Hacker News Reader

Comment by bjackman | Hacker News Reader

Cool! Why?

(I have also spent the last year working on a homebrew OS kernel, so please trust that my "why" isn't intended as an attack! For me, the answer was simply "to see if I can", and I really enjoyed the project, but I'd love to know what motivated you in particular - especially as you seem to have achieved a very impressive amount)

View on HN

Sortix is because I want to write a good operating system that I like myself. I do like if it's generally useful.

To paraphrase Cave Johnson, OS dev isn't about why, it's about why not!

Also to get your OS into space. That's definitely on the bucket list.

Yes, is there any particular niche this OS is targeting? What are these traditional design mistakes that are being avoided?

If the purpose of creating Sortix is simply the challenge of writing an operating system, that's all good too. Best of luck.

First mistake to avoid; writing it in C.

What would you use, then? D? C++? (Rust is not ready yet, alas.)

For a POSIX-compatible operating system, more exotic choices are probably not available.

Is Rust not ready? Maybe part of the challenge is writing it in Rust

Imagine that you used std::thread::scoped for a few months in some critical places of your project.

If you use a stable language, you can concentrate on the challenges of the actual OS design.

In some cases, especially new projects (like a new OS), I think the pain of using an immature-and-powerful language is less than using a mature-and-weak one.

> Rust is not ready yet, alas

Was C ready when Ritchie & Thompson wrote Unix with it?

> For a POSIX-compatible operating system, more exotic choices are probably not available

Why limit to POSIX compatibility?

Ritchie and Thompson also sort of wrote C, at least, some parts of it. I fully understand when Mozilla writes Rust and Servo using Rust.

No need to limit things to POSIX compatibility, but merely achieving it is no mean feat — and POSIX is seriously C-oriented. Some people invent a new OS architecture, then implement POSIX on the top of it (see NT) — but this is an order of magnitude harder, I suppose.

When you implement a POSIX-compliant OS, you suddenly have a lot of ready-made open-source software that both serves its primary purpose and as an "integration test suite". For a new OS, you have to write most basic userland things yourself. This is again a lot of effort.

> Ritchie and Thompson also sort of wrote C, at least, some parts of it.

Given its open nature, there's no reason (except limits of time/skill) someone writing an OS in Rust couldn't also be advancing Rust and its libraries in the directions needed to support that effort, even if they are outside of Mozilla.

> except limits of time/skill

I won't mind to be a billionaire with superhuman abilities and unlimited lifespan, too! Unfortunately, we mere mortals often have to limit the scope of our endeavors somehow.

Order of magnitude strikes me as too high if you think about doing it ahead of time and factor it into your design. The harder part would probably be not letting POSIX compatibility cramp your design. For example, the use of fork() is in part an artifact of making a practical system run on very memory limited hardware (like 128KiB max for a program, half code, half data), there's no intrinsic need for it to develop solid software.

Some very smart and pretty motivated folks at GNU/Hurd project struggled for years with a similar, but a bit less ambitious, approach. You never know how deep a puddle is.

http://www.projectoberon.com/

POSIX? We don't need no steenking POSIX...

Too bad this project did not develop further.

Modula-2 and of course Oberon the language were rather fine for their times, and I'm glad to see some pieces of them reused in e.g. Go.

Says every pundit who's never tried to write an operating system.

Reply on news.ycombinator.com