At my current employer, Rescale, we were using the Jest framework by Facebook to test our React and Flux application. Jest is a simple-to-get-started testing framework with an API similar to that of Jasmine, but with automatic mocking of CommonJS modules baked in. However, like many developers have noted, we’ve found Jest to be unbearably slow when running a non-trivial test suite. For our test suite of about 60 test cases, it takes well over 10 seconds to finish! This article explains how we’ve ditched Jest in favor of Karma, Webpack, and Jasmine. The same test suite running under our new setup takes only a little under 200ms to execute in PhantomJS, after the initial Webpack bundling.
Our testing setup is based on the one explained in this article: Testing ReactJS Components with Karma and Webpack. Read that if you are unfamiliar with Karma and Webpack. To summarize, this setup uses karma-webpack to bundle all of our tests into a single file which Karma loads and runs in the browser. The npm packages needed for this are:
- karma
- karma-cli
- karma-jasmine
- karma-phantomjs-launcher
- karma-webpack
- jasmine
- webpack
- core-js (optional, for ES5 shims for PhantomJS)
- babel-loader (optional, for ES6 to ES5 transpilation)
The Karma configuration file, karma.conf.js
may look something like this:
module { config;};
And the entry point for the test suite Webpack bundle, tests.webpack.js
,
should look something like this:
// ES5 shims for Function.prototype.bind, Object.prototype.keys, etc.;// Replace ./src/js with the directory of your application code and// make sure the file name regexp matches your test files.var context = require;context;
And here’s a sample test, located at
./src/js/components/__tests__/MemberList-test.js
which tests the component
./src/js/components/MemberList.jsx
:
var React = ;var TestUtils = ;var MemberList = ; ;
That’s all well and good for testing React components, but testing Flux applications, specifically Flux Stores, requires a little more setup.
Testing Flux Stores
The problem with Flux stores is that stores are usually singletons with state, and singletons with state can quite possibly be one of the worst things to test because their state persists between tests. It becomes hard to predict the state of the stores when many parts of our application interact with them throughout the test suite. You might be thinking that we could just implement some method to reset the state of the store before each test case, but doing this for every store is a maintainability nightmare and is very prone to errors.
Jest solves this problem by running every test case in its own environment. In
otherwords, when we require
a module in a test case, the module imported is a
fresh instance. This is exactly the behavior we want when testing stores!
With Webpack, we can do just that: clear the require
cache before each test
case so that calls to require
load a fresh instance of a module. As an
optimization, we wouldn't want to remove third party modules, such as react
,
from the require
cache. Doing so would slow down our test suite significantly,
and none of those third party modules should be singletons with state anyway. We
can add cache busting before each test in the tests.webpack.js
file like so:
// Create a Webpack require context so we can dynamically require our// project's modules. Exclude test files in this context.var projectContext = require;// Extract the module ids that Webpack uses to track modules.var projectModuleIds = projectContext; ;
That’s all there is to busting the require
cache. Now modules are cached
within each test case, but aren’t in between. Stores, actions, and dispatchers
are all isolated between tests.
Testing the Actions/Stores Boundary
We use Reflux as the Flux implementation for one of our projects, and
Actions are pretty much the public API for Stores. Triggering an action is an
async operation though, so we need a way to control this in our tests. This
should be easy with any testing framework or library that provides facilities to
mock out the native setTimeout
and setInterval
functions and manually
advance them. Here is an example of testing the Action/Store boundary with
Jasmine:
;
Mocking a Module’s Dependencies
Say we want to test a module that has a dependency on another module that we
want to mock. One example could be that we want to test an api module that
depends on some http module like axios. This is where Jest shines because
they make it easy to specify mocks with its jest.setMock(moduleName, moduleExports)
API or with their automatic mocking facilities.
One way to achieve this outside of Jest is to use rewire and its webpack
version rewire-webpack. Rewire can be used to change private variables in
a module. For example if we want to mock out the axios
module within our
api.js
module, we can write something like this:
// api.jsvar axios = ; var API = { axios; }; moduleexports = API; // In some testvar API = ;API;
Now API.doSomething()
will call the mocked axios' get()
.
Another option is to, once again, manipulate Webpack’s require
cache:
; // Make sure the module is loaded and cachedrequirecacherequireexports = mockAxios;
The benefit of this over rewire is that every call to require('axios')
is
mocked, not just in the module that we rewired.
Note: Mocking an http request library is probably a bad example. For that you should use something like jasmine-ajax.
Closing
We saw around a 50x speed improvment in the execution of our React and Flux application test suite using Karma and Webpack over Jest. However, it does take a bit more knowledge and effor to set up compared to Jest, and mocking a module’s dependencies isn’t as easy. Jest is really nice since it’s easy to set up, but that it’s so slow is a deal breaker for us, and we dont know if it’ll ever be fixed. In the mean time, our current setup allows us to effectively TDD.
Edit: Sample repository demonstrating this technique, react-flux-testing.