"Starting Over": my FOSDEM 2021 talk

[liam_on_linux]

My talk should be on in about an hour and a half from when I post this.

«

A possible next evolutionary step for computers is persistent memory: large capacity non-volatile main memory. With a few terabytes of nonvolatile RAM, who needs an SSD any more? I will sketch out a proposal for how to build an versatile, general-purpose OS for a computer

Comments

[liam_on_linux]

Ha! Excellent. :-D

I had one very brief play with Taos at an Acorn User show about 25Y ago.

A few of us are hunting for a demo/cover CD of AmigaDE 1.2.1 that was on Digital Magazine. It would be the latest extant version of Elate, I think...

Speaking of which, have you seen ChrysaLisp?

https://github.com/vygr/ChrysaLisp

[waistcoatmark]

Deleted my comment as it used real names, and LJ won't let me edit a comment after it's been replied to :-(. Will ask around for the demo, but I suspect you'll be lucky after this amount of time...

[liam_on_linux]

That's a shame. You could have quoted yourself and censored it...?

[waistcoatmark]

Bloody LJ and binding Enter as "Send comment"...

I read an (internal to Google) email from Ken Thompson (might have Rob Pike. One of the Big Names anyway) about using Plan 9 and it really was very different. IIRC, you could send running instances of programs to colleagues on a different machine and they could open it and carry on running it from where it left off on their machine. He did make it very clear that he was just covering the good bits in the mail, there were plenty of bad bits he was skipping over that just wouldn't work for one reason or antoehr in today's environment.

[liam_on_linux]

:-D

I have tried it a few times but it is quite so weird -- a little bit of that weirdness inherited from the original Oberon, which Cmdr Pike admired -- that I find it very hard to operate at all.

I used to have a bootable image of Inferno and had a very short play with it. It made a lot more sense, and had a vaguely familiar GUI.

The comparison is rather like between Oberon and A2.

I still reckon Plan 9 has a lot of potential. It has APE, a POSIX environment. I wonder if it might be possible to take some code from Joyent's SmartOS -- they upgraded Sun's long-neglected Linuxulator code, to make it able to run 64-bit Linux binaries compatible with kernel 4.something on Solaris.

If Plan 9 could run Linux-binary Docker containers, say, ISTM it could be used to make a very lightweight replacement for Kubernetes.

My wilder idea is to wonder whether it might be possible to in effect merge Plan 9 with Inferno to get something with the best of both worlds. Nobody needs Inferno's ability to run as a browser plugin or a program any

Inferno runs on Plan 9

[ext_5761010]

Inferno already runs on Plan 9, or it used to.

I don't think you would get very far merging one into the other, especially if you wanted to keep the GUI from Inferno. My impression is that many Plan 9 users are quite happy with rio. I don't think there are many people so attached to wm in Inferno.

Re: Inferno runs on Plan 9

[liam_on_linux]

Thanks for the comment

[anonymous]

A few other systems that might be worth looking at as precedents:

Racket -- a Lisp with alternate concrete syntaxes, like Dylan, but with a great deal of more recent activity (if nothing else, it's the focus of a lot of active university research).

Multics -- a system with everything in a single large address space, which programs could selectively map in; there was no conventional read/write style I/O at the kernel level.

[liam_on_linux]

Thanks for the comment!

Yes, I am aware of both.

I have looked at Racket, and my concern basically is this: I think its syntax improves things _for Lisp programmers_ but does nothing to lower the very considerable barrier to non-Lispers. My interest was in things which serve the role of the planned MLisp, so I looked at CGOL, PLOT, SweetExpressions, etc. But I did also read up on Racket, ARC, Clozure, Clojure, etc.

Multics was and is interesting, especially now that there are working emulators. It was of course not the only single-level-store OS. IBM i, formerly known as OS/400, is another, and in much wider use today. But as far as I can tell, the way that Multics and OS/400 treat single-level-storage is that it's all disk. They don't make RAM visible; the OS managed paging, demand-loading etc. transparently.

This is good stuff, but sort of the reverse of what I am considering. :-)

[waistcoatmark]

Surprised that you didn't mention HP's The Machine https://www.labs.hpe.com/memory-driven-computing or was it not worth mentioning after HP killed it off?

[waistcoatmark]

And I'm not sure I agree that one language all the way from the top of the stack to the bottom is a good idea

[liam_on_linux]

So there are multiple possible responses to this. In no particular order:

• The attentive reader/listener will have noted that my proposed system design includes _two different languages_ within it ;-) and I submit that this is not merely a compromise but also a pragmatic one. If someone were to do it the way I'm suggesting, bit-bashing could be accomplished in Oberon, with the fancy GUI stuff in Smalltalk or Lisp.

I do not currently have a plan for how to mix the two but I suspect that it would not be that hard to accomplish.

• Later LispMs supported C, Fortran and a few other languages. This too was a pragmatic compromise.

• AFAIK, Smalltalk machines didn't, but then, Smalltalk machines never really took off, they just inspired the shape of the entire industry from then on. OTOH, Xerox's ones ran Mesa or Cedar underneath, so in principle you could use that.

[tpear]

Thanks for interesting presentation

[liam_on_linux]

> Thanks for interesting presentation