Does a hobby OS need to be useful?

You probably do want to speak IP, but if you want to do all the work, you don't need to use BSD sockets.

POSIX and other such things will let you use more of other people's software... but that's a choice.

You can also do a lot of reuse if you just want to focus on some parts. There's no need to write a bootloader unless you want to. You can pull in lwIP for the IP stack. You can do a lot of interesting (I think) things with barely any userland if you want.

If it's supposed to be a general purpose OS, you end up looking a lot like existing general purpose OSes though, cause it's hard to build everything and the closer you get to existing OSes, the more existing software you can leverage.

If you only want to run a subset of existing software, you might not need to cover that many of the syscalls from whatever OS it was targetted at. My hobby OS runs one specific FreeBSD executable, and it doesn't take that many syscalls to do it.

> Imagine we discover software from another planet that would not even work with 8-bit bytes, for instance.

8-bit bytes didn't become fully dominant until maybe the 1970s. It's a bit hard to find software older than that, but it's around somewhere.

> I wonder what a hobby OS would have looked like it if it assumed nothing

There's lots of stuff that was around in the 1960s that became dead-ends because everybody piled in on another way of doing something, and then things got optimised for that way, then every other way was less performant, and soon the lucky one way became the only way.

I've never really had much time to explore these early ideas from well before I was born whilst working, but definitely think it'd be a fun way to spend my time when I eventually retire - to try to recreate some of the stuff from that era that got forgotten and see what it could be developed into nowadays.

For example, just watch Douglas Engelbart's 1968 demo [1] for some ideas - some of the things presented in that took decades before it was rediscovered and implemented on systems available outside of research environments.

[1] https://www.youtube.com/watch?v=yJDv-zdhzMY

Rob Pike of Plan 9 and Go fame lamented this back in 2000 in his “Systems Software Research is Irrelevant” talk:

http://herpolhode.com/rob/utah2000.pdf

I’d love to see a new operating system that explores radically different APIs for applications. The trouble is writing an operating system is a large effort. Barring market effects, OS has to be heads-and-shoulders better than existing ones in order to convince application developers to write software for it. Windows, macOS, and Linux are good enough for most people, even techies. Additionally, it is often easier to modify an existing operating system such as Linux than to go through the trouble of writing a brand new operating system.

For a hobby OS .. nobody is stopping you. I have my own hobby OS where everything is custom, I don't even have a standard C library, I made my own, I don't copy posix conventions, everything is just implemented as to what makes sense to me. I never understood the people that go through all the effort to work on their own OS, and all their work is to just create a linux/windows clone from scratch. Of course even accomplishing that is a huge achievement, but if you're going through all the effort, why not do something different?

I wonder what a hobby OS would have looked like it if it assumed nothing, that is, as a thought experiment ...

There are some hobby OS projects that do this. The best known example is Terry Davis' TempleOS.

https://en.wikipedia.org/wiki/TempleOS

If you only want to please yourself, you can dispense with all that legacy stuff. You can stay busy and enjoy yourself in your own world. Some of the hobby OS listed here look like they might be standalone worlds:

https://github.com/jubalh/awesome-os

If you have a goal of making your OS usable, you have to design it with this in mind. Build your kernel and driver subsystem in a way that legacy OS interfaces can be implemented on top of them. Find ways to repurpose existing infrastructure so you don't have to do all the work yourself.

It would probably look like the Third Temple.

This, I assume?: https://github.com/vinc/moros

Pretty impressive, you've gotten much farther than I ever did (I didn't have the patience to implement all the borderline boilerplate an OS needs).

I bet it's just running llvm-pipe to render in software.

You mean a packaged file that one simply runs? Isn't that the installer?

You mean ISOs? ;)

ScummVM has effectively done this.