Over the past 9 years of Sprout Social’s existence, our Selenium framework has grown into 100’s of scenarios with as many page classes and factory containers. We’ve noticed it has become cumbersome to manage and as such, our QA Engineers thought the models in use needed a refresh. So in recent times, we adopted what we like to call the Component Model.

Note: All code examples can be found in the following Github Gist.

The idea for this component model was a result of a few things:

  • Modern websites aren’t really built strictly of static “pages” any more
  • A Grow@Sprout talk on the SOLID principles by Uncle Bob
  • Our web app’s use of React.js, which uses a similar idea of components for encapsulation

    Modern websites (i.e. web applications) are highly dynamic pieces of software with global bars, side bars, content views, and pop-outs. Many of these pieces are yet composed of other smaller pieces. Reflecting on this and the SOLID principles, we realized that the monolithic classes the page-object model created were not working to our advantage anymore.

    We noticed we had multiple functional zones and really only one spot for actual page content. We had a global top nav bar that also included settings access. Additionally, there is a secondary left nav bar for navigating within functional areas. A lot of pages also contained a rail on the right side commonly housing date pickers and filters for the content view. Please refer to the mockup below to help visualize this:

We noticed plenty of areas in the framework where we could DRY (Don’t Repeat Yourself) up the code. This is included in the SOLID software development principles. Buttons, inputs, and data visualizations all share common code in the UI implementation. Why, then, should our tests implement multiple ways to get and interact with filters across multiple page classes?

Instead, we agreed that the components our selenium tests use should be a 1:1 representation of their equivalent web UI component. Take this example CheckboxV1 (click for the gist code snippet) class. The most beneficial aspects of this model are: the component houses its unique selenium selector, and specific constructors and methods that your tests can consume. This significantly enhances the readability of code and the maintainability of the framework.

After using these components for a while, we started to notice that code was being repeated between classes. In order to DRY the repetition up, we created a class called BaseComponent to control much of the basic behavior of any component. One can then extend all other components from this BaseComponent to ensure every component implementation is as simple as possible.

Combining implementation from our original CheckboxV1 with inheritance from BaseComponent further simplifies the component and increases the maintainability of the framework. CheckboxV2 allows BaseComponent to do most of the heavy lifting not specific to the particular component. We also have one less spot to check when fixing issues as most of the behavior is within BaseComponent instead of multiple versions of similar functionality scattered across multiple components.

A quick note on the above CheckboxV2 implementation - we tend to use the String argument constructor the most with components that have attributes in the UI code. The By and WebElement argument constructors are backups for elements in our UI that might not contain special data attributes, what we refer to as QA attributes, but the follow same/similar component structure in the DOM. However, we encourage our team to add their own QA selectors if able to or further collaborate with the developers to get them added. Having them usually allows the component code and subsequent Selenium automation to be much simpler overall.

Here is a quick example of how we then use the components within a Page and Step Definition classes.

It has been a fun and interesting journey getting to this point through many refactors of our framework. However, it has proved time and time again to make our framework a bit easier to grok as well as maintain. We are already beginning to think of ways to improve upon this further such as with a wrapper class called Selector to help with formatting and making BaseComponent more friendly to work with. We look forward to providing more information on those in the future

Have you or your organization used or are currently using something similar? Please share your experiences in the comments!

2017, it's been too short... 2018, where did you come from?

It goes without saying, I didn’t exactly get to all the things I wanted to get to this year; however, I did get to many things I didn’t k...… Continue reading