22
loading...
Using React Testing Library with AG Grid
Having written a Podcast Player using AG Grid, we should really add some tests. This post shows how to use Jest and the React Testing Library with AG Grid.
All the code to support this blog post is available on Github in the React Data Grid Podcast Project, in particular the version 8 folder is the version where Testing Library was used.
A previous blog post described the development of the Podcast Player application. You can use the Podcast Player Online. This post describes how to add Automated Testing with Jest and the React Testing Library.
The Testing Library is a JavaScript testing library which works with multiple frameworks. It is not specifically a Unit Testing library because it is designed to target user visible interactions and changes, rather that internal object changes.
In the AG Grid Documentation there is an example of using Enzyme for Unit level testing. This uses low level API Access on the grid to allow access to the internals for the AG Grid.
In the examples in this post I use the Testing Library and do not access the AG Grid API. All the testing is performed through DOM interaction and DOM events. This makes the automating more like end to end tests but the difference is that we don't instantiate the full application, instead we instantiate the component or subset of components that we want to test and build 'just enough' DOM to make the components usable.
The Testing Library is added to our project when we use
create-react-app.We can run all the tests for our project with:
npm testI often use the verbose mode:
npm test -- --verbose
Since
create-react-app creates a default test for each project in App.test.js, we may find that our tests are broken as soon as we start development because one of the first things we do is change the heading for our application.import { render, screen } from '@testing-library/react';
import App from './App';
test('renders learn react link', () => {
render(<App />);
const linkElement = screen.getByText(/learn react/i);
expect(linkElement).toBeInTheDocument();
});For this reason, a lot of people immediately delete the
App.test.js file after creating their application with the create-react-app bootstrap.Instead, if we keep the test up to date with our application then we can gradually add more tests to our project without too much work.
For the Podcast Player application, I amended the default test to read:
test('renders the app', () => {
render(<App />);
const headerElement = screen.getByText(/Podcast Player/i);
expect(headerElement).toBeInTheDocument();
});This change doesn't really help me with my development, since it just looks on the page for the existence of some text, but since the text matches my
h1 I have a working test which I can then use to learn more about the Testing Library.The Testing Library is a set of packages to support testing UI components, without requiring a full browser or integrated environment.
It consists of:
user-event which makes it easier to simulate user interaction with the DOM.The Testing Library does not contain a test runner, but when
create-react-app is used, Jest is configured as the default test runner.The default test created by
create-react-app looks as follows:test('renders learn react link', () => {
render(<App />);
const linkElement = screen.getByText(/learn react/i);
expect(linkElement).toBeInTheDocument();
});To pull this apart to explain the main concepts.
test comes from Jest and allows us to write 'tests' and report on pass/failuresrender comes from the react wrapper of Testing Library and renders a React component in a virtual DOM, waiting for the initial events to complete.screen comes from the core Testing Library and is a convenience object which maps to document.body and has the query methods provided by Testing Library pre-bound e.g. allowing screen.getByText("Podcast")
expect comes from Jest and allows creating assertions with matchers
toBeInTheDocument is a matcher from the jest-dom library installed when we used create-react-app
After using
create-react-app most of the libraries are installed.To test the user interactions I also installed the
user-event and dom libraries:npm install --save-dev @testing-library/user-event @testing-library/domTests are stored in files with
test in the name, this allows them to be found by Jest when npm test is run.In the example source I created:
App.test.js this contains tests that explore the interaction between form fields and the grid.PodcastGrid.text.js this contains tests that look at the Grid component in isolation.Rather than use
AgGridReact directly in my application I wrap it in a component to make it easier to configure with properties and handle the support methods unique to the rendering of podcasts.This also allows me to test
PodcastGrid rather than AgGridReact.One of the decisions we have to make when testing using 3rd party components like AG Grid is the scope of what we will test.
I don't want to write tests to check that AG Grid works. I want to write tests that assert AG Grid renders the expected data, and any cell renderers I have created display the underlying data as I expect.
I don't want to test AG Grid, I want to test the interaction of my code with AG Grid. This fits in with the general philosophy of the Testing Library which focusses on the functionality from the user perspective, rather than the implementation.
Because AG Grid is a complex component I have to differ from the recommended approaches put forward by the Testing Library authors regarding locating elements.
Locating elements is a key part of automating any web application or component. We have to 'find' an element before we can interact with it to 'click', get attributes, check the text, etc.
To 'find' an element we have to query the DOM, and the Testing Library provides an extensive set of queries e.g.
getByLabelTextgetByRolegetByPlaceholderTextgetByTextThe recommendations from the Testing Library authors include:
data-testid attribute to identify elements on the screenI have a background in GUI and Web Automation using libraries like WebDriver. I tend to avoid locating elements by text because I often have to perform complex synchronisation and I want to have a high degree of control over the elements selected, so I prefer to use CSS Selectors to find elements in the DOM.
