Yeah, modern computers often feel like a scam. Obviously, some things are faster and obviously, we can calculate more complex problems.
But so often, programs are only optimized until they reach a level of “acceptable” pain. And especially with monopolistic, commercial software that level is close to infinity, because well, it’s acceptable so long as customers don’t switch to competitors.Either way, the slowness that was acceptable twenty years ago is generally still acceptable today, so you get much of the same slowness despite being on a beefier PC.
This isn’t a very good article IMHO. I think I agree (strongly) with what it’s trying to say, but as it’s written, it just isn’t it.
Wrappers and VMs and bytecodes and runtimes are bad: they make life easier but they are less efficient and make issues harder to troubleshoot.
Runtimes/“VMs” like the JVM also allow nice things like stack traces. I don’t know about the author but I much prefer looking at a stack trace over “segmentation fault (core dumped)”. Having a runtime opens new possibilities for concurrency and parallelism too.
The COSMIC desktop looks like GNOME, works like GNOME Shell, but it’s smaller and faster and more customisable because it’s native Rust code.
This just doesn’t make any sense. COSMIC is more configurable because it wants to be, this has absolutely nothing to do with Rust vs Javascript.
Dennis Ritchie and Ken Thompson knew this. That’s why Research Unix evolved into Plan 9, which puts way more stuff through the filesystem to remove whole types of API. Everything’s in a container all the time, the filesystem abstracts the network and the GUI and more.
And here the author just contradicts themselves. So wrappers, runtimes and VMs are bad, except when it’s Ken Thompson doing it in which case adding containers and a language runtime into a kernel is a great idea actually?
Lastly, I didn’t address the efficiency arguments in the quotes because it’s mostly just true… but I do think it requires some more careful consideration than “JS bad Rust good”. Consider this unscientific sample of different apps on my PC and how much of my (expensive!) RAM they use:
- Spotify (Electron): 1G
- Ghostty (Zig/GTK): 235M
- Decibels (Typescript/GTK): 140M
- Anyrun (Rust/GTK): 110M
Note that Electron, and only Electron, is a supermassive black hole of bloat. Whatever is going on here, it’s not Javascript.
Runtimes/“VMs” like the JVM also allow nice things like stack traces. I don’t know about the author but I much prefer looking at a stack trace over “segmentation fault (core dumped)”. Having a runtime opens new possibilities for concurrency and parallelism too.
Rust has stacktraces without needing a runtime. Don’t ask me what exactly is going on behind the scenes, but there is a way to request a stacktrace for a given point in the program. And unless you’re doing embedded stuff, a stacktrace is automatically generated for errors.
And as for concurrency/parallelism, it’s correct what you wrote, but I just wanted to point out that it doesn’t have to be a language runtime. Using Rust as an example again, you typically spawn the Tokio async runtime on program start, if you’re gonna do
async/awaitstuff.
Less bugs, less maintenance work. All code is a debt, either you or the users have to pay.
- The fastest code is the code you don’t run.
Not really. The code can be slow, even if I do not run it. Also, sometimes additional code can do optimization (like caching), which is more code = faster. Or additional libraries, complexity and code paths can in example add multicore execution, which could speed up. So, I do not buy the less code is faster logic.
I think it really depends on what your code is doing. I do agree that less isn’t always faster or as they mentioned in the post, safer. Taking a raw input is fast, but not very safe for a variety of reasons. I personally make “simple to understand” as the highest priority for my code.
On a slightly different example, the suckless project has a huge emphasizes on lightweight code, which they call “suckless”. I don’t think in this case faster is the goal, but having less code and be simple as possible (not even configuration files allowed, you just recompile program) and almost no documentation in the code either. But the idea is the same, of having “lightweight” code.
I think the most obvious example is loop unrolling. An unrolled loop can be many times more code, but runs faster because you’re not updating a counter or doing conditional jumps.
