A discussion about writing tests for frontend applications.

Context: my frontend coworkers don't write tests, at all. Yeah, really. Our testing process is very manual. We test manually when developing. We test manually when reviewing code. After merging, the application is deployed to a staging server and the design team does a QA Sprint. Lots of manual testing and some bugs still crawl by.

So I decided to start pushing my coworkers to start writing tests. One of the reasons I constantly hear them say to not write tests in the frontend is: "It's not worth the time, because design keeps changing, which means we have to take time to fix the tests. Time that we usually don't have."

I've been thinking about this a lot and it seems to me that this is more related to bad tests than to tests in general.

Tests should not break with design changes (small changes at least). They should test funcionality, not how things look. A form should not break if the submit button's style changes, so why should its tests fail? I also think that tests help save time, as they prevent some back and forth because of bugs.

Writing good tests is the hard part. Tests that cover what's really important and aren't frail and break with things that shouldn't break them. What (and how) should we test? And what shouldn't be tested?

Writing them fast is another hard thing. Are you doing it right if they take more time to write than the actual code?

What do you think about this? Do you write tests for your frontend applications? What do you test? How much time do you spend writing tests? What are your testing tools/frameworks?

Fuck. My boss left for a meeting with the client to show him the new frontend. 5 minutes after he left I saw that on the staging server half of the page entries were gone for the non-default localizations. FUCK.

It took me 20 minutes to restore the fuckery. I hope the client didn't see it...

Finally made my node production server stable enough that I could focus on writing tests*. I start by setting up docker, mocking cognito, preparing the database and everything. Reading up on Node test suites and following a short tut to set up my first unit test. Didn't go smoothly, but it's local and there are no deadlines so who cares. 4 days later, first assert.equal(1+1, 2) passes and I'm happy.

I start writing all sorts of tests, installing everything required into "devDependancies," and getting the joy of having some tests pass on first try with all asserts set up, feels good!

I decide to make a small update to production, so I add a test, run and see it fail, implement the feature, re-run and, it passes!

I push the feature to develop, test it, and it works as intended. Merge that to master and subsequently to one of my ec2 production servers**, and lo and behold, production server is on a bootloop claiming it "Cannot find module `graphql`". But how? I didn't change any production dependencies, and my package lock json is committed so wth?

I google the issue, but can't find anything relevant. The only thing that I could guess was that some dependencies (including graphql) were referenced*** in both, prod and dev, and were omitted when installed on a prod NODE_ENV, but googling that specific issue yielded no results, and I would have thought npm would be clever enough to see that and would always install those dependencies (spoiler: it didn't for me).

With reduced production capacity (having one server down) I decided to npm uninstall all dev dependencies anyway and see what happens. Aaaaand it works.....

So now I have a working production server, but broken local tests, and I'm not sure why npm is behaving like this...

* Yes I see the irony.
** No staging because $$$, also this is a personal project.
*** I am not directly referencing the same thing twice, it's probably a subdependency somewhere.

When the DBA restores the live db to staging and all your connections fail on passwords. Every fucking day!