Rob Pike, one of the creators of Plan 9 from Bell Labs, spoke about this in his 2000 talk "Systems Software Research is Irrelevant" (http://doc.cat-v.org/bell_labs/utah2000/utah2000.pdf).

Writing an OS from scratch is hard work, and without complying with pre-existing standards, the new OS won't be able to take advantage of existing software or communicate with computers running different operating systems, making it a non-starter for most users. Even supporting a wide range of hardware is very challenging, especially given the fact that many hardware devices lack documentation sufficient to write drivers, which results in challenging reverse-engineering projects or relying on the hardware manufacturer to create drivers for the new OS; this is a challenge even for Linux, which has the best driver support outside of Windows. Most people who want to experiment with new ideas for computer systems end up building on top of Linux since they can leverage Linux's existing hardware support and its implementation of a wide range of standards allowing interoperability.

I'm working on an experimental desktop computing project that ironically uses Plan 9 as its base platform since Plan 9's everything-is-a-file interface and its overall simplicity allows me to build my project with less effort than dealing with Linux/X11/Wayland/dbus/etc., but a big part of me is tempted to eventually write some 9P services to emulate enough of the underlying Plan 9 architecture to run my project on Linux and the BSDs in order to take advantage of these operating systems' better driver support.

A possible path would be something initially running on an SBC, like a RISC-V board or even an FPGA. People buy these boards to experiment and use for simple tasks. Something really interesting could gain a small initial following and slowly grow from there.

It does not have to be completely isolated. You can still have support for some common communication protocols and maybe some client software running on a mainstream system to do various things.

I'd like to see a minimalist os that just runs multiple instances of wasm micro or maybe wasmer. It would only have to support the wasi. A scheduler, time support, a sockets implementation and a file system. Targeting a sbc makes a lot of sense for that.

Urbit.

IMO, the interesting future is RISC-V and FPGA. Potential is huge. Just needs better availability of devkits. Long lead times on much of it.

On plan9/9front, I'd love to port Nethack/Slashem with NPE:

https://git.sr.ht/~ft/npe/tree/master/item/include/npe

But not with a TTY UI. With libdraw:

http://wiki.9front.org/programming-gui

This reminds me that Alan Kay really hates Linux: https://youtu.be/rmsIZUuBoQs

This stings since I really like Alan Kay and have been influenced by his ideas, but am also working on a Linux-based VR headset as a thinking tool[1]. I think Alan would approve of the hardware but not the software, perhaps ultimately saying something like "you can spend your whole life dicking around in Linux, but still not understand anything about computing".

[1] https://simulavr.com

> Linux has done more to kill exotic OSes than anything Microsoft or Apple has ever done.

I'm pretty sure commodity hardware, standards like POSIX, and the fact Unix-like OSes being taught in operating systems classes really did more than just Linux alone.

> Unix won so hard most everything else is dead; even Windows is very "unix-like" in ways people don't even realize.

I'm not so sure. I learned to code on VMS and then AT&T Unix, and I see a lot more VMS in Windows than anything. VMS was actually pretty interesting itself.

I worked for 5 years on VMS for a real-time satellite image processing HW/SW system for NASA. VMS was wonderful. At another company, I then worked on a Unix system for a few years, then 2 years on a port of our software to Unix. (The port was smooth and quick, thanks to VAX C and its relatively complete library; most of the 2 years was spent adding new capabilities for a particular customer.) The rest of my career has mostly been under Unix.

Microsoft hired the main architect of VMS, Dave Cutler, away from DEC to design Windows NT. (VMS++ = WNT!) I haven't worked on any in-depth Windows programming projects. I did read an article some years ago about Cutler, VMS, and WNT. The author pointed out an insightful distinction between VMS and Unix with regard to I/O operations: VMS tells you when an operation completes, whereas Unix tells you when you can begin an operation. As a result, I think, asynchronous I/O under VMS was there and available from the git-go, but always seemed like some odd thing grafted onto Unix, making it more painful than it should have been. I have never used it, but I believe WNT's overlapped I/O was patterned after the VMS model?

> Microsoft hired the main architect of VMS, Dave Cutler, away from DEC to design Windows NT.

Not sure why you mice words. Cutler developed NT at Digital. That's better. In 1988, Cutler took his work (the literal data, the MICA OS code from DEC's cancelled Prism RISC project) and his engineering team at DEC with him to Microsoft. I believe it was much less like poaching and much more like defecting. DEC ultimately forced Microsoft into an alliance under threat of lawsuit for the IP theft to migrate the VMS user base to NT, with Microsoft paying quite a lot, $180M, to avoid lawsuit, for, get this... training DEC engineers to use NT. In 1995, someone at MIT found large chunks of DEC MICA code unaltered, including comments, in Windows NT.

There is a book that addresses the reasons for Cutler's departure from DEC, but it does not go into the details that you have mentioned.

https://www.goodreads.com/en/book/show/1416925.Show_Stopper_

But then either VMS is much closer to UNIX than to things like IBM i (or IBM mainframes, which are however themselves very, very different from IBM i), or Windows is closer to UNIX than VMS. Because Windows is very close to UNIX compared to either of those.

> But then either VMS is much closer to UNIX than to things like IBM i

I'm not really sure that is the case at all. VMS had versioning file system that had support for multiple file types including stream, sequential, indexed and relative. It had a very different security model, four levels of processor access (unix there's kernel and userland) and it had very different networking capabilities, including clustering baked in (in 1983).

> Because Windows is very close to UNIX compared to either of those.

