Experience GraphQL Summit 2024

You are viewing documentation for a previous version of this software.

Switch to the latest stable version.

Testing React components

Have peace of mind when using React Apollo in production

Running tests against code meant for production has long been a best practice. It provides additional security for the code that's already written, and prevents accidental regressions in the future. Components utilizing React Apollo, the React implementation of Apollo Client, are no exception.

Although React Apollo has a lot going on under the hood, the library provides multiple tools for testing that simplify those abstractions, and allows complete focus on the component logic. These testing utilities have long been used to test the React Apollo library itself, so they will be supported long-term.

An introduction

The React Apollo library relies on React's context to pass the ApolloClient instance through the React component tree. In addition, React Apollo makes network requests in order to fetch data. This behavior affects how tests should be written for components that use React Apollo.

This guide will explain step-by-step how to test React Apollo code. The following examples use the Jest testing framework, but most concepts should be reusable with other libraries. These examples aim to use as simple of a toolset as possible, so React's test renderer will be used in place of React-specific tools like Enzyme and react-testing-library.

Note: As of React Apollo 3, all testing utilities can now be found in their own @apollo/react-testing package.

Consider the component below, which makes a basic query, and displays its results:

JavaScript
JavaScript
hooks
1import React from 'react';
2import gql from 'graphql-tag';
3import { useQuery } from '@apollo/react-hooks';
4
5// Make sure the query is also exported -- not just the component
6export const GET_DOG_QUERY = gql`
7  query getDog($name: String) {
8    dog(name: $name) {
9      id
10      name
11      breed
12    }
13  }
14`;
15
16export function Dog({ name }) {
17  const { loading, error, data } = useQuery(
18    GET_DOG_QUERY,
19    { variables: { name } }
20  );
21  if (loading) return <p>Loading...</p>;
22  if (error) return <p>Error!</p>;
23
24  return (
25    <p>
26      {data.dog.name} is a {data.dog.breed}
27    </p>
28  );
29}

Given this component, let's try to render it inside a test, just to make sure there are no render errors:

JavaScript
1// Broken because it's missing Apollo Client in the context
2it('should render without error', () => {
3  renderer.create(<Dog name="Buck" />);
4});

This test would produce an error because Apollo Client isn't available on the context for the useQuery Hook to consume.

In order to fix this we could wrap the component in an ApolloProvider and pass an instance of Apollo Client to the client prop. However, this will cause the tests to run against an actual backend which makes the tests very unpredictable for the following reasons:

  • The server could be down.

  • There may be no network connection.

  • The results are not guaranteed to be the same for every query.

JavaScript
1// Not predictable
2it('renders without error', () => {
3  renderer.create(
4    <ApolloProvider client={client}>
5      <Dog name="Buck" />
6    </ApolloProvider>,
7  );
8});

MockedProvider

The @apollo/react-testing package exports a MockedProvider component which simplifies the testing of React components by mocking calls to the GraphQL endpoint. This allows the tests to be run in isolation and provides consistent results on every run by removing the dependence on remote data.

By using this MockedProvider component, it's possible to specify the exact results that should be returned for a certain query using the mocks prop.

Here's an example of a test for the above Dog component using MockedProvider, which shows how to define the mocked response for GET_DOG_QUERY:

JavaScript
1// dog.test.js
2
3import { MockedProvider } from '@apollo/react-testing';
4
5// The component AND the query need to be exported
6import { GET_DOG_QUERY, Dog } from './dog';
7
8const mocks = [
9  {
10    request: {
11      query: GET_DOG_QUERY,
12      variables: {
13        name: 'Buck',
14      },
15    },
16    result: {
17      data: {
18        dog: { id: '1', name: 'Buck', breed: 'bulldog' },
19      },
20    },
21  },
22];
23
24it('renders without error', () => {
25  renderer.create(
26    <MockedProvider mocks={mocks} addTypename={false}>
27      <Dog name="Buck" />
28    </MockedProvider>,
29  );
30});

The mocks array takes objects with specific requests and their associated results. When the provider receives a GET_DOG_QUERY with matching variables, it returns the corresponding object from the result key. A result may alternatively be a function returning the object:

JavaScript
1const mocks = [
2  {
3    request: {
4      query: GET_DOG_QUERY,
5      variables: {
6        name: 'Buck',
7      },
8    },
9    result: () => {
10      // do something, such as recording that this function has been called
11      // ...
12      return {
13        data: {
14          dog: { id: '1', name: 'Buck', breed: 'bulldog' },
15        },
16      }
17    },
18  },
19];

Your mock request's variables object must exactly match the query variables sent from your component.

addTypename

You may notice the prop being passed to the MockedProvider called addTypename. The reason this is here is because of how Apollo Client normally works. When a request is made with Apollo Client normally, it adds a __typename field to every object type requested. This is to make sure that Apollo Client's cache knows how to normalize and store the response. When we're making our mocks, though, we're importing the raw queries without typenames from the component files.

