Why even use data attributes as test selectors?

Integration and acceptance tests usually interact with and assert on the presence of certain elements in the markup that an application renders. These elements are identified using CSS selectors. Most projects use one of three approaches for CSS selectors in tests:

Selectors based on HTML structure

This approach simply selects elements by their position in the rendered HTML. For the following template:

<article>
  <h1>Post Title</h1>
  <p>Post Body…</p>
</article>

one might select the post's title with the selector 'article h1'. Of course this breaks when changing the <h1> to a <h2> while the functionality being tested is probably not affected by that change.

Selectors based on CSS classes

This approach selects elements by CSS classes. For the following template:

<article>
  <h1 class="post-title">{{post.title}}</h1>
  <p>{{post.body}}</p>
</article>

one might select the post title with the selector '.post-title'. This of course breaks when the CSS class is changed or renamed, although that would only be a visual change which shouldn't affect the tests at all.

Many projects use CSS classes with special prefixes that are only used for testing to overcome this problem like 'js-post-title'. While that approach is definitely more stable it is often hard to maintain. Also it doesn't easily allow to encode additional information like e.g. the post's id.

Selectors based on data attributes

This approach uses HTML 5 data attributes to select elements. For the following template:

<article>
  <h1 data-test-selector="post-title">{{post.title}}</h1>
  <p>{{post.body}}</p>
</article>

one would select the post's title with the selector *[data-test-selector="post-title"] (which selects any element with a data-test-selector attribute that has the value "post-title"). While the selector is arguably a bit longer this approach clearly separates the test selectors from the rest of the markup and is resilient to change as the attribute can be applied to any element rendering the post's title, regardless of the HTML structure, CSS classes etc. Also it easily allows encoding more data in the markup like e.g. the post's id:

<article>
  <h1 data-test-selector="post-title" data-test-resource-id="{{post.id}}">{{post.title}}</h1>
  <p>{{post.body}}</p>
</article>

ember-test-selectors makes sure to remove all these data attributes in the production environment so that users will have perfectly clean HTML delivered:

<article>
  <h1>My great post</h1>
  <p>Bla bla…</p>
</article>

Future Plans

We have some future plans for ember-test-selectors to make working with data attributes as test selectors even more convenient:

• custom test helpers that find elements by data attributes so that you don't have to write the quite long selectors yourselves; probably sth. like findViaTestSelector('selector', 'post-title'), https://github.com/simplabs/ember-test-selectors/issues/8
• template helpers that generate data attributes for elements, e.g. <h1 {{test-selector="post-title"}}>{{post.title}}</h1> which would result in <h1 data-test-selector="post-title">{{post.title}}</h1> or <div {{test-selector post}}>…</div> which would result in <div data-test-selector="post-1">…</div>, https://github.com/simplabs/ember-test-selectors/issues/9