Your Ad could be here. I want to connect my readers to relavant ads. If you have a product targeted at developers, let's talk. [email protected]
Theo's response puts a lot of my feelings about unit testing into words. Many of us have grown up in this world preaching unit testing. We often hear these statements "Everything must be unit tested, tests make code more maintainable." In reality when we are not writing complex low level code unit tests are probably the wrong approach.
So much of software engineering is assembling existing well tested code. Crud applications, UI, Data Pipelines, building on top of battle tested code.
This kind of reminds me of Manufacturing. Individual components are QA tested with tests that look more like unit test. Parts like bearings, pistons, shafts, valves, they are all tested against sophisticated statistics of sample measurements. This is quite similar to unit testing.
You see measuring the individual sizes of these components does not guarantee the actual function of the component. Before the Henry Ford assembly line parts like this were all different sizes, and they were all custom matched to each other. The one industry that I can thing of that still exists today like this is gunsmithing. parts are made very close to size, but they need some lapping and final finishing for high quality function.
So how are pumps, engines, valve assemblies, and so on tested? They are integration testsed based on their function. They are not generally unit tested to measurements of thier size. A pump is tested on a test bench and will be required to output a spec flow, a valve will have air or fluid applied and tested for leaks. An engine is a more complicated case where it will be ran on a suite of cycles and should output a spec level of power, while emissions, temperature, fuel consumption and hundreds of other measurements stay within their specs.
There is 100% a place for unit testing in the software engineering industry, just like manufacturing, but I don't think it makes sense everywhere. Library development and lower level algorithms look a lot more like parts that can be physically measured. But that's probably not the code most of us are writing most of the time.
We are likely assembling a lot of code to build a product such as a data pipeline, or a web application for instance. For these cases we should be thinking about the final function of the product. Like Prime says in the video, think of the product as a black box, as if you have no idea how anything works in the box, but you know how it should function.
I'm starting to lean more into simple tools for this qa work on we applications that I build. They are easy to build and maintain, and do not get in the way when the direction of the project changes. If I rip out half my code it would be likely to see 3/4's of my unit tests failing. Not a fun thing to maintain.
This is the simplest form of integration testing that I am pulling into fokais, and it's nearly fully automatable without much thought on my behalf. I am building web applications in fastapi, these api's have endpoints that should be reachable. Tools like Uptime Kuma make it trivial to pull in all of my endpoints and simply ask if they are reachable once per minute.
With uptime tests I can more confidently move through dev to prod. If dev has been running for a number of minutes and alarms are not going off, it's probably functioning.
I've used tools built upon selenium in the past, such as testproject.io. It is a pretty great experience, and makes it pretty easy to define, when a user logs into the dev app, and clicks the dashboard, can they see their name and account information.
These tests are likely to remain true for the life of the project. No matter what changes inside the black box, when a user goes to their account they should see their account information. It might be server rendered with jinja, fetch with htmx, or client side rendered with reactjs, it does not matter.
The last thing I want to really put as a tool in my toolbelt is load testing. It's pretty easy to degrade the experience of a webpage with a bad database query or api call. Load testing can ensure that your application can handle the load that you want it to in dev before you release to prod. I am looking into bringing in Locust.io for load testing fokais.com.
Inevitably users will use your product in ways that were never intended. It will be nearly impossible to cover everything with any sort of test. This is where logging and monitoring come in. If you can set up error monitoring and alerts, then you can check production logs to see what the heck happend and potentially resolve the issue. Without monitoring, you will have wild bugs that you never thought could exist.
No tests will never be perfect, they will not catch everything. The point that I was inspired to make after watching Theo's video is that unit tests may not the right thing for your product, and then requiring coverage metrics on PR may be more of a detriment to the product than help.