Hacker News Reader

Comment by nikolay | Hacker News Reader

I've never written unit tests in my life, and my code works most of the time. There used to be times where I'd write code for an entire week in Java without compiling (because it was taking forever) and back then we used pretty basic IDEs that didn't help as much as they to today to help you prevent stupid errors. And, usually, my code compiled just fine and worked. Today's developer is a trial-and-error one, tweaking a few lines of code, refreshing the browser, and seeing if stuff works or not. Today's developer spends disproportional amounts of time writing unit tests and, yet, producing buggy code.

View on HN

I've never written unit tests in my life, and my code works most of the time.

How do you know?

I agree with your point about taking more time to think about the way your code works, to design data structures, to use rigorous methods to make sure the code works, but ultimately you still have to check things. If you haven't automated those checks then either you're checking things manually or you're not checking them. On any significant project there's too much code to check manually.

How do I know? No bugs were reported.

People take % of code coverage as some insurance and spend time writing stupid tests instead of writing robust code.

People also don't think that unit tests are constant maintenance burden and a time waster - you change this and that and then you need to spend time entertaining the tests.

Again, use your brain, not your fingers - it's that easy! Oh, and don't program using a dynamic language saving keystrokes, but wasting, even more, writing brainless unit tests!

At least in my personal career, most, though, and real issues come from race conditions. But I've been fortunate to catch them quickly by using my brain. And that's what we lack today - developers who use their brains, who don't spend too much time on emoji reactions in Slack and finding the right GIF.

There's plenty lots of code monkeys and code gluers today that can't even implement a bubble sort without googling it!

People also don't think that unit tests are constant maintenance and time waster - you change this and then, you need to spend time tweaking tests.

That's why you test interfaces, that you've spent time designing so they don't change often. You, or someone else who isn't as good as you, can change the internals of a block without updating the tests.

Given the choice of working with a dev who knows how to write a bubble sort[1] and a dev doesn't but does write tests, I'd go for the latter.

[1] To be honest, if someone in my team even considered writing their own sort function, from memory or by googling, I'd have a quiet word. Use a library. Preferably a well-tested one.

This is not about memorizing useless stuff, but about knowing the foundation of Computer Science. Our brains are full of emojis today, not algorithms. It's been scientifically proven that the Internet is making us more stupid by tweaking our brains' modus operandi to offload memory to Google and StackOverflow. We think memorizing has just a storage function, but it doesn't - when knowledge is planted in our brains, it just doesn't sit passively in there, but its accumulation improves our thought process by using things like generalizations, drawing analogies, etc. So, again, this is not about being dependent on the Internet as much as it is about growing better brains. I don't think we're smarter than people 100 years ago - we just have more tools to compensate for our shrinking brains.

By the way, I also test my code - just not the way the newly-bred "ninjas" do it - pair program, peer review, and all this fancy useless stuff! By the way, I have over 30 years of software development (not coding!) experience. I've attended many programming contests (not hackathons) since childhood and won most of them. You don't hear about such contests anymore - gluing up a quick POC together is the big deal today, and this is just pathetic. And no wonder we have so much POC-quality code in Production (with a capital "P") today! And a craziness like CD to Production is considered the cool new trend - Production is not sacred anymore, and everybody can dishonor it as they wish!

Again, I'm talking about the large gray mass of developers; there are still a lot of true developers alive today, but they are becoming extinct, unfortunately!

How do you test your code?

Honestly? Mainly, reviewing it and debugging it in my mind. It works - you should try it! It makes you a smarter and a better developer as well. The try-refresh-repeat cycle is for amateurs.

But that comes with the way I write my code - modular, i.e. broken into small, readable and reusable units. Well, if I'm doing a something that requires shaving off any redundant CPU cycle, that's another story, but as we know, premature optimization is the root of all evil. You won't see anywhere in my code long functions/methods, endlessly nested IF-THENs, and similar junk - writing simple, unfancy, DRY code drops the need for unit tests to almost zero. Using long, but meaningful identifiers drop the need for inline documentation.

I mean, I don't have to repeat things that have been known for ages - just buy some old books, it's all in there!

There are way too many coders today and very few engineers!

I mean, I don't have to repeat things that have been known for ages - just buy some old books, it's all in there!

