MINTIA (not vibecoded)

[flat dust]

i have my window open and some dude's loudspeaker is loud enough i hear it from the second floor 😭

that compiles to 8086 anyway, 286 protected mode is either assembly or other torture methods

[twilit smelt]

i860 and i960 are completely unrelated by the way

ik its confusing

theyre completely separate ISAs

no relation

[flat dust]

oh

i got fooled :(

[twilit smelt]

i960 is a RISC-ification of iAPX 432 basically lol

[flat dust]

wasnt that some early 80s cpu

[blissful smelt]

iAPX 432 with jackal when

jAPX 432

[flat dust]

you need to give him a deathwish or something first

so he has nothing to live for

4% chance he'll do it in that instance

[warm mural]

Watcom does

[flat dust]

o

[warm mural]

I think

[flat dust]

idk

[warm mural]

Just use asm tho

[flat dust]

true

[tame phoenix]

I'm sad that I lost my first ever os source.

[flat dust]

i remembered some dude going insane in #lounge-0 over making a 286 OS

[tame phoenix]

it was before I even knew what CVS was

[flat dust]

the worm obsessed dude

[flat dust]

:(

[flat dust]

i think i lost mine too but it doesnt work

i had a kernel shell 💔

and a VGA 16 color planar mode driver 💔

the source code of which i stole from that one example on the wiki

[tame phoenix]

mine worked, it had a filesystem and all sorts

[flat dust]

fair enough

[tame phoenix]

all in assembly with nasm

[flat dust]

ooo

[tame phoenix]

it was all real mode, separate shell executable

[flat dust]

it seems like a good idea however with his experience i doubt it'll be in any way even a funny esoteric OS, sadly

[mortal thunder]

i guess i must be one of the greats then

[flat dust]

closest thing to user mode shell in real mode i guess?

[flat dust]

you probably have the worst userspace and the worst best kernel

[mortal thunder]

boron is certainly not the "best" kernel and i was actually talking about my previous project which is nanoshell

[flat dust]

oh

idrk anything about nanoshell, i just know it had a GUI

[mortal thunder]

Its kernel is pretty bad and its userspace is not that bad

[flat dust]

fair

the worst kernel by far is linux, it has source code that holds itself together by the pope praying over it 7 times

literally cant read it 😭

[mortal thunder]

I disagree

[tame phoenix]

only thing left on archive.org is the docs
https://web.archive.org/web/20010505050017/http://www.proweb.co.uk/~crypt98/

ugh that isn't the right url

[flat dust]

i mean its more of a subjective thing however i still have problems understanding it

[flat dust]

i have problems with reading any source code that isnt mine honestly

[mortal thunder]

huh

[mortal thunder]

it's normal

[flat dust]

2001

[flat dust]

i guess so

[tame phoenix]

ah found it
https://web.archive.org/web/20010825011238fw_/http://www.proweb.co.uk/~crypt98/bxos/index.htm

the frames weren't giving the right url

the os was 1999

[flat dust]

bxos

[flat dust]

it probably compiles in a microsecond nowadays

[tame phoenix]

it was the first letters of mine and my friends nicknames

brain and x3r0

[flat dust]

i wonder who brain is

[tame phoenix]

yes I've had the same nickname since college lol

[flat dust]

ive had mine since i was like 9 or something

[tame phoenix]

yeah it would lol it was tiny

[flat dust]

honestly despite it possibly being tiny it might still take longer

[tame phoenix]

my current os doesn't compile fast any more

[flat dust]

my powerpc bootloader compiles in about uhh

0.5 seconds ok thats fast

[tame phoenix]

actually now I'm curious

[flat dust]

i have xorriso afterwards, and then its also my build system kconfig ripoff compiling menus

[flat dust]

then after a semicolon theres a kilometer long QEMU command

as intended

[flat dust]

?

[flat dust]

that is kind of an achievement of either how crappily oversized your codebase is or how much features are in there

holy crap my laptop screen is so dark

[tame phoenix]

hmm 15 seconds not so bad,but is a 32 core machine cheating?

