How do I... Do browser UI testing?
Can the bots use a browser? Yes they can. It's pretty simple actually, you just need to use Hound, an Elixir browser automation library, in your actions. In that way, it's not much different from using an HTTP library. In fact, you could create both an HTTP adapter, and a Hound adapter with the same API for your actions to call, and switch them out at runtime as needed. In other words, you can use the same bot trees and actions to run both API level tests and browser UI tests!
Getting set up is super easy - I've created a UI testing demo repository with full instructions and examples that you can just clone and modify as needed. Be sure to read the Hound docs.
UI testing is notoriously flakey. You can mitigate problems if you are disciplined. Here are a few tips:
- Be mindful of what you use as selectors.
:textis simple and closest to what the user will actually see, but if it changes (or you try to run your tests on a different internationalization language) your tests will break.:classes are a bad choice because they too can change, just like:tags.:ids or:names are better. - Clear separation of layers goes a long way. Luckily, the bots are already set up in clear layers: the tree describes the test, the actions perform the work, and if you've set up an adapter as suggested, that has the implementation specific calls.
- Just like actions, being atomic is a very good practice in your adapter's API. For a complex page, creating a page object is very useful.
- The hard part about browser testing is that sometimes elements haven't loaded yet when you try to test for them. Hound has a built-in retry system for most actions that involve inspecting the DOM. But some actions, like checking the page's title do not have built-in retries, and you might sometimes need them. See the retry_with_timeout example in the demo repo's actions for a simple generic retry implementation.
Happy automating!