I don't really understand this point. UnitTest and SUnit, that all subsequent unit testing frameworks are based on, were released in about 1990. "Extreme Programming" was a big thing in the mid-90s and that featured test frameworks as a core component. Automated testing isn't new. There are plenty of old books that advocate exactly the "new and trendy" processes you decry. If your argument is "Write code like engineers did 20+ years ago!", then that includes automated testing.

I strongly suspect that we're not actually that far apart in our thinking. I don't believe that unit tests magically make your code better. You can't develop a solution to a problem you don't fully understand by throwing more and more tests at it. Developers still need to take the time to think. I believe a lot of my enthusiasm for testing comes from the belief (...experience...) that there are plenty of developers out there who, as you put it it, are coders rather than engineers. My tests pick up their mistakes (and mine, because sometimes I'm not at my best). Maybe, if you're very fortunate, you've managed to surround yourself with a team who are all really good engineers who genuinely don't need to test things because they're rigorous to the point of infallibility. I haven't, and I work in an industry that means I probably never will, so tests are necessary.

Advocating testing is much, much more productive, and will lead to better software in the long run, than advocating developers write better code.

Just because unit testing existed in the 90s, it doesn't mean it was widely (ab)used as it is now! Many technologies were invented well before they gained adoption.

I didn't mean unit testing is new or useless - just that it's being abused ignoring the benefit-cost ratio and negatively affecting you thought process to rely on something unreliable instead on focusing on code quality - tests are no panacea as code today is more tested than ever, but is also buggier than ever as well.

As a summary: Just think more when you write code, don't get into the vicious circle of try-refresh-repeat, and having tests in place is no excuse for poor code.

Oh, I forgot to mention the ever-shrinking attention span. Our lives become fuller and fuller of distractions and interruptions. With the increasing load of notifications, alerts, and news feed updates, we can't really concentrate and grasp hard concepts. And we, the people involved in startups and being behind these distractions, are eating our own dogfood and suffering from the same evil.

> Today's developer spends disproportional amounts of time writing unit tests and, yet, producing buggy code.

Unit testing make you a lot more efficient. You can choose whether to spend that efficiency producing less buggy code in the same time with the same functionality, equally buggy code quicker with the same functionality, or equally buggy code in the same time with more functionality. Most of the time businesses quite reasonably decide the second or third option is more valuable.

Are you suggesting that an increase in unit tests is correlated with a decline in code quality? That would be a surprising correlation.

I do think sometimes about TDD as Think less driven development. The problem is not that much in TDD itself, but how its usually implemented - as a safety net and excuse for producing all kinds of bad code.

Unit tests do not correlate to quality, and even at 100% coverage only guarantee that you've looked at every line of code twice (usually). Not only that, a lot of projects skip adding testing until well after a proof of concept is working... why? because we don't want to not ship features for two weeks while we backport in tests... Now you have less than optimal code and less than optimal tests.

This persists, every suggestion to refactor structure rebuffed, and this persists until the code debt is so high, it takes 4-5x as long to add a feature as it should. Where, if you're lucky, you can make a new version using only 3-5 year old technology, instead of the decade old stuff you've been supporting for half a decade. Even then the 2-5yo stuff isn't as nice as it was sold, but it's "enforced" by the company powers that be (god I hate angular sometimes).

While I wouldn't suggest that, I will say spending more time manipulating my code to work with the automated unit testing tool of the month has definitely lead to an occasional decrease in code reliability. Not all code can be unit tested [with every unit testing library], and being compelled to shape everything such that it works with all once and future automated tools often makes the mess worse.

That's what happens when you rely too much on something unreliable.

That's what I have come to call FDD, Feedback Driven Development. Try something... Google/Stackoverflow it... maybe find a tangential solution to the problem at hand, fit it in the context of the problem, try something... ad nauseum

Never also, don't understand at all. Bugs usually visual and business process. If wanted to type all business process to unit test not sure client willingness to paid the time.

Unit tests are no silver bullet. This trend was started by consultancies that found more ways to milk customers - write unit tests and do inline code (!) documentation. Both are a real cash cow when the client doesn't know what they want and you need to constantly tweak code and respectively tests and documentation! I understand API documentation, but internal code that's obvious like what "width" and "height" parameters of a method are for is beyond stupid!

Reply on news.ycombinator.com