[flat dust]

idk

i compile my thing with what i assume to be -j 1

[tame phoenix]

that would take me ages, I have lots of files

[flat dust]

gcc compiled for hours until i did a -j 12

(i have 12 cores i think)

(sounds appropriate)

[flat dust]

fair

[digital pivot]

btw have you looked into xnu scheduler?

an interesting fact about is that it uses some freebsd ule heuristics iirc

[flat dust]

doesnt XNU literally have freebsd inside it in some form or idk

[hollow musk]

just so you know I didn't even know they had open source code 😄

[flat dust]

apple open source a bunch but nobody knows about it

their powermac bootloader is also open

BootX

[hollow musk]

well, I knew, but I didn't know they updated it more often than every 10 years or so

[flat dust]

okay yeah thats true

the left side of my laptop is being deepfried from sunlight

[digital pivot]

yeah but the scheduler is not in the freebsd land

[flat dust]

understandable

[digital pivot]

they call it clutch scheduler

[blissful smelt]

XNU might have some leftovers from 4BSD

[digital pivot]

maybe this will be interesting fro you https://github.com/apple-oss-distributions/xnu/blob/main/osfmk/kern/sched_clutch.c

[blissful smelt]

I doubt they have ULE

[digital pivot]

they dont have ule

they use their own scheduler which has a piece of logic used in ule as well

and they took it from there

i think it was confirmed by someone knowledgeable about xnu

( @hybrid condor i think it was you )

[warm mural]

it definitely does

XNU is a mix of old bsd code + code imported from freebsd + mach + its own stuff

[shadow ridge]

💀

[raven drift]

It's true that most of the time people NIH things, it correlates with them being unaware of proper solutions, especially in the hobby space

or uninterested in reading about / researching the state of the art

I've not only seen this phenomenon with OSes but also with other kinds of algorithms etc

[shadow ridge]

i think exceptional programmers will stumble onto the same solutions anyways

or better solutions even

[raven drift]

I think what differentiates exceptional programmers is more the ability to see where you can innovate

instead of trying to come up with an improvement / NIH solution for every possible problem

[shadow ridge]

i mean, most people wont be like "i have already created a mini universe with a great particle physics simulation. now i can simulate my own computer in it and finally write hello world from scratch"

[twilit smelt]

anyone who does nih is an irredeemable idot.

[raven drift]

I'm not saying that NIH is bad, I'm saying that even if ppl NIH, they should research the state of the art first (unless they don't care about quality which is also fine for a hobby project)

I think one of the things where I saw the most severe cases of "ignore SotA, just NIH" was when I wrote a stock trading program a few years ago. Like, there are known trading strategies that make money with reasonable consistency (but with returns that are not high enough to make you rich) and that work because there are fundamental reasons why they have positive expected value. And you can learn about them by reading a few finance books etc. and there is nothing secret about the basic idea. Yet if you look at what ppl do online it's the wackiest shit ever that consistently just loses money.

Like, these ppl do not even want to reach their supposed goal (even though they claim that they want to) because doing so would require them to do some boring studying

[shadow ridge]

i agree

[twilit smelt]

i dont understand doing NIH without studying existing state of the art

the whole fun of NIH is learning the state of the art by doing

you cant have "ohh thats why thats like that" moments without knowing what its like

[shadow ridge]

"the state of the art sucks, i could do it better" immediately followed by "ooooh" moments are so good ngl

[twilit smelt]

also i feel like any "wow thats cool!" factor an NIH project would have would come from the fact that its entirely NIHed and still does all the cool modern stuff because that means you were extremely autistic about learning how those things work

[twilit smelt]

ive never had that kind of arrogance i think i tend to assume the experienced ppl who wrote existing kernels were smarter than me

[digital pivot]

sometimes you need to invent your own stuff, right?