If we don't disable the adding of typenames to queries, the imported query won't match the query actually being run by the component during our tests.

In short, if queries are lacking __typename, it's important to pass the addTypename={false} prop to the MockedProviders.

Testing loading states

In this example, the Dog component will render, but it will render in a loading state, not the final response state. This is because MockedProvider doesn't just return the data but instead returns a Promise that will resolve to that data. By using a Promise it enables testing of the loading state in addition to the final state:

JavaScript
1it('should render loading state initially', () => {
2  const component = renderer.create(
3    <MockedProvider mocks={[]}>
4      <Dog />
5    </MockedProvider>,
6  );
7
8  const tree = component.toJSON();
9  expect(tree.children).toContain('Loading...');
10});

This shows a basic example test that tests the loading state of a component by checking that the children of the component contain the text Loading.... In an actual application, this test would probably be more complicated, but the testing logic would be the same.

Testing final state

Loading state, while important, isn't the only thing to test. To test the final state of the component after receiving data, we can just wait for it to update and test the final state.

JavaScript
1const wait = require('waait');
2
3it('should render dog', async () => {
4  const dogMock = {
5    request: {
6      query: GET_DOG_QUERY,
7      variables: { name: 'Buck' },
8    },
9    result: {
10      data: { dog: { id: 1, name: 'Buck', breed: 'poodle' } },
11    },
12  };
13
14  const component = renderer.create(
15    <MockedProvider mocks={[dogMock]} addTypename={false}>
16      <Dog name="Buck" />
17    </MockedProvider>,
18  );
19
20  await wait(0); // wait for response
21
22  const p = component.root.findByType('p');
23  expect(p.children).toContain('Buck is a poodle');
24});

Here, you can see the await wait(0) line. This is a utility function from the waait npm package. It delays until the next "tick" of the event loop, and allows time for that Promise returned from MockedProvider to be fulfilled. After that Promise resolves (or rejects), the component can be checked to ensure it displays the correct information — in this case, "Buck is a poodle".

For more complex UI with heavy calculations, or delays added into its render logic, the wait(0) will not be long enough. In these cases, you could either increase the wait time or use a package like wait-for-expect to delay until the render has happened. The risk of using a package like this everywhere by default is that every test could take up to five seconds to execute (or longer if the default timeout has been increased).

Testing error states

Since they can make or break the experience a user has when interacting with the app, error states are one of the most important states to test, but are often less tested in development.

Since most developers would follow the "happy path" and not encounter these states as often, it's almost more important to test these states to prevent accidental regressions.

To simulate a network error, an error property can be included on the mock, in place of or in addition to the result.

JavaScript
1it('should show error UI', async () => {
2  const dogMock = {
3    request: {
4      query: GET_DOG_QUERY,
5      variables: { name: 'Buck' },
6    },
7    error: new Error('aw shucks'),
8  };
9
10  const component = renderer.create(
11    <MockedProvider mocks={[dogMock]} addTypename={false}>
12      <Dog name="Buck" />
13    </MockedProvider>,
14  );
15
16  await wait(0); // wait for response
17
18  const tree = component.toJSON();
19  expect(tree.children).toContain('Error!');
20});

Here, whenever the MockedProvider receives a GET_DOG_QUERY with matching variables, it will return the error assigned to the error property in the mock. This forces the component into the error state, allowing verification that it's being handled gracefully.

To simulate GraphQL errors, define errors with an instantiated GraphQLError object that represents your error, along with any data in your result.

JavaScript
1const dogMock = {
2  // ...
3  result: {
4    errors: [new GraphQLError('Error!')],
5  },
6};

Testing mutation components

useMutation based components are tested very similarly to useQuery components. The only key difference is how the operation is fired. With useQuery the query is fired when the wrapping component mounts, whereas with useMutation the mutation is fired manually, usually after some user interaction like pressing a button.