There's deep VMS heritage in Windows NT: both VMS and NT were written by David Cutler.

I know! But then VMS is maybe closer than it seems? Have a look at IBM i.

Agreed. I've only limited experience with i, but more with z. After using them, the other OSes you might use start to look and feel pretty similar to each other.

Microsoft hired the main architect of VMS, Dave Cutler, away from DEC to design Windows NT. (VMS++ = WNT)

You've got it backwards. Digital cancelled Cutler's pet project and he shopped himself, and Digital's IP, to Microsoft. VMS and NT have a lot of similarities, but NT is not VMS, though they have a common ancestor: MICA = NT

[0] https://en.wikipedia.org/wiki/DEC_MICA

Windows has very stringent file locking; Unix doesn't.

I believe this major design difference came from VMS.

> even Windows is very "unix-like" in ways people don't even realize.

WSL is a tacit acceptance of UNIX dominance.

WSL is Microsoft doing its classic "embrace and extend" strategy, yet again.

https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguis...

Even before that, much of the UI of DOS itself is Unix inspired (DOS even had “files” for certain devices such as PRN and COM)

There are VMS-inspired parts also, but even that is somewhat in the Unix sphere of things.

> Even before that, much of the UI of DOS itself is Unix inspired

The inspiration rather came from CP/M.

...which in turn was strongly influenced by a previous product from DEC.

"Various aspects of CP/M were influenced by the TOPS-10 operating system of the DECsystem-10 mainframe computer, which Kildall had used as a development environment."

https://en.m.wikipedia.org/wiki/CP/M

Not exactly - several "killer features" of DOS 2.0 were features inspired by Xenix and which were not present in CP/M, including a /DEV/ directory, the mere concept of hierarchical filesystem and optional use of forward slashes in directory paths - backslashes were introduced because of CP/M legacy in commands

No they weren't.

This is an urban myth of computing, and it needs to die.

CP/M commands did not accept command-line switches in any standard way. You are trying to "correct" people who are telling the real story by repeating a myth.

http://www.os2museum.com/wp/why-does-windows-really-use-back...

Ah, I seem to have been waylaid by some of the older mentions there - thanks for the link with mention of M-DOS. I was certain of DEC connections, but thought it came through CP/M and forgot to take into account PDP-10 history of Microsoft.

With the upgrades to WSL in Windows 11 you can run fsv[1] with minimal hassle and finally get the true UNIX experience[2].

1: http://fsv.sourceforge.net/ 2: https://www.youtube.com/watch?v=dxIPcbmo1_U

Here's hoping Haiku gets more momentum. BeOS was a little less UNIX-like, I think.

I kinda feel a modern System 7 [0] like OS could be neat. But with a terminal included.

It was kinda easy to get your head around the OS just navigating through the system folder. The UI was also more consistent than anything we have these days.

———

[0]: https://en.m.wikipedia.org/wiki/System_7

Not sure if that's old people painting the past rose, but System 7 was not awesome. Cooperative scheduling and lack of memory protection were anachronisms already in 1991.

I think we're talking about different things here. The UI might have been superior in MacOS, but the technical foundations of Unix clearly were. Something OSX tried to (belatedly) reconcile. Nothing stops you from adding an awesome UI on top of Linux (although I'd say UI will always be a matter of personal taste and prior experience).

System 7 was so awesome. I was still developing Macintosh GUI software when it came out. It was great to use, and visually pleasing. And as you say, the user interface was consistent, also discoverable, and things did what you expected. Attributes I miss as GUI software inexorably moves into the browser.

I have some PPC macs that have OS 9 on them, I love the UI. I look for excuses to use them whenever I can.

However, Haiku is gaining a ton of GNU/Linux software. It becomes less interesting to me when people are just porting the same junk to it.

Serenity is far more interesting.

I think Haiku biggest mistake was to try to be a regular desktop OS instead of focusing on a specific hardware like RaspberryPi.

In their defence they’ve been at it since 2002 before any of those fun little pieces of hardware existed.

Well, they were also quite hostile to the idea of targeting the Pi. I remember being involved in a forum discussion where it was dissed as a toy, another around ARM never being a relevant platform, and another about the Broadcom bits being proprietary.

To this day, Haiku has been dragging its feet on the Pi (and yes, I know it’s a community effort, that it requires sponsorship, suitable volunteers, etc. - I’m just pointing out that they dropped the ball in even acknowledging the need for a port).

‘Need’ is a big word to use for something voluntary.

Most of the Linux crap it's QT based, which looks good enough in any OS.

For the rest there are some interesting native tools such as video editors.

But Haiku needs 3 things to get more real life use cases:

- GL/Vulkan for older Intel (both old and new)

- Async USB

- UVC webcams

Serenity is still a very Unix-like OS though.

It is, but it doesn’t use any of the existing open source software, which makes it fun, different, and interesting. Heck, just having a non-blink/non-webkit/non-goanna/non-gecko browser is cool.

Their ports system (which they didn't invent, of course) offers a pretty nice compromise of continuing their "everything from scratch" ethos while letting people have fun porting. And who knows, maybe the ports will allow it to be a practical OS someday.

Agreed.

This amazing demo of a in-progress operating system (writen by a single person) which has been discussed previously on HN. The developer says near the end of the video:

"I'm not a big fan of unix myself and so 'essence' is designed without any sort of unix influence"

Essence OS demo (2021): https://www.youtube.com/watch?v=aGxt-tQ5Bt

In what ways do you see Fuchsia heavily influenced by Linux?