[shadow ridge]

surely there were things that seemed so weird and out of place that you thought it must not be the best solution

[twilit smelt]

the only good idea ive ever independently come up with thats an objective improvement was if NT's driver stacks were traversed iteratively rather than recursively during IO enqueue

and it turned out shortly thereafter that an NT guy already came up with that idea

and that they were working to implement minifilters that way

and already wished theyd done the IO enqueue iteratively during NT's original development

and not only that but theyd been well aware of it for 20+ years before i came up with it lol

its not even original in the hobby space - we found out minoca's IO packet enqueue is iterative

like mintia's

i didnt notice that until after i did it

[digital pivot]

if you want to do something no one else really does or doesnt do it right and then you can come up with better solution

[shadow ridge]

idk i think it's still original if you were the one who came up with it

i think multiple people can have the same original idea

like a tree with multiple roots

[twilit smelt]

pretty much any time ive judged a thing as superior to another thing it was not my own invention but just an idea from another existing thing

[warm pine]

it's all good and well if you actually invent something better but most people doing it just have a robinson crusoe complex, but invent something worse

[twilit smelt]

i mean i think in this case its just obvious the IO enqueue should have always been iterative and that the original NT designers just had a rare case of misdesign they failed to notice in time

[shadow ridge]

im talking about how you think something is badly implemented, try it yourself then realize that due to some X condition, the correct way is actually the way you thought was bad

[warm pine]

now and then you see people post a hot new OS concept to hackernews or wherever, and they disparage all of the things traditional OSes do

[twilit smelt]

the tpot grifter types on twitter would sometimes talk about OSes to their oblivious audiences who would praise them as geniuses and say like "waow... alternative non-unix os design ideas that might actually improve on things... welcome back terry..."

[warm pine]

and what they post is something silly, like an OS which has no supervisor/user distinction and only runs safe rust code

[twilit smelt]

and their ideas were invariably like

"what if syscall arguments were in the form of json so that LLMs can emit them and all the programs can be LLM native?"

[shadow ridge]

what the fuck

[twilit smelt]

real example of one i saw

from a guy lauded as a tech god with like 20k followers

[digital pivot]

do you think in professional os dev all ideas are just circulating and they don't invent new stuff but all the time borrow from someone else?

[twilit smelt]

i mean

how can you borrow from someone else something that wasnt invented?

they are inventing stuff its just not at an astronomical pace

[digital pivot]

and then like rotate

[twilit smelt]

they do constantly borrow from one another

[shadow ridge]

borrow implies theyll give it back

[twilit smelt]

there is occasional or regular direct communication between individual active contributors to every major kernel

linux, nt, xnu, the BSDs

some of which is just interpersonal

between pals

and they exchange ideas informally

in other cases they get the idea because its a well-publicized obviously good idea someone else just did

[digital pivot]

maybe but I think it's not really legal if they share confidential inventions

[twilit smelt]

i dont think they always care

engineers love talking about stuff theyre working on to ppl who know what theyre talking about

and are interested

and its extremely unlikely for that to actually blow back in their face

with something as low stakes as kernel design

where theres no like cutthroat competition on features really lol

since shareholders dont pay any attention to kernel design details

they get distracted by whatever Current Thing is

[digital pivot]

but officially I'm sure nt devs aren't even allowed to look at xnu sources and same for xnu

[twilit smelt]

like blockchain or LLMs

[twilit smelt]

why cant they look at xnu source

its open

xnu guys probably arent allowed to look at NT though yeah

since its not open

[shadow ridge]

at least techbros havent come up with blockchain syscalls

[twilit smelt]

i mean i have it on good authority that NT people regularly check up on what the XNU people are doing

which they probably dont need to hide

since as i said its open and legal for anyone to look at

theres nothing in xnu's license that says "except microsoft!"

[shadow ridge]

it's legal to reverse engineer it though

no?

[twilit smelt]

legal grey area i think

[shadow ridge]

if it was illegal, half of open source would go poof

i think it should be legal to look into how something you own works

[digital pivot]

because apple patents its products, and it's not safe if they start snooping around

[twilit smelt]

the license takes precedence

im fairly sure you cant release something under a free and open source license

then patent something it does

[digital pivot]

it's not as simple as just license

[twilit smelt]

and then that somehow retroactively stops anybody else from being able to look at that

[digital pivot]

i don't remember what it's called, but roughly speaking when you get a job you probably sign a special document that restricts such things

[shadow ridge]

does the license apply to the algorithms etc. too?

i thought it only applied to the code

[twilit smelt]

software patents are like fake anyway

[digital pivot]

it works the other way as well so that you wouldn't share private details as well

[twilit smelt]

nobody enforces them

theyre basically an award given to an engineer who had a good idea

[shadow ridge]

NDAs or whatever?

