Behaviour-Driven Development of Rails viewsEdit

Note: This article was originally written in early 2007; for a more up-to-date overview of Rails view testing, see Rails view testing cheatsheet.

Introduction

Like models, views are fairly easy to test independently. Despite the fact that they depend on their controllers to feed them data, well-designed views contain very little code and merely reference model attributes in simple ways. At most, the most complicated logic in a view should be limited to iterating over a collection, or conditionally including content.

How much testing is appropriate?

Take the following piece of HTML as an example:

<div id="userlist">
  <div class="odd">
    <div class="label">Login name:</div>
    <div class="info">user1</div>
    <div class="label">Display name:</div>
    <div class="info">user1</div>
    <div class="label">Email address:</div>
    <div class="info">user1@example.com</div>
  </div>
  <div class="even">
    <div class="label">Login name:</div>
    <div class="info">user2</div>
    <div class="label">Display name:</div>
    <div class="info">user2</div>
    <div class="label">Email address:</div>
    <div class="info">user2@example.com</div>
  </div>
</div>

The high-level specifications that led to this HTML are:

  • Users should be displayed in a list
  • The list should have alternating "odd" and "even" rows
  • Each row should show the login name, display name and email address for the user
  • The login name, display name and email address should each have corresponding labels
  • Each element should have an appropriate class or ID so that the output can be styled using CSS

The Rails view that generates this view is:

<% for user in @users %>
  <div id="userlist">
    <div class="<%= cycle('odd', 'even') %>">
      <div class="label">Login name:</div>
      <div class="info"><%=h user.login_name %></div>
      <div class="label">Display name:</div>
      <div class="info"><%=h user.display_name %></div>
      <div class="label">Email address:</div>
      <div class="info"><%=h user.email_address %></div>
    </div>
  </div>
<% end %>

The question is, how many specs should we employ while going from our high-level goals to the final implementation? If we take a pure Behaviour-Driven approach then we won’t add anything to the view at all without first writing a failing spec that requires that something be added to the view in order to pass.

For example, we might start with a spec that states:

  • A user list with 2 users
    • Should display an odd and even row

And begin with this view code:

<% for user in @users %>
  <div class="<%= cycle('odd', 'even') %>"></div>
<% end %>

And then successively add specs and additions to the view until we have the final result. A final spec might look like this:

  it 'should display the login name, displays name and email address in each row' do
    response.should have_tag('div.odd>div', 6) do
      with_tag('div.label', 'Login name:')
      with_tag('div.info', 'user1')
      with_tag('div.label', 'Display name:')
      with_tag('div.info', 'user1')
      with_tag('div.label', 'Email address:')
      with_tag('div.info', 'user1@example.com')
    end
    response.should have_tag('div.even>div', 6) do
      with_tag('div.label', 'Login name:')
      with_tag('div.info', 'user2')
      with_tag('div.label', 'Display name:')
      with_tag('div.info', 'user2')
      with_tag('div.label', 'Email address:')
      with_tag('div.info', 'user2@example.com')
    end
  end

The problem with this approach is that we now have a spec which is extremely tightly bound to our view implementation, so tightly bound that changing the view in any way will almost certainly break the spec. This is good in the sense that mistakes introduced in the view will not likely escape our attention, but bad in the sense that it impedes change (and possible improvement) in the view.

It is also very duplicative: the view is the embodiment of the specification, written in ERB, and the spec is effectively identical but written in another language. If we are going to perform this level of pedantic, duplicative specification then we may as well just compare the output to a pre-computed string representation and require them to match.

It is my opinion that pure BDD shouldn’t apply to views; or to put it another way, adding text to a view shouldn’t necessarily require you to first write a failing spec that requires it. That kind of rigorous, line-by-line BDD is for code, in my opinion. Views are not code, or at least, if they are well written they shouldn’t be code (for example, see Terence Parr’s article, "Enforcing Strict Model-View Separation in Template Engines").

My rule of thumb is that you should write specs for everything in your view that involves code. The specs are there to help you catch errors in your code. So, in the example above, the items in the view that should be checked include:

  • That the list of users is produced (that the for construct works)
  • That the table rows alternate (that the cycle method has been correctly applied)
  • That the user attributes are shown

It would also be possible to check things such as:

  • That all the elements appear in the expected order
  • That each element has the appropriate identification for [[CSS]

But I think that going that far is not a wise investment of programmer time; such visual aspects are easily checked by inspection and automating their verification comes at a high cost because it makes future change more difficult.

See also