Hacker News Reader

Comment by 0xbadcafebee | Hacker News Reader

Software has a "built on top of" problem. When you're doing construction, you can either re-use an existing building and all its existing components, or upgrade a few components inside the building, or you can knock it down and build from scratch. Software doesn't do this. Instead it takes the lead pipes, CAT-3 wires, horsehair insulation, and single-glazed windows, and installs them in a new skyscraper. We keep reusing and building on top of shitty old tech from decades ago because it's easier than inventing something better (PVC, CAT-6, blown insulation, triple-glazing), or we're avoiding making some old standard obsolete because some old buildings still use it and the landlords don't wanna pay for upgrades. So we get a working skyscraper, but for some reason we still need to train new engineers on horsehair insulation.

View on HN

People try to build all new, better software tech all the time! But doing so breaks backwards compatibility with existing stuff, which greatly hinders adoption. So it mostly remains relegated to experimental research projects, and if you're lucky, the best parts of it get incorporated into existing projects gradually over time.

Probably where the "new tech needs to be several times as good as the old stuff if it want to replace it" to overcome the momentum and resistance to change.

I love how most of the comments here seem to be from people who don't think there is anything valuable to learn from the previous 60+ years of computer evolution.

What you're missing is that the reason WHY lots of this "shitty old tech" is still around is because it has stood the test of time. Nobody keeps all of this around in modern technology out of pure nostalgia. This stuff is still useful because it largely solves a problem elegantly, and/or provides a useful abstraction with more modern systems. Or because there's simply no escaping the need for compatibility with existing systems.

Yes this "shitty old tech" has warts. ALL of your code will in a few decades too.

What was that XKCD about standards again?

Id argue a counterpoint to this, and its that replacing things like software are expensive, and time consuming. Take for example a CNC machine. The whole hog is a pricey piece, and you cant swap it out to some competitor without a lot of work, usually (in this case) in the way of all the program files that have to be retranslated, operators retrained, etc. AND THEN you add to that the software itself. Some of these machines have a "more modern" (read: at least from 1990) computer. Which while great in some aspects, the software still cannot be upgraded, due to compatibility. And so, these machines continue to put thousands of operating hours on the clock. The thing is, the software is usually pretty bad. Operators have just gotten used to the bad, and work around it. And its only gotten worse as time goes on. Some ancient windows os, needing to connect to a newer version as the rest of the business modernized. And worse, that ancient os, still having some network connectivity in some cases.

Its not that the tech is great, its that its become so engrained into the business that replacing it would be harder than starting a new business with the new tech. And so, it continues to operate well beyond its lifetime, and thats how you end up having cobol and fortran being actively used in industries in 2023.

And.. lets talk about the modern standards *cough* USB, but wait did i mean USB 2.0, or maybe 3.1, but maybe i meant 3.2+superspeed, but wait theres also USB 4.1! Old standards havent been entirely great, and that's why we have iterations (my usb example was really a rant more than example), but we have unfortunately built a lot of bad standards, that we still very heavily rely on, and wont easily get away from. The overhead of TCP for example, or how about IPv4?

I blame business though, we have much higher velocities in modern development than we did 50 years ago. Having years to plan a thing vs two weeks it seems to be today. If given a good time frame, software and standards can be more concretely designed, leading to less shitty new tech, but im a dreamer and it will never happen except in small scale areas.

Reply on news.ycombinator.com