[twilit smelt]

sponsored by the company

so they can call themselves an inventor

and boast their legally recognized inventions

thats the only real use of software patents ive ever seen

there is widespread software patent infringement already that has gone unenforced

there are like basic algorithms used in every kernel that have been patented by IBM since like 1967

[shadow ridge]

it's not infringement if you dont even know the patent exists though

[twilit smelt]

someone patented the concept of a paging vmm

for sure

and that patent probably still stands

[digital pivot]

did they do anything with that information

[twilit smelt]

i mean

[digital pivot]

yeah stuff like that is more important than patents

[twilit smelt]

from what i heard, the unfair blocking lock revolution in NT was directly downstream of XNU doing that

[digital pivot]

no kelower confirmed on your server it was Microsoft research not xnu

he said microsoff research in California iirc

[twilit smelt]

theres something else i cant cite because it was sent to me once and then deleted where XNU was directly mentioned as the inspiration

and it came from a more authoritative source on the matter than kelowerirql

someone who worked on the locking more directly

at the time

[digital pivot]

#1062027635611422860 message

[twilit smelt]

no thats not an invitation for you to make guesses as to who it is lol

[twilit smelt]

oh yeah i mean

hes not even refuting the xnu involvement

hes just saying it was an MS research guy who got it off XNU

rather than an NT kernel guy

[digital pivot]

i understand it not like that but it's not my first language

[digital pivot]

i wonder who and how can be more authoritative source

must be cutler himself telling you he stole the nt kernel from apple

is it true??

[twilit smelt]

kelowerirql would agree this guy is more authoritative on that particular thing because he did not work there in the '00s when this happened

and this guy did and was directly involved in that work

kelowerirql is not the only NT person ive talked to online

[digital pivot]

ok it's not interesting who

but what?

what was the thing

[twilit smelt]

in any case kelowerirql seems to agree with me that xnu was involved

he literally says "thats true"

if he disagreed that xnu was involved then none of what i said would be true

and hed say "thats false"

[digital pivot]

he also said it's not right about windows

[twilit smelt]

he corrects the part where the xnu person talked directly to an nt kernel team member

and says instead the xnu person talked to an ms research member

im like 99% sure this is what he meant

dont bother him over this though

[digital pivot]

i wanted lol

[twilit smelt]

details like this are not things anybody should be pressing him about lol

thats why he left here

like 2 yrs ago now never to return

[digital pivot]

true true

[digital pivot]

let me ask if this thing from xnu is still in nt

[twilit smelt]

if what thing from xnu is still in nt

my understanding is that what they got from xnu was basically just the insight that lock convoys were caused by atomic handoff of locks, and that this happened some time in the '00s or early '10s

which changed their locking design strategy in the nt kernel

they didnt like

directly yoink any xnu code

[digital pivot]

i meant this

or you talked about the same convoy stuff

[hybrid condor]

Did they lie to me about age verification on discord?

[shadow ridge]

same i dont seem to be getting any prompts

and my settings are all set to allow nsfw

[hybrid condor]

I logged in to check and everything is working

[hybrid condor]

Keeping track doesn’t mean studying the code

I have a very rough idea of what happens in Linux, but I am not allowed to read its code

even though both projects are on GitHub

[twilit smelt]

i understand

i have to imagine it still happens though

for some

since apple/ms/whoever ultimately doesnt have ninjas staring at their employees through their window at home making sure they dont go to github.com/torvalds/linux

and the temptation would be there

[shadow ridge]

it's also not unlikely that youll see some little part of linux code somewhere

like many hacking movies / stock videos etc. have linux source code

i know because i checked some and it lead to a filesystem driver iirc

[twilit smelt]

i dont think anybody fears or is making a problem out of tiny snippets you accidentally see

[shadow ridge]

well true lol

what if you previously worked on linux? do they just not hire you for a kernel dev position?