Consider this component that calls a mutation:

JavaScript
JavaScript
hooks
1export const DELETE_DOG_MUTATION = gql`
2  mutation deleteDog($name: String!) {
3    deleteDog(name: $name) {
4      id
5      name
6      breed
7    }
8  }
9`;
10
11export function DeleteButton() {
12  const [mutate, { loading, error, data }] = useMutation(DELETE_DOG_MUTATION);
13
14  if (loading) return <p>Loading...</p>;
15  if (error) return <p>Error!</p>;
16  if (data) return <p>Deleted!</p>;
17
18  return (
19    <button onClick={() => mutate({ variables: { name: 'Buck' } })}>
20      Click me to Delete Buck!
21    </button>
22  );
23}

Testing an initial render for this component looks identical to testing our useQuery based component.

JavaScript
1import DeleteButton, { DELETE_DOG_MUTATION } from './delete-dog';
2
3it('should render without error', () => {
4  renderer.create(
5    <MockedProvider mocks={[]}>
6      <DeleteButton />
7    </MockedProvider>,
8  );
9});

Calling the mutation is where things get interesting:

JavaScript
1it('should render loading state initially', () => {
2  const deleteDog = { name: 'Buck', breed: 'Poodle', id: 1 };
3  const mocks = [
4    {
5      request: {
6        query: DELETE_DOG_MUTATION,
7        variables: { name: 'Buck' },
8      },
9      result: { data: { deleteDog } },
10    },
11  ];
12
13  const component = renderer.create(
14    <MockedProvider mocks={mocks} addTypename={false}>
15      <DeleteButton />
16    </MockedProvider>,
17  );
18
19  // find the button and simulate a click
20  const button = component.root.findByType('button');
21  button.props.onClick(); // fires the mutation
22
23  const tree = component.toJSON();
24  expect(tree.children).toContain('Loading...');
25});

This example looks very similar to the useQuery based component above, but the difference comes after the rendering is completed. Since this component relies on a button to be clicked to fire a mutation, the renderer's API is used to find the button.

After a reference to the button has been obtained, a "click" on the button can be simulated by calling its onClick handler. This will fire off the mutation, and then the rest will be tested identically to the useQuery based component.

Note: Other test utilities like Enzyme and react-testing-library have built-in tools for finding elements and simulating events, but the concept is the same: find the button and simulate a click on it.

To test for a successful mutation after simulating the click, the fulfilled Promise from MockedProvider can be checked for the appropriate confirmation message, just like the useQuery based component:

JavaScript
1it('should delete and give visual feedback', async () => {
2  const deleteDog = { name: 'Buck', breed: 'Poodle', id: 1 };
3  const mocks = [
4    {
5      request: {
6        query: DELETE_DOG_MUTATION,
7        variables: { name: 'Buck' },
8      },
9      result: { data: { deleteDog } },
10    },
11  ];
12
13  const component = renderer.create(
14    <MockedProvider mocks={mocks} addTypename={false}>
15      <DeleteButton />
16    </MockedProvider>,
17  );
18
19  // find the button and simulate a click
20  const button = component.root.findByType('button');
21  button.props.onClick(); // fires the mutation
22
23  await wait(0);
24
25  const tree = component.toJSON();
26  expect(tree.children).toContain('Deleted!');
27});

The result in a mocked mutation may be a function rather than an object. This gives you a simple way to check that a mutation has been called:

JavaScript
1it('should delete and give visual feedback', async () => {
2  const deleteDog = { name: 'Buck', breed: 'Poodle', id: 1 };
3  let deleteMutationCalled = false;
4  const mocks = [
5    {
6      request: {
7        query: DELETE_DOG_MUTATION,
8        variables: { name: 'Buck' },
9      },
10      result: () => {
11        deleteMutationCalled = true;
12        return { data: { deleteDog } };
13      },
14    },
15  ];
16
17  const component = renderer.create(
18    <MockedProvider mocks={mocks} addTypename={false}>
19      <DeleteButton />
20    </MockedProvider>,
21  );
22
23  // find the button and simulate a click
24  const button = component.root.findByType('button');
25  button.props.onClick(); // fires the mutation
26
27  await wait(0);
28
29  expect(deleteMutationCalled).toBe(true);
30
31  const tree = component.toJSON();
32  expect(tree.children).toContain('Deleted!');
33});

For the sake of simplicity, the error case for mutations hasn't been shown here, but testing useMutation errors is exactly the same as testing useQuery errors: just add an error to the mock, fire the mutation, and check the UI for error messages.

Testing UI components isn't a simple issue, but hopefully these tools will create confidence when testing components that are dependent on data.

For a working example showing how to test components, check out this project on CodeSandbox:

Edit React-Apollo Testing
Feedback

Edit on GitHub

Forums