Ask HN: Is average code getting worse?

#1

Post by Aeolun » Mon, Mar 01, 2021, 8:48 AM UTC

Hi HN,

There used to be a time, in the dark dark ages of history, 10 years ago or so, when I would encounter issues during the course of my work, and I could fairly confidently assume I was doing something wrong, or I just hadn’t read the manual correctly.

Contrast that to now, when I regularly write code, or include a library to do a certain thing, and I find it just does not work. Following my historical logic, I spent a day trying to figure out what I did wrong, only to figure out that it was a bug, or edge case that someone just hadn’t covered (never mind that my life seems to consist of edge cases now) or the documentation is just plain out of date (or nonexistent).

Is this a trend? And does it have to do with the average code, or myself? Have you experienced something similar? It’s driving me nuts.

I want to rely on other code, but more and more I find that it’s safer to just assume it’ll be broken from the start.

Re: Ask HN: Is average code getting worse?

#2

Post by jpgvm » Mon, Mar 01, 2021, 9:53 AM UTC

I find this varies by ecosystem. Certain ecosystems have incredibly high quality libraries on average. However ecosystems also vary in their breadth and quantity of available libraries also so that may play into it.

I find Rust and Perl to both have very high quality libraries available. JVM/Java ecosystem has a huge number of libraries and some extremely high quality ones but also a massive number of abandoned or low quality libraries left to decades past.

The ecosystems I find troublesome are JS, Python and to a much lesser degree Ruby. Huge numbers of low quality libraries and dependency hell.

The differences seem to stem from the culture of said ecosystems. Java code is often flexible to a fault, Rust code is strictly correct and fast but with less emphasis on flexibility/ease of use. JS/Python hail from the "get shit done" culture and as such there are many half-baked libraries.

These are just generalisations. Even the ecosystems I don't enjoy working with have some very high quality libraries of course. JS has React, Python has Tensorflow, Requests, etc.

If you want to be able to rely on library code though I would go for Java/C#/Rust maybe Go if it doesn't annoy you for other reasons.

Re: Ask HN: Is average code getting worse?

#3

Post by snicker7 » Mon, Mar 01, 2021, 9:53 AM UTC

Have you vetted the vendors of the library? Is it something that a reputable company offers that you /payed/ for? Or is it some side-project of some rando?

Re: Ask HN: Is average code getting worse?

#4

Post by dfn43589 » Mon, Mar 01, 2021, 9:54 AM UTC

Where are you getting these libraries compared to 10 years ago? What languages do you use?

Re: Ask HN: Is average code getting worse?

#5

Post by krageon » Mon, Mar 01, 2021, 10:02 AM UTC

snicker7 wrote: post #3

Have you vetted the vendors of the library? Is it something that a reputable company offers that you /payed/ for? Or is it some side-project of some rando?

These metrics don't matter. Almost every library you care to pull for a popular (read: widely used and written) language will be trash. Yes, even if you pay for it. Even if it was a "reputable" company making it.

Re: Ask HN: Is average code getting worse?

#6

Post by lhnz » Mon, Mar 01, 2021, 10:02 AM UTC

I generally assume code is broken from the start now, but I don't think that this is because average code is getting worse. I think it's because there are more "solutions" to problems online (also, I'm more critical than I was a decade ago).

In the past, I had to write almost everything from scratch, and that meant I'd have worse code than what is open-sourced nowadays for the time period that I was implementing (and possibly longer if the problem is gnarly).

Nowadays, I can do a search on GitHub and find something that appears to solve my problem, but when I use it it'll turn out the author was solving a slightly different problem than mine and so his solution only partially solves my problem or doesn't handle certain edge-cases/bugs that will be a problem for me.

That doesn't mean that the code is worse. I am just using something that was created to solve another person's problem and expecting it to solve my own problems with no extra work.

Basically, you can still choose to implement everything from scratch if you want to. In these situations you should make sure to learn from the people that came before you, by reading other people's code first and spending time thinking about your design/implementation.

But if you want to use other people's code off-the-shelf to implement something faster, you can't just assume that because their solution solved their problem without running into edge-cases, that it'll do the same for yours. You have to be responsible for every line of code that you integrate with, and should assume that is not guaranteed to work for your problem/data.

---

What is an issue and wasn't before, are problems that occur due to a particular combination of packages and incompatabilities. These problems are difficult because they can require impossible levels of co-ordination to fix upstream. I think in many situations we should rewrite things to be more monolithic than they are.

Re: Ask HN: Is average code getting worse?

#7

Post by krageon » Mon, Mar 01, 2021, 10:03 AM UTC

In my experience yes. Average code is much, much worse than it used to be. Regardless of the origin. Then again, there's also much more of it. It could be argued to be a mixed blessing.

Re: Ask HN: Is average code getting worse?

#8

Post by turing_complete » Mon, Mar 01, 2021, 10:07 AM UTC

This talk deals with the issue you are describing as well as the underlying reason for it: "Preventing the Collapse of Civilization" by Jonathan Blow https://youtu.be/pW-SOdj4Kkk

It's very informative and engaging. Highly recommended.

Re: Ask HN: Is average code getting worse?

#9

Post by samhuk » Mon, Mar 01, 2021, 10:08 AM UTC

Preface: I've worked on code ranging between FORTRAN 77 to a rather nuts react, typescript, webpack,... combination.

From my experiences, what I have observered first and foremost is that the problems that software is solving are getting more complex as time progresses, causing software complexity. That is often proportional to the number of bugs.

Many popular FORTRAN packages out there solve a particular (or set of) well(ish)-documented mathematical or physical problems. Matrix manipulation, eigenvalues, weather simulation, etc. What i've observed is that these problems are, although sometimes complex, require minimal dependencies. The primary dependency is some journal paper somewhere, which you don't have to install as a software dependency!

Constrast that to a couple recent codebases i've worked on, easily >300MB of npm dependencies for a mix of react, typescript, webpack, sass (gyp!! >:(), jest, cypress, etc.

Having said all of that, for both of the above eras of software, i've encountered annoying bugs. It's just that nowadays i see most bugs are package compatibility issues (looking at you, steaming pile of webpack 5), whereas with FORTRAN it has instead been a mathematical error.

My point is that software is solving very complex problems right now. Back then, a web application was 3 types of files, because web applications were simpler. But now entire businesses use them, all their processes can rely on a single web app. Important processes sometimes with regulatory involvement. This necessitated typescript (scalable js), testing (jest and cypress), better performance (webpack), etc....

Re: Ask HN: Is average code getting worse?

#10

Post by stared » Mon, Mar 01, 2021, 10:09 AM UTC

It is similar to "are teenagers worse than they used to be?".

But in all fairness - from my personal experience - the average code is getting better. It is mostly thanks to tooling: languages (write a readable code in PHP vs Python, or C vs Rust), version control (no need to keep tons of unused code that one does not want to throw away), linters getting more widespread, tests and CI becoming a standard.