microsoft
[Feature] Allow mocking third party and built-in modules (for non-E2E tests)
Feature Request
Playwright Test can also be used as a "generic" test runner to run unit/integration tests (see #14268).
For such tests it is often helpful, if not necessary, to mock modules loaded by CJS require or ESM import statements. This allows to test a module in isolation.
At the moment, Playwright does not have a way to replace modules with mocks when they get loaded. As discussed in #14398, Node.js solutions to mock modules like proxyquire or testdouble cannot be used because Playwright has a custom implementation to load modules (for technical reasons) and therefore these solutions don't work.
It would be great if Playwright Test would have some sort of API to mock CJS/ESM modules, similar to the things testdouble is capable of.
Their documentation shows that there would be some things to think through, for example:
- is it possible to replace a module after the subject under test was imported/required?
- is it possible to access (parts of) the original implementation after a module mock was applied?
- should there be an API to "reset" mocked modules?
Idea
Example is derived from https://jestjs.io/docs/mock-functions#mocking-modules, API inspired by https://github.com/testdouble/testdouble.js#specifying-a-custom-replacement.
users.ts:
import axios from 'axios';
class Users {
static all() {
return axios.get('/users.json').then((resp) => resp.data);
}
}
export default Users;
my-test.spec.ts:
import { test, mock } from '@playwright/test';
import Users from './users';
test('should fetch users', async ({}) => {
const fakeUsers = [{ name: 'Bob' }];
/* mock axios.get (would be great if that works even though ./users was imported already */
mock.replace('axios', {
get: mock.func(() => Promise.resolve({ data: fakeUsers })),
});
const actualUsers = await Users.all();
expect(actualUsers).toEqual(fakeUsers);
});
@pkerschbaum We do not prioritize this feature since we focus on e2e tests. Let's wait and collect some upvotes.
Not sure if that would work, but maybe manually compiling tests with custom tsconfig.json where modules are mapped to their mocked implementations? See here
Example:
{
"compilerOptions": {
"baseUrl": ".",
"paths": {
"axios": ["./playwright/mocks/axios"]
}
}
}
@arvigeus your idea is interesting!
I think this way you could create "global mocks", still it would not have the same feature set as my proposed idea (e.g. you could not define a separate mock per test).
One note on this: TypeScript does not offer any option to replace path aliases in the compiled output, see https://github.com/microsoft/TypeScript/issues/26722 and the more recent https://github.com/microsoft/TypeScript/issues/45862.
There are different solutions to this problem, my current recommendation is to use github.com/nonara/ts-patch in conjunction with github.com/LeDDGroup/typescript-transform-paths.
Like here: https://github.com/pkerschbaum/react-pdf-document-generator/tree/b32fa3b452cda2c0320607e96c3bc7c71965a399/packages/config.
Personally I would find this useful to mock-out some back-end services/logic to simplify test setup. I know there are different ways of going about this, but one way it would help me would be to allow me to largely preserve my authentication stack while only mocking a single function server-side.
Let me understand this correctly. There is no way to mock an Axios call to a remote server at present? This is a dealbreaker. I won't be standing up every remote service I need, or build my test suite to be robust enough to make use of arbitrary results returned by a remote service, when what I really need is to run my client-side test suite. If I understand the state of play correctly we won't be using Playwright at all. Which is sad.
Let me understand this correctly. There is no way to mock an Axios call to a remote server at present?
This ask is about mocking abitrary JS modules in the context of being a "generic" unit test runner, not network requests. Mocking network requests works today via standard e2e testing: https://playwright.dev/docs/mock
Let me understand this correctly. There is no way to mock an Axios call to a remote server at present?
This ask is about mocking abitrary JS modules in the context of being a "generic" unit test runner, not network requests. Mocking network requests works today via standard e2e testing: https://playwright.dev/docs/mock
Not if the network requests are made with Axios. Mock them all you like in your test, you'll continue to make real requests. I migrated my app to use fetch instead of Axios as an experiment, and the mocks started working.
That's odd 🤔. Mocks work with both fetch and XHR (which is what Axios uses IIRC). Not sure why it'd go around that unless something weird was happening. I'd file a ticket or start a discussion in case there's a legit bug there.
Cypress has this module/function mocking built-in, and it's very useful.
Imagine that we want to do integration/end-to-end testing of a feature that uses feature flags, and we don't want to rely on the network response from the feature flag service, which would return data that's not stable, i.e., data that's not controlled by the test case. The ability to mock the feature flag SDK module function would facilitate testing those features and forcing the feature flags into a repeatable known state.
While I agree that all module mocking can be replaced with making the code more parameter driven (instead of import driven), nonetheless modules are a heavy part of JavaScript ecosystems and module mocking is fundamental to modern Testing frameworks.
I think this is also reflected by the number of 👍🏻 it received :rose: https://github.com/microsoft/playwright/issues?q=is%3Aissue+is%3Aopen+sort%3Areactions-%2B1-desc
I also see this useful in E2E tests. For example, we call third-party library that resolves a promise and I would very much be able to mock that.
The api in this case looks like this:
import {showZXCConnect} from '@zxc-api/connect';
and then you call it:
const resultToken = await showZXCConnect(link)
I should love to be able to mock showZXCConnect so it just returns some dummy token.
This would be useful in our visual regression tests. We generate lorem ipsum in some storybook stories which are checked for visual diffs. I'd like to mock this lorem generation method to return a constant string when running playwright tests so it doesn't cause a png comparison failure.
It is impossible for now to get access or mock at least chrome.storage.local/sync store, which is being used for storing users' authentication data, thus it would be really useful to be able to mock imports for emulation of its behavior. So I'm totally upvoting this feature request.
We're using an Auth0 package for our authentication, and it's very cumbersome to test authenticated states otherwise. It would be very much appreciated if this feature cannot be implemented, an alternative provided in the docs for something to replace this pattern.
+1 Any workarounds for this? Trying to stub a simple query param used in the component test here.
+1
I use Playwright to unit test the rules of Firebase Database and Firestore.
Difficult to use Playwright for serious testing without this feature. Only being able to mock external APIs instead of modules forces us to mock implementation details for SDKs whose API calls we cannot make during testing.
I can't run my e2e tests using Nextjs 14 with the directive "server only" as it fails if any of my services include it.
in Vitest i'm able to mock the module
vi.mock("server-only", () => {
return {
// mock server-only module
};
});
This feature would be really useful.
+1 This is very important to avoid mocking a lot of requests. We want to mock auth0 for authentication.
+1 Any workarounds for this? Trying to stub a simple query param used in the component test here.
You can wrap your hole app with MSW before starting the E2E process.
We do not prioritize this feature since we focus on e2e tests. Let's wait and collect some upvotes.
Jun 2, 2022
This is a necessary feature for a testing suite. I'm around 6 hours into working with playwright and already have a use-case where this would be incredibly useful. In fact, it's pretty much a blocker.
Even if you're doing e2e, there are certain functions that you need to overwrite to do e2e effectively.
As a possible solution/alternative, there is SafeTest:
- https://netflixtechblog.com/introducing-safetest-a-novel-approach-to-front-end-testing-37f9f88c152d
- https://www.npmjs.com/package/safetest
It combines Playwright + Jest/Vitest as a usable out-of-the-box unified testing suite, supporting mocks.
Probably switching to Cypress - https://www.cypress.io. They have most of playwright's features with mocking included. They even have out of the box mocking for things like system clock, etc.
It's a shame this isn't a priority.
