Black Sheep Code

What is the role of QA and how should their tests fit into the development process?

This is a topic that I don't have a strong opinion to land on.

Instead I've got two contradictory ideas that I'm struggling to reconcile.

I like having QAs in the organisations.

Some people say, 'just don't have QAs, devs should test their own work'.

I agree that devs should test their own work - and specifically be producing suites of:

With all that said - I like the safety net that the QAs provide. Let's talk about how something might slip through, even if the developer is testing their own work.

Additionally, I have found that often it is the QA who knows the product the best. There may be a section of the application that as a newer developer on the team, I'm not actually sure how it works/what it is meant to do. The QA is often the person who knows best.

Should QAs write automated browser tests?

A fairly standard process is that:

  1. Business Analyst/Product Owner defines requirements
  2. Developers implement the requirements, write their own tests, and release the code/make a release candidate.
  3. QAs do a 'regression suite' checking that every part of the application still works as intended.

In its most basic form, this regression suite consists of the QAs manually clicking through to explore every aspect of the application. They may be following a check list of steps to follow written in a word document.

This is repetitive, boring work, and prone to error.

So this is where QA might propose - 'Hey we can automate this process using tools like Cypress or Selenium'.

So far, no problems, the automated suites are just an extension of the manual process that the QAs are otherwise running.

Should these automated browser tests be added to the CI pipeline?

Adding those tests to the CI pipeline is where I've seen the process breakdown.

The idea is, 'we'll run the automated browser tests when the dev submits a PR, and if they've broken anything they'll know about it sooner'.

The problem is that unless the person writing the automated test, and the developer creating the code being tested are aligned, it can be easy to for tests to fail, without actually breaking anything.

As an example, lets say we have some code like:

const StyledWidget = styled.div` 

    .action-button {
      color: red; 
      border: solid 1px red; 


export const Widget = () => {

  return <StyledWidget>
      <button className ="action-button" onClick ={() => {
        //does something
      }}>Click me</button>

And lets say the QA has written a cypress test like:

const actionButton = cy.get('.action-button'); 
// Do something with the action button 

This test is working fine.

Now, lets say that the devs have agreed that it's really dumb to manually style those buttons everywhere, instead lets create a reusable design system that adds that red text and border for us.

So they change the code to:

export const OurButton = styled.button`
   color: red; 
   border: solid 1px red; 

export const Widget = () => {

  return <div>
    <OurButton onClick ={() => {
      //does something
    }}>Click me</OurButton>

The cypress tests now fail because the .action-button class no longer exists on the element, and the selector no longer works.

Now you might argue that 'Hey, the dev shouldn't remove that class, the class should be considered part of that components functionality, because it is relied on for tests'.

In this particular instance you might be right; especially if you've got a codebase that already has a lot of tests of this style, I would suggest that when the developers do this refactor to their new design system, as a rule they retain the classNames on their interactable elements, in order to prevent breaking existing tests.

However, let's make the example a little more complicated.

export const Widget = () => {

  return <div className = "widget-outer">
    <div className ="widget-inner-left">
      <div className ="action-button-container">
        <button className ="action-button" onClick ={() => {
          //does something
        }}>Click me</button>
    <div className = "widget-inner-right">


Here there are a lot of class names, and it's unclear which of these class names are safe to remove, and which are not.

There's a few rules you could go for, and none of them are particularly satisfying:

What I recommend - use testing library style accessible selectors.

I recommend going all in on the Testing Library philosophy of testing your frontend by using accessible selectors.

That is - instead of selecting by class name or data attribute, you select by aria attributes like 'role'.

As a major additional benefit is that this quite naturally guides your developers to writing an accessible application, without even intending to. This can be an important point of product differentiation.

In the above example, the cypress test could be written as:

cy.findByRole('button', {name: "Click me"}); 

Now it doesn't matter if the class name changed, or those surrounding divs changed or were removed, all that matters is that there is a button with the label "Click me".

This still won't prevent tests from breaking.

You might be thinking 'Ah! But what if you change the text of your button, then it will break!'. And you'd be right.

On this - two points:


export const Widget = () => {

  return <div>
    <OurButton onClick ={() => {
      //does something
cy.findByRole('button', {name: TEXT_CONSTANTS.CLICK_ME}); 

Devs should write examples of the accessible selectors.

I'll acknowledge that writing accessible selectors isn't always as straight forward as advertised, especially if your devs or QAs haven't done it before.

So what I suggest is that when devs write their components, they're also including a frontend unit test, that includes demonstrating the use of the accessible selector.

eg, for the OurButton component, the dev should write an RTL test that looks like:

describe("OurButton", () => {
   it("has an accessible selector", () => {

     const fakeHandleClick = jest.fn(); 

     render(<OurButton onClick = {fakeHandleClick}>foobar</OurButton>); 

     const button = screen.getByRole("button", {
       name: "foobar"

Now, if someone is writing a test that needs to interact with one of these buttons - they can see how the write the selector right there.

This is especially important for things like checkboxes and radio buttons - which can actually be quite difficult to write accessible selectors for.

Maybe devs should write all the browser tests

But this brings us kind of full circle - if we're saying that only way to write automated browser tests in a practical way that keeps them aligned, is that developers and people writing the automated browser tests need to be aligned on how they write components such that they're selectable via an accessible selector, doesn't it just make sense that the developers would write the tests?

Possibly, but:

Shift left

Where this seems to take us is into the shift left philosophy - where we're getting testers to be more active in the code base, and their job resembles a developer a bit more closely.

I think where organisations can start struggling here is:

A counterpoint

After writing this post, I come across this video which makes a good point. It makes the argument that a QAs role isn't test execution it should be telling people what tests need to be covered.


Sorry, this post ends abruptly. This post was half finished and I didn't end up coming back to it to finish my thought, which I've since forgotten. Maybe a part two will be coming.


Questions? Comments? Criticisms? Get in the comments! 👇

Spotted an error? Edit this page with Github