How can I fail Karma tests if there are unhandled rejections? - karma-runner

Question

A couple of places propose this solution:
window.addEventListener('unhandledrejection', function(err) {
window.__karma__.error(err); // yeah private API ¯\_(ツ)_/¯
});
But it throws:
Uncaught TypeError: Cannot read property 'error' of undefined

Answer 1

I'm able to get reports of unhandled rejections with the following setup:
karma.conf.js:
module.exports = function(config) {
config.set({
basePath: '',
frameworks: ['mocha'],
files: [
'setup.js',
'test.js',
],
exclude: [],
preprocessors: {},
reporters: ['progress'],
port: 9876,
colors: true,
logLevel: config.LOG_INFO,
autoWatch: true,
browsers: ['Chrome'],
singleRun: false,
concurrency: Infinity
});
};
setup.js:
window.addEventListener('unhandledrejection', function(ev) {
window.__karma__.error("unhandled rejection: " + ev.reason.message);
});
test.js:
it("test 1", () => {
Promise.reject(new Error("Q"));
});
it("test 2", (done) => {
setTimeout(done, 1000);
});
Separating setup.js from test.js is not necessary. I just like to have such setup code separate from the tests proper.
When I run karma start --single-run I get:
25 01 2017 07:20:07.521:INFO [karma]: Karma v1.4.0 server started at http://0.0.0.0:9876/
25 01 2017 07:20:07.523:INFO [launcher]: Launching browser Chrome with unlimited concurrency
25 01 2017 07:20:07.528:INFO [launcher]: Starting browser Chrome
25 01 2017 07:20:08.071:INFO [Chrome 55.0.2883 (Linux 0.0.0)]: Connected on socket g-BGwMfQLsQM128IAAAA with id 22107710
Chrome 55.0.2883 (Linux 0.0.0) ERROR
unhandled rejection: Q
Chrome 55.0.2883 (Linux 0.0.0): Executed 1 of 2 ERROR (0.006 secs / 0.001 secs)
Caveat
Reports of unhandled rejections are asynchronous. This has a few consequences.
The example I gave has a 2nd test that takes 1 second to complete. This gives time to the browser to report the unhandled rejection in the 1st test. Without having this delay, Karma terminates without detecting the unhandled rejection.
Another issue is that an unhandled rejection caused by test X may be discovered while test X+1 is running. The runner's report may make it look like X+1 is the test the caused the issue.