTML templates;
  3. Software Engineer writes controller tests, which would actually test the routes, the controllers and the templates (such controller 'unit tests' are actually fairly standard across MVC frameworks, for some reason - e.g. in Zend)

Those controller tests were based on CSS selectors or XPaths. Therefore, any time a small front-end tweak needed to be done, the controller tests would break, which is very annoying for everyone.

We had two problems:

  • Our controller 'unit tests' were not really unit tests - front-end developers shouldn't have to understand the whole routing, controllers, models for making a front end change.
  • Using CSS selectors or XPaths for template tests is brittle. We don't want our tests to break every time we change the name of a CSS class.

In order to solve the first problem, we divided our controller tests in route tests (here is a route, assert that the application forwards it to the right controller/action with the right parameters), real unit controller tests (mock the models, call an action with some request parameters, check that the right data is sent to the view), and template tests.

In order to solve our second problem, we made those template tests rely on RDFa markup embedded within the page. In order to test a template, we create some mock data, generate the template using this data, and check the right RDF triples can be extracted from the resulting page. It ensures that tests are actually based on data - front-end changes won't have an impact on them. We just want to make sure we present the right data to the user. As the tests are not relying on other application code, it also means that someone writing the templates can maintain his own test suite.

A simple example of one of these tests is the following one:

    public function testLetterSearch()
        $data = (object) array(
            "by"      => "by",
            "search"  => "b",
            "slice" => "player",
            "letters" => array('@', 'a', 'b', 'c'),

        $this->assertTriples($data, array(
            array('/programmes/a-z/by/b', 'rdfs:seeAlso', '/programmes/a-z/by/%40/player'),
            array('/programmes/a-z/by/b', 'rdfs:seeAlso', '/programmes/a-z/by/a/player'),
            array('/programmes/a-z/by/b', 'rdfs:seeAlso', '/programmes/a-z/by/c/player'),

which can be read as if you are displaying a list of letters on an A-Z page and you have selected one, you shouldn't link to that one

Another very nice side-effect is that developers have a motivation to put lots of RDFa inside our pages! Compare the 168 triples extracted from a new /programmes page, including full tracklist information, programme metadata and broadcasts, to the 18 triples extracted from an old one. And as we add new components to this page, more RDFa will become available.

Also, the speed at which our developers picked up RDFa (1.0, not even 1.1, which is apparently simpler) defeats the eternal argument about RDFa being too complicated, but that's just my opinion :) The RDFa cheat sheet has proved immensely helpful.