Hacker News Reader

Comment by amatecha | Hacker News Reader

As someone who's been programming for years, one of my biggest complaints in software is over-engineering. It's so universally prevalent, I literally expect to get downvoted for even making this comment. I strongly advocate for simplicity in software implementation, allowing for maintainability and flexibility... the cleanest solution that reaches the goal and doesn't lock us too far into a narrow approach that may be difficult to deviate from. Sometimes that's not realistic, but it's actually feasible a lot more than people seem to believe.

View on HN

It's a weird term, "over engineering". Think about over-engineering a bridge. What does it look like? Huge and bulky, way too expensive, able to hold much more weight than it should, lots of walkways and access points and design features that nobody really wants? But wait: "any fool can build a bridge that doesn't fall down. It takes an engineer to build a bridge that just barely doesn't fall down." The whole point of engineering is to figure out exactly what the bridge needs to do and make it do only that. So the "over-engineered" bridge is actually under-engineered -- they haven't thought enough about the actual problem. So too is most code that is called "over-engineered" -- it's far more common that people haven't though enough about the problem than that they've over-thought it.

I like to call it "poor engineering" and there are many forms of poor engineering I've seen. Following the analogy... painting the bottom of the bridge in same color as the terrain. Then repainting because there were color differences.

Spending time and money looking for lighter materials even when the bridge is designed to support a lot more weight than traditional time tested materials. Justification : "Maybe cars will get heavier, and bigger saving 3 pounds is a good pursuit"

I think that in a way that view is like the entry level engineer's lie. It makes it seem like the engineer is there to make shoddy things, always a hair's breadth from failing.

What an engineer is actually there for is to run the numbers and predict behavior without having to go through the mess or expense of doing it any more than absolutely necessary. To plum the depths of our collective technical knowhow to drive a project that makes all the right tradeoffs.

It's not about making a bridge that can hold a semi out of matchsticks and bubblegum. It's about knowing it won't work without having to do it to find out.

When I say "over-engineering" I mean stuff like this: https://github.com/MarkSFrancis/enterprise-fizz-buzz Building a whole ton of stuff, just to do something fairly trivial.

In this example, it's black and white obviously over engineered. The reality is, in an application of moderate complexity, whether or not it is indeed over-engineered is a far more subjective calculation.

E.g., >90% of engineers would agree enterprise-fizz-buzz is over-engineered.

But perhaps ~50% of engineers agree my SaaS app is over-engineered.

Just to be very clear, that application is intended as a joke. (just in case)

> he whole point of engineering is to figure out exactly what the bridge needs to do and make it do only that. So the "over-engineered" bridge is actually under-engineered -- they haven't thought enough about the actual problem. So too is most code that is called "over-engineered" -- it's far more common that people haven't though enough about the problem than that they've over-thought it.

It is different in the physical world where materials are a big factor and engineering will definitely not make free use of them. There are inefficiencies here and there but they're largely reduced. In the digital world of programming accidental and intentional complexity often slips through without anybody noticing and that is for many reasons. It's a relatively new field and is still quite inefficient. I worked in many places whose codebases are a giant maze designed with no clear architecture and sometimes I suspect this complicated mess of intentional moat building engineering.

But people will hate your suckless bridge and instead will use a more feature rich bridge with more bells and whistles. https://stroustrup.com/P0977-remember-the-vasa.pdf

I like to believe that experience will eventually drive people towards using scientific parsimony when creating solutions to engineering problems.

I don't think over-engineering is the only problem: shiny-new. Shiny-new syndrome wherein the typically junior programmer cannot help but jump from one new thing to the next, advocating that some new system must use the new tech that they recently discovered. While the enthusiasm is ok, working with people like this can be a drag.

Another problem: we are facebook too! In the sense that small companies heartily believe that they actually have problems at the scale of a company like facebook (or will ever get to that level). Again, hearing someone say, "We should do xyz because Netflix does it!" makes me want to give up on this profession.

> Another problem: we are facebook too! In the sense that small companies heartily believe that they actually have problems at the scale of a company like facebook (or will ever get to that level).

Precisely! It's tiresome how many people setup Kubernetes for a network of 2-6 computers.

I was with you once, but last year's I got tired when people set up networks of 2-6 computers with daemons that run in a wildly different ways, and it's not clear how to fix their failures, where configs and metrics and logs are. I'd honestly prefer overengineered kubernetes setup nkw, only for some more standard way of doing things between projects.

That's a false dichotomy though. Having a standard way of doing things between projects does not require running a complex distributed system with dozens of pieces all engaged in constant chatter. It's a testament to the failings of our profession that we apparently cannot agree on a sane baseline toolset less complicated than Kubernetes.

Infrastructure as code is much easier to handle than full backups of the configuration of your mutable machines, even for 1 machine.

The only problem it has (and why I don't use stuff like kubernetes at home) is that the configuration formats aren't stable and thus require a lot of ongoing maintenance.

"over-engineering" is a fuzzy term though. For someone, using a third-party service to handle users auth flow is a time saver, and building the whole flow themselves would be over-engineered. For another, using a third-party service for auth would be adding layers of complexity on top of a simple db request, and would thus be considered over-engineered.

Largely agree, but I've found distilling complex requirements down to a simple implementation takes effort, and sometimes longer than the more circuitous route.

A coarse rule of thumb from a few decades of my own experience in software development is that version 1 will be quick, dirty and (in retrospect) naive. Version 2 will improve it with what you learned the first time around. And version 3 tends to be the clean, solid, polished winner. I've learned to expect and encourage refactoring to arrive at this point.

"I would have written a shorter letter, but I did not have the time."

In my experience, you often start with a somewhat over engineered solution. Then you take the (long) process of really understanding it and slowly distill it down. This is true for a new feature and also the whole codebase.

Yeah, that's a great point. The final code I submit is often 2/3 the size of what I initially got running (perhaps less, even). I'm always happy to see a code review that is removing more code than it's adding, haha :) I like to use the phrase "measure twice, cut once" in regards to shipping code. Take the time to pare it down to the truly essential functionality, trimming away extraneous stuff that complicates things and may even be premature optimization. It's certainly a skill that takes time to improve at, haha

Wholeheartedly agree. The best system is no system at all - problem is it doesn't really satisfy any requirements. Anything added beyond that is extra room where fragility creeps in & the tradeoff shouldn't ever be taken lightly.

Reminds me this comment: https://news.ycombinator.com/item?id=19977678

There are other coping mechanisms you can use, although some of them are arguably worse than procrastination (eg overengineering).

In my opinion over-engineering is a symptom of not enough thinking. Complex, over engineered solutions don't happen because a problem is too well understood, they happen because a problem is not understood well enough.

This is based on the assumption that more thinking == better understanding.

Reply on news.ycombinator.com