/tech/ - Technology★

>>969155 (OP)

Daily reminder that "proving code correct" means proving a relationship between a particular specification and an implementation, in no way meaning the specification has anything to do with the reality of the system. In many cases the specification is LARGER than the implementation.

You said shells twice. It would have a shell based on a "memelang" like a lisp or forth. Forth is often used as a shell on embedded systems.

>kernel

Something based on L4, indeed.

>filesystem

Something like XFS with self-healing properties would be nice. Leave volume management, raid management, compression on deduplication out of it.

>shell

I'm partial to sh with some of the most popular features/fixes of bash/zsh/ksh implemented. Getting out of the quoting hell might be nice too (arrays kind of solve it, though).

>POSIX

Provides a compatibility layer while fixing and improving the whole mess.

>>969159

>Forth is often used as a shell on embedded systems.

It's not used "as" a "shell", there's no shell, and it's not imitating the shell that isn't there.

>>969173

You do realize the bash shell is just a shitty programming language right.

>>969155 (OP)

sel4 or microkernel-ified openbsd modified for an (obviously managed) pan-network memory address space so that multiple computers can be wrapped in one OS. In this sense, it would probably be more of a meta-os.

The filesystem would probably be HAMMER2 or ZFS. Otherwise, it would probably need to be custom-written.

Shells would be the same as most systems. There would need to be modifications to accommodate for the distributed nature of the OS.

>>969179

>microkernel-ified openbsd

I thought the BSD's already used microkernels. I thought that was one of their main selling points?

>>969194

Not at all. The only real microkernel OS out there that anyone uses is QNX.

Gentoo < Arch < Ideal OS > ReactOS > Windows > Ghetto SBC > OSX > Fridge

>>969195

OKL4 is more used than QNX will ever be, m8.

>>969202

I used QNX for a few years. It was reasonably supported.

Say this in the seL4 manual, can someone explain this to an OS noob? This sounds like the opposite of what should work?

>There are no arbitrary resource limits in the kernel apart from those dictated by the hardware and so many denial-of-service attacks via resource exhaustion are avoided.

This seems like it would delay resource exhaustion at best?

>>969202

OKL4 is a hypervisor you fuckwit. Your VM runner does not count as the base of an OS.

>>969221

>This seems like it would delay resource exhaustion at best?

Yes anon your hardware is not literally infinite. What the fuck do you expect.

>>969224

Avoid is different than delay. Avoid means it won't happen. Delay means it will just happen later than compared to other operating systems.

>>969221

It means that a userspace program (which in a microkernel architecture is basically everything) can't starve the kernel for resources and disrupt the system. In Linux and other OSs its possible under some conditions to consume more and more system resources until a kernel panic or system crash occurs.

Docker is an example of a program which does this, if the application running in a docker container logs to stdout the docker daemon will leak memory until there are no more free resources available and the system crashes.

>>969155 (OP)

An ideal operating system would be microkernel based but the userspace applications will be run with a thin wrapper which translates the system calls so that the underlying microkernel can be anything rather than a specific microkernel. The wrapper will also be dynamic so that critical aspects of the system such as drivers and the filesystem can be easily swapped out, even while the system is running, without having to modify the applications. For god mode you can also have all the userspace applications distributed as LLVM intermediate code and then JIT them at runtime.

In terms of POSIX compliance probably not, POSIX is actually a bottleneck in highly concurrent systems and a few proposals for replacements exist. The UNIX style shell works and most people are familiar with it.

>>969155 (OP)

I'd prefer Minix over seL4 since it has real world use on millions of machines.

Genode, ZFS, Qubes like interface/VMs sel4 kernel, Power9 or Riscv arch

>>969427

Why would you need a Qubes-like interface with VMs if you have a Genode system built on seL4?

Why not just use pure userspace programs on Genode?

Probably something like Qubes with a "microkernel" of sorts (while we're dreaming, I'd like it to be written in Ada). Theo says virtualization is buggy and exploitable, but I'd like to see him crack into a Qubes system. Until then, I'll believe virtualization is more secure than bare-metal.

>>969419

Minix probably wouldn't be terrible, but the drivers need some serious retooling to make anything useful for general computing.