[twilit smelt]

im p sure it only applies if youre currently working there

otherwise every student graduating from any worthwhile cs university would be unemployable as a kernel dev because their OS course looked at linux

[twilit smelt]

every reactos dev has probably accidentally seen a number of snippets of leaked NT code just because it inevitably jumpscares you when you get into researching NT

and its super unlikely that the fact this occurs could be used against them in a court of law

successfully

[twilit smelt]

also apple has tried to hire prominent linux kernel devs to work on xnu

for example: linus torvalds

so evidently, no

[shadow ridge]

who even is that? it's probably just a small contributor

[twilit smelt]

something funny in his book is that he mentions he went in for an interview with steve jobs and avie tevanian about working on the mac os x kernel around like 1999 and he says he wasnt interested because he thought their kernel sucked

and he makes the implication that he thought that because he thought it was a microkernel

like minix

and that as of writing the book he still thought it was a microkernel

so its possible he turned them down out of a misunderstanding

because he fell for the marketing when they kept calling it "the mach microkernel"

[digital pivot]

apple xnu?

for apple its a bit weird

because your system literally has other kernels stuff

like freebsds rcu thing

and turnstiles

[twilit smelt]

those arent copypasted

[digital pivot]

turnstiles maybe yes you could do without looking at the code

but freebsd one has very little information outside the sources

so you had to look at their code

[warm pine]

Freebsd is compatibly licenced while linux is under a gnu public licence

[hybrid condor]

yeah that's a different story

[digital pivot]

also i think itll happen just later

[twilit smelt]

I think they also changed their identity verification provider because the last one turned out to be a Palantir data harvesting op or something

That sounds made up but it's at least 30% true

[shadow ridge]

@twilit smelt you should open a matrix room for mintia

[flat dust]

persona or whatever is supposedly secure enough to be used by the US govt (certificate or something), and it like collects utter kilotons of data so its sorta a reasonable assumption they also work with the government

also ties with chat gippity

which are also used by the government

this feels like a conspiracy theory

[hollow musk]

if I have interpreted correctly Darwin can also donate QoS

it seems that their scheduler being heterogeneously oriented like Windows also relies on QoS

what's quite curious is that Windows PI doesn't donate (CPU) QoS, I don't know why

would be an interesting experiment

[mortal thunder]

the EU allows reverse engineering AT LEAST for interoperability purposes

if not more

[shadow ridge]

i find it hilarious that code is IP in the first place tbh

imagine if we treated math like this

where mathematicians invent new equations or whatever

and they "close source" it

and looking into how it works is somehow illegal

[warm pine]

that we have the current extreme situation around software is mostly a function of the fact that the software industry has a lot of money

if they did not then it would not have been possible to throw money at the finest lawyers

[digital pivot]

im sure theres a lot of math models kept private

and protected by law

[shadow ridge]

thats the dumbest thing ive ever heard of

[digital pivot]

well im not saying people own math itself, they can keep the implementation proprietary

[digital pivot]

what would that give ?

[craggy spire]

This must be intentional, as the QoS donation mechanism exists separately from AutoBoost (see WaitNextClearWobPriorityFloor)

Changing the priority (like AutoBoost's floor) can also trigger a QoS adaptation, in practice only if the new priority >= 15 (or >= 8 for system threads)

[twilit smelt]

im implementing kernel text mirroring right now

[twilit smelt]

thats where the read-only parts of the kernel and modules are replicated across each numa node

theyre mapped in the same place on each

but the physical pages they point to are different

ive implemented mirroring of the page tables used to map them

now i have to mirror the data itself

[twilit smelt]

ok its in

that took way longer than it should have cuz i kept getting distracted

[shadow ridge]

how does this mirroring work exactly

do you have like a list of physical addresses per virt page

or do you have a list of virt pages per node

[twilit smelt]

for what

[shadow ridge]

mirroring the page tables

or is it completely hardwired

[twilit smelt]

the main code that does it is fairly straightforward https://github.com/xrarch/mintia2/blob/main/OS/Loader/XRstation/_BlXrComputer.jkl#L1396

[shadow ridge]

ok so you dont have like a struct page

[twilit smelt]

i do but this has nothing to do with that

this is just the bootloader copying the text sections of kernel modules to the remote nodes and making their page tables refer to that instead

each node has a system process with its own system page tables, some of which are shared between nodes and some of which are not

[shadow ridge]

so it's hardwired

what if u wanted to do this in a more flexible way in the kernel

where the user can ask some pages to be mirrored etc.