The Testing Library suggestions make a lot of sense when dealing with components, and in particular help avoid the Testing Library tests overlap too far into the internal Unit testing. Component testing with Testing Library will usually require less synchronisation because the component will be instantiated in a specific state, and then the visual elements and user focussed parts can be tested.
The main component I am testing is a data grid. A data grid will often render the same values in multiple cells in the grid, which can make querying by visible text difficult.
When dealing with 3rd party libraries that we are extending, we don't have the option of adding
data-testid attributes into the 3rd party components. As a personal preference I try to avoid adding automation specific attributes into my code, so I would tend to avoid data-testid for that reason alone, but with a locator strategy built around CSS Selectors it doesn't really matter.As a result of all of the above, rather than use the recommended query methods most of my element locators are built around
querySelector or querySelectorAll.One reason people avoid
querySelector as a query strategy is that it is vulnerable to DOM changes or, if we are basing the selectors on CSS Styling, that the CSS Styling locators can change independently of the functionality and we may find functional based tests failing due to styling changes.One mitigation approach for this is to use query selectors like
getByRole.The approach I have used is to add an abstraction layer so that our test code does not use the CSS Selectors directly, instead these are contained in an abstraction. The tests may fail if styling names change, but I only have to 'fix' this in one place in the code.
This can be as simple as:
const AudioLocator = {
source: "audio source"
}Where I could assert on an element attribute value with:
expect(
element.querySelector(AudioLocator.source).
getAttribute("src")).
toEqual("https://eviltester.com")One advantage of using CSS Selectors with AG Grid is that it has a 'theme' concept, which means that many of the CSS Selectors remain stable in order to avoid breaking existing styling themes. We are at risk of AG Grid changing between versions, if we use locators which are not used in theme styling.
Abstraction layers are also useful with complex 3rd party components because they can be shared between projects. For example Kerry McKeever has an AG Grid abstraction for Cypress which can be found on GitHub.
In the podcast project, the abstraction code is in the
AgGridTestUtils.js code.This has custom selector functions to query the DOM and locate elements, e.g.
const columnNamed = (cellName)=>{
return `.ag-cell[col-id="${cellName}"]`
}It also has the code used to synchronise the grid.
AG Grid performs most of its operations asynchronously so we have to synchronise our test execution with the grid, otherwise we may try and access an element before it is ready for interaction.
Most of these are functions which use the Testing Library's generic
waitFor functionconst waitForGridToBeInTheDOM=()=>{
return waitFor(() => {
expect(document.querySelector(".ag-root-wrapper")).toBeInTheDocument();
});
}Testing Library also has a built in synchronisation method to wait for an element to be removed:
waitForElementToBeRemoved.Because of the way I wrote my
PodcastGrid component, I have to mock out fetch requests when testing the component.Writing test code is a good way to identify architecture decisions in your code which may not be optimal.
If a component is hard to automate in a
test function, then it may be time to rewrite it.But... ideally write a
test function to cover the functionality first, then when you amend the code, you have a something that checks if you made any errors during any refactoring.I'll break down one of the tests for the
PodcastGrid to make it easier to understand how to automate AG Grid.it("renders user data from a url", async () => {The tests are written to describe the condition being tested as an assertion i.e. the PodcastGrid can render user data from a url.
Then I need to mock out the
fetch request made by the PodcastGrid in a useEffect to load the RSS feed.First I create the data for the RSS feed in the
fakeRSSFeed constant.const fakeRSSFeed =
`<channel>
<item>
<title>Fake Episode</title>
<pubDate>Thu, 23 Sep 2021 10:00:00 +0000</pubDate>
<enclosure url="https://eviltester.com"/>
<description>
<![CDATA[<p>Fake Description</p>]]>
</description>
</item>
</channel>`;Because we are testing at a component level it is very easy to control the data we need for the testing without having to setup any web servers to deploy files.
A standard function of
jest is used to spy on any fetch calls, and instead of issuing an HTTP request, simply return an object that represents the results of having made a fetch.jest.spyOn(window, "fetch").mockImplementation(() =>{
return Promise.resolve({
text: () => fakeRSSFeed
})}
);Then I render the component being tested in the DOM.
render(<PodcastGrid
rssfeed="https://fakefeed"
height="500px"
width="100%"
quickFilter=""/>);And with all the setup out of the way, I can then start to automate AG Grid.
The first thing I have to do is wait for the grid to be fully rendered and populated in the DOM.
await AgGridTest.waitForGridToBeInTheDOM();
await AgGridTest.waitForDataToHaveLoaded();Both of the above synchronisation methods are contained in the
AgGridTestUtils.js code.waitForGridToBeInTheDOM waits for the basic grid to be present in the DOM.waitForDataToHaveLoaded waits until any 'loading' indicator is no longer present.At this point AG Grid will then be asynchronously populating the grid with data, so if I start to interact with the Grid now, I'll likely get flaky tests that will work sometimes but at other times fail.