>>969431

sek4 can be used as a hypervisor, and there's plans to make a qubes fork with it instead of xen.

>>969429

to run other OSs when needed

>>969201

What if we can make reactOS become arch's subsystemd?

>>969411

> POSIX is actually a bottleneck in highly concurrent system

Care to elaborate?

>>969202

QNX is used in over 20,000,000 automobiles.

>>969155 (OP)

this thread is the most cringeworthy buzzword salad i have ever read in my life

>>969772

https://web.archive.org/web/20120211210405/http://www.ok-labs.com/releases/release/ok-labs-software-surpasses-milestone-of-1.5-billion-mobile-device-shipments

>>969664

https://www.nextplatform.com/2017/09/11/whats-bad-posix-io/

>>969411

>An ideal operating system would be microkernel based but the userspace applications will be run with a thin wrapper which translates the system calls so that the underlying microkernel can be anything rather than a specific microkernel. The wrapper will also be dynamic so that critical aspects of the system such as drivers and the filesystem can be easily swapped out, even while the system is running, without having to modify the applications. For god mode you can also have all the userspace applications distributed as LLVM intermediate code and then JIT them at runtime.

i came

>>969446

>sek4 can be used as a hypervisor, and there's plans to make a qubes fork with it instead of xen

Assume you meant sel4 here, but source pls. Can sel4 run on POWER9?

>>970265

what plans? where?

>>969995

People are actually already doing something like this with POSIX compatibility. Most POSIX compliant programs can make syscalls that either the linux, openbsd, or net/freebsd kernels can interpret. Since userspace calls for linux never change, only the internal kernel system calls, someone should implement the userland GNU/linux system calls as a translation layer for a intermediatary which you can then fork and fill in with your own custom kernel. Like you could run windows NT under it if you so wished *why would you ever do this?**. Or you could port SEL4 to the intermidiatary. To make all the userland code a LLVM intermidiatary would be trivial to do on the fly. Then you just need to write a super secure/performant JIT compiler.

>>969194

Nope, BSDs are monolithic. The only thing remotely close is DragonFlyBSD, which is hybrid-ish

I want something with a package manager like Pacman, but not systemd.

A kernel like Hurd that actually works.

A filesystem like ZFS.

A DE that's like KDE4 or NEXTStep.

Bash is okay I guess.

A virtual assistant would be cool I guess, but if it has to be included it should have allegiance to me.

A VM that's not VMware or Oracleshit with a non-cucked TempleOS image out of the box.

A decent fucking browser, holy fucking shit can I just get a good fucking browser.

While I'm at it, nuke the web and start over.

>>969174

Why the fuck did no one replace it with luajit yet?

All that "compatibility" crap gets on my nerves.

>>969155 (OP)

My ideal OS would be an OS that seamlessly work in all my devices and integrate with each other well like Mac, just werkz like Windows and have the transparency and security of Linux.

Allow you to easily isolate apps

Is lightweight, doesn't use scripting languages for core components

GUI doesn't treat you like you're and idiot

Package manager is functional, like nix, but also simple

Package manager knows app isolation

>>969155 (OP)

It should be a unix like system

It uses micro kernel

No systemd and have proper init system

Open Source development model

The GUI should have two modes- Tablet mode similar to Gnome/Unity when I remove it from the keyboard and a PC mode similar to KDE/XFCE when attached to keyboard.

Software should jest werk including professional software like Office and Photoshop but that might be asking too much

Should use the rolling release model to update the core system(DE, Kernel, bash, GNU tools, compilers, etc...) while user applications such as office, games, browser, etc... installed at system level using application stores such as Steam, Synaptic, etc... along with support to even install windows applications like how proton did on steam and those are installed at the user level instead of system level

Cloud integration for all the popular platforms

Should have most developer tools working especially libraries for C/C++, Perl, Python, Java, Lisp, JS, SQL, etc...

>>981993

He didn't say he wanted to use Xorg.

>>981993

Most definitly not Xorg, also where the fuck did xorg fit into his post?

>>969155 (OP)

Windows 10, it keeps me safe from all the bads.