then mark them not mirrored

or whatever

[twilit smelt]

wdym hardwired

[shadow ridge]

when you hand off to kernel, how does it know it's mirrored?

what physical addresses does it even store in its struct pages for the kernel's table?

or do you just have the kernel's page tables (your representation not the actual pts) duplicated per node

been a while since i did this stuff so my bad if my terminology is off / what im saying is nonsensical

[twilit smelt]

The actual page tables are duplicated per node

The mmu page tables

I should probably make a diagram of my setup for that

[shadow ridge]

i see

so how would you handle it if you wanted to mirror pages in the kernel for whatever reason

like the user wanting a read only section to be mapped mirrored or smth

and then later change that

or is that not something that's necessary at all

[twilit smelt]

It is possible to accomplish this by mirroring the process page tables per node but I don't think this is commonly done

[shadow ridge]

probably not very useful then

[twilit smelt]

in the kernel there ends up being a per-node system process

which contains the per-node kernel virtual memory

[shadow ridge]

that makes sense

does it scan the page tables to know that

or is it passed by the loader

[twilit smelt]

wdym "know"

why would it need to know

basically the kernel's text section appears at the same virtual address on each node

its just mapped to different page frames

this is so that threads that get migrated between nodes dont get scrambled

[shadow ridge]

it doesnt need to know the physical address?

[twilit smelt]

for one, what does it need that for

before i answer that question (which does have an answer)

[shadow ridge]

if it didnt, wouldnt it leak when u unmap it

not that youll unmap kernel sections

but in general dont you need to store it for this reason

also for DMA

maybe

[twilit smelt]

why would you need to DMA into kernel text pages

also if you need the physical address you can get it by looking at the PTE that maps it

[shadow ridge]

as i said

it's not that you'll ever do this stuff with kernel pages

it's more about consistency

[twilit smelt]

"consistency" is unnecessary here

[shadow ridge]

i see

[dense vigil]

btw people that did it for linux didnt find it boosted performance that much

[twilit smelt]

its basically free at runtime so i dont see any reason not to do it

they also might have done it wrong

[dense vigil]

i guess if u dont have runtime text patching

[twilit smelt]

i dont think this is done often enough for the performance hit of "you have to patch multiple copies instead of just the one" to be a big deal

am i wrong

[dense vigil]

yeah its not done often

it just increases the complexity of multiple text copies

[twilit smelt]

what i read about their kernel text mirroring was that they got a nearly 17% general speedup

and a 50% speedup in userspace programs that were highly numa-aware

like dbms

[dense vigil]

yea

[twilit smelt]

thats substantial

[dense vigil]

and it mattered more on other arches for some reason

on x86 it didnt do much

[twilit smelt]

also linux's numa segregation isnt as complete as mine

[dense vigil]

for synthetic tests still

[twilit smelt]

my whole node space thing with fully segregated page tables for the node-local heaps and whatever is not a thing at all on linux

which surprised me

[dense vigil]

i mean yeah

[twilit smelt]

in fact im not sure its a thing anywhere else although i could be wrong

[dense vigil]

but data often escapes node local

unless you dont allow thread migration or same address space to exist across numas

[twilit smelt]

using bits of the VA to determine which node a virtual address is local to might be novelish

saying that cautiously

just cuz im not aware of any other kernels that do that off the top of my head

[twilit smelt]

my plan for numa affinitization is to allow process-level affinitization and thread-level affinitization which affect the placement of process-wide structures and thread structures + scheduling respectively

by default a thread will be affinitized the same as its process

right now there is no cross-node scheduling or memory allocation

affinitization is not implemented yet

that is coming up after getting the basic mechanisms working

which is like 90% done

[dense vigil]

yeah i guess

I'm assuming you've seen this https://man7.org/linux/man-pages/man2/set_mempolicy.2.html

[twilit smelt]

well affinitization is implemented

i should say more like

non-affinitization is whats not implemented

that is, doing anything on a remote node when the local one is overloaded

i have dug trenches between the nodes and now i need to build drawbridges over them

[shadow ridge]

do you have to write a man page if you contribute new functions etc. to linux

ik thats like a shit question but im curious

what if you do some huge core thing

[dense vigil]

dont think theres such a requirement

manpages have their own maintainers

[shadow ridge]

i see

[twilit smelt]

its probably nice if you start one

i wouldnt be surprised if a large amount of linux documentation like the man pages is done by corporations like IBM or Google

