Skip to main content

Things Fall Apart

How bad would it be if your code broke? That’s a question I wish we asked ourselves more. Sometimes the answer is really really bad. If you write code for airplanes or medical devices, failure is quite literally fatal. Or maybe you’re writing software that trades billions of dollars of financial instruments. Failure is bad there, but maybe not as bad as in the previous group.

However, for the majority of us, the answer is not that dramatic. Maybe someone’s website won’t work for a bit. Maybe a dashboard will be out of sync. Maybe you lose a few users. Yes, it’s bad but it’s not bad.

Why is this a question worth asking? Because the potential downsides of code breaking dictate your strategy for software development. It dictates how often you release, how much testing you do, what you tell your users, and so on. It dictates your tolerance for risk.

I remember discussing with a former colleague about shipping a developer tool. They hemmed and hawwed, and eventually confessed that they thought the tool was not ready for production. I was honestly pretty baffled. I asked them, “what does production mean here?” Because you see, our tool was not in any shape or form in the direct path for a user’s product. It would not affect the correctness of their code. The worst it would do is create some small amount of inconvenience. It was not running on a server. Nevermind that we had extensive testing and quite good code review.

The issue here was twofold. The first is that errors are more damning if you don’t believe in your product’s inherent value. It’s like that old rule of dating: You’ll know if they’re interested. If someone is interested in your product, then you can have a couple bugs here and there and they’ll still use it. If the condition for the product is perfect execution, well, that’s an impossible condition. It’s like someone saying they’ll date you if you behave perfectly. Maybe they just don’t want to date you.

The second issue is pride. As software developers, we have a certain pride in our code. Our code is correct, not like those other developers. It’s why not-invented-here syndrome exists. It’s why developers reinvent the wheel. We believe that we can do better. As such, it’s not so nice to admit that you can’t write code that’s always better than other people’s code. It’s not so nice to admit that sometimes you mess up.

But that is the reality of programming. By shipping your code, you are opening yourself up to bugs, to ridicule, to people questioning your intelligence because how could you have missed such a simple bug?! Too bad. You will mess up. Projects with far more at stake have messed up. Bridges have collapsed; space shuttles have exploded; planes have crashed. Failure is how we learn.

You should not avoid failure at all costs. There is always too high a cost. Any sane manager understands this. If you asked them whether they would delay their release for a year if they could guarantee almost zero bugs, who would do it? What about two years? Ten? Never? Failure is something to be weighed and balanced. This is why we don’t build buildings with two feet thick walls out of poured concrete. There is a cost to infallibility.

I wish we discussed this more in my CS major. In the rare case study we’d study something like the Therac-25, which, while a good example of how concurrent programming can be tricky, is also a total strawman. I’ll be sure to review it when I start writing code for radiation therapy machines, but until then, I think the stakes are a little less dire.

And sometimes it’s okay to willingly create bugs for your users. I love writing Rust because it’s a wonderful, expressive, fast, well-designed language. I also love it because it’s totally obsessive about getting your code right. It will nitpick you on details like strings must be valid UTF-8 while filesystem paths do not have to be valid UTF-8. That can save your bacon. It can also send you into a spiral of obsessing about whether your users pass you a non-UTF-8 path and what should you do? And oh no, maybe we should figure out a way to rewrite the path and inform the user or maybe we could just pass bytes and then only convert to string when necessary, and so on. I love writing Rust, but on occasion I have to remind myself that it’s okay to just throw an error and tell the user to use a UTF-8 path. Sometimes it’s okay if your code doesn’t catch all of those edge cases.

Not to get all philosophical, but we’ll all be dead someday. It’s not worth wasting your time trying to ship perfect code.