So I add an additional level of synchronisation to check that the pagination has rendered as I expect.
await AgGridTest.waitForPagination().
then((pagination)=>{
expect(pagination.firstRow).toEqual("1");
expect(pagination.lastRow).toEqual("1");
expect(pagination.rowCount).toEqual("1");
});At this point I will check that the cell that represents the
mp3 for the podcast episode has been rendered as an HTML audio control and has the URL from the RSS feed.This code is also wrapped in a synchronisation
waitFor function because the audio control can be rendered asynchronously.// the audio component may take a little extra time to render so waitFor it
await waitFor(() => {
expect( AgGridTest.getFirstRowWithNamedCellValue("title", "Fake Episode").
querySelector(AgGridTest.columnNamed('mp3')).
querySelector(AudioLocator.source).
getAttribute("src")).
toEqual("https://eviltester.com")
}
)Finally we remove the mock and override from the
fetch function.// remove the mock to ensure tests are completely isolated
global.fetch.mockRestore();
});Some important things to note about the test code are:
querySelector is a very flexible way to work with 3rd party componentsAlso, these tests are first draft. The more tests that I add to the project, the more abstraction layers I would build up so the code asserting on the audio component mp3 url would move into a 'waitForAudioComponent' abstraction function to make it re-usable and simplify the test.
The components would be automated and checked individually in isolation first, prior to checking the interaction as an application.
The components are integrated through the use of the
App component, which renders:PodcastGrid component which renders AG GridMost of the functionality for
PodcastGrid would be tested in isolation. But I still wanted to make sure that when I click a feed from the dropdown, the grid is populated with the RSS feed.This is not a full integration test because the
App component will be tested in the DOM in isolation, but this approach demonstrates that it is possible to gradually build up 'integration' coverage without needing to do all the testing on a deployed application.At this point I extend my use of Testing Library to cover user events.
Testing Library comes with built in support for low level event firing which can be useful for testing components in isolation.
When working in a more integrated setting then the User Event Extension can help offer a more realistic user interaction event sequence e.g. a 'click' event from the
userEvent will hover over the element, priori to issuing the mouse events.To demonstrate an integration test, I'll annotate one of the tests below.
The test begins with a description of the expectation.
it("loads feed into grid when button pressed", async () => {We have to setup mocking of the
fetch again. In this case I have two RSS feed variables, because I want to make sure that when the drop down is selected, the url is populated in the text field and a call is made for the correct feed, rather than just returning the same RSS Feed regardless of the URL chosen.jest.spyOn(window, "fetch").mockImplementation((aUrl) =>{
if(aUrl==="https://feed.pod.co/the-evil-tester-show"){
return Promise.resolve({text: () => fakeEvilFeed});
}else{
return Promise.resolve({text: () => fakeWebrushFeed});
}
});We then render the application component.
render(<App />);At this point we can use the recommended Testing Library methods for element location because the HTML is simpler and under our full control with no 3rd party libraries used. It's basically just a
select drop down and url input field.I can find the drop down using its label text with
getByLabelText and then the userEvent abstraction layer to select a specific option from the drop down with selectOptions.userEvent.selectOptions(screen.getByLabelText("Choose a podcast:"),'The Evil Tester Show');Having selected the drop down, other code will be executed which populates the url and when the
Load Feed button is clicked the props on the PodcastGrid will be updated, causing the grid to re-render and load data into the grid.const loadButton = screen.getByText("Load Feed");
userEvent.click(loadButton);Since the grid will load data we need to synchronise on its updating:
await AgGridTest.waitForGridToBeInTheDOM();
await AgGridTest.waitForDataToHaveLoaded();Because the
PodcastGrid component has its own set of tests, all I do in the Integration test is the minimal assertion to check that the title of the correct podcast episode is present in the grid.// wait for first cell to expected data
await waitFor(() => {
expect(AgGridTest.getNamedCellsWithValues("title", "Fake Evil Tester Episode").length).toEqual(1);
});And finally tidy up and remove the mock fetch.
// remove the mock to ensure tests are completely isolated
global.fetch.mockRestore();
});There are many ways to automate applications which are using AG Grid components. We can use external automating with libraries such as Selenium WebDriver, Puppeteer, Playwright and Cypress as demonstrated by Kerry McKeever's AG Grid Cypress Extension.
We can also incorporate tests into our Unit automation, using Testing Library as this example shows, or using Enzyme as this Enzyme example in AG Grid Documentation demonstrates. The Enzyme example shows the use of the AG Grid API directly offering a tighter interaction with the grid, which might be useful if you are concentrating on complex filtering code.
The Testing Library example here works purely at a DOM level and uses abstraction layers to make the tests readable and maintainable.
If you are interested in Testing Library then you should certainly check out the Testing Library Web Site, and might find the React Testing Recipes useful.
All the code to support this blog post is available on Github in the React Data Grid Podcast Project, in particular the version 8 folder is the version where Testing Library was used.
22