[twilit smelt]

on numa machines i will probably not page the kernel text

2 complicated to implement and kneecaps some useful optimizations

and is basically useless

because on a numa system youre going to have enough ram that you dont need to worry about paging out some kernel code

[twilit smelt]

for example when switching between threads that both belong to different per-node system processes

you can stay on the cpu's local node's system page tables

rather than switching to those of the remote system process

if youre paging the kernel text on a per-node basis, this isnt really doable or is very annoying

[night needle]

when x86 numa

[twilit smelt]

when it run on x86

some mechanism stuff still needs to be finished

like supporting memory-only nodes properly

currently i have no idea what it would do with a memory-only node but im pretty sure it wouldnt be correct

wtf just tried it and it called me fat

[digital pivot]

do you maybe have some numa computer

[twilit smelt]

a real one you mean

[digital pivot]

yeah

x86 numa system

[twilit smelt]

no but possibly you can get older multi-socket numa mobos pretty cheaply

few hundred bucks idk

ones from like the early 2010s

could be a good development vehicle

[night needle]

i do

you can get x99 mobos for dirt cheap

[hollow musk]

I don't know about this one

could you explain?

also, I knew there must be an online community where people know more about modern NT than I do, even though searching the internet won't yield many results

[craggy spire]

I know from a very reliable source that the Wob mechanism doesn't solve regular priority inversions, but rather big.Little inversions

[twilit smelt]

The Wob Mechanism

menacing name

[digital pivot]

KiWobQosResponseRequired ?

is it like working over autoboost

or a different thing altogether

[hollow musk]

when you think about it, AutoBoost is also a bit silly name

[balmy zealot]

I heard about big.LITTLE inversion for the first time.

How is it related to classic priority inversion?

[twilit smelt]

probably when a thread on a high performance core wants a lock held by a thread on a low performance core

the latter should be bumped up to a high performance core

[balmy zealot]

Hm, big.LITTLE is more about thread placement, and if inversion is related to the choice of core type, then it is no longer a priority inversion as such.

Especially considering that the choice of core type (big or LITTLE) is unlikely to depend on just priority.

[blissful smelt]

where did you hear about it?

[balmy zealot]

Right here in this room.

the Wob mechanism doesn't solve regular priority inversions, but rather big.Little inversions

[blissful smelt]

oh i was thinking that there was some other place where it was discussed

🤔

[balmy zealot]

I don't know about that, as I don't open Discord very often.

I'm not sure I understand the point of solving this kind of inversion, as I think that it brings a lot of complications.

Background and low-priority threads are more likely to end up on LITTLE cores, so I think this situation can occur quite often.

Is it worth trying to move to a different type of core every time?

[twilit smelt]

https://github.com/nathan237/TrustOS/issues/8#issuecomment-4079602568

my effortful crashout at a cs major defending the largest pile of ai slop ive ever seen

[shadow ridge]

john kernel saves the day

[shadow ridge]

btw i love how he starts with "nobody else does this" then instantly has something that has been done before...

[mortal thunder]

im not sure AI has a gender

it/its are its pronouns

[devout geyser]

holy yap

but holy shit 264k loc of KERNELSPACE slop

[warm pine]

pub // SÉCURITÉ : Bloc unsafe — contourne les garanties mémoire de Rust. Vérifier les invariants manuellement.
unsafe fn jump_to_ring3(entry_point: u64, user_stack: u64) -> ! {

why is it in french

[devout geyser]

mmm

[warm pine]

and surely it should be bloc dangereux or even bloc non sur

[warm mural]

no, the french love use english words

[warm pine]

the academie doesn't love it

[mortal thunder]

It probably talks about the unsafe block below

[flat dust]

The personal attacks and armchair psychology ("AI psychosis", "script kiddie", "incapable of understanding") aren't constructive and say more about you than about me.
do they?

its the guy after your post

idk what it "says more" about hyenasky however the dude was certainly offended

do constructive arguments really matter in the case of your kernel being utter balls

[twilit smelt]

I wasn't even talking to that guy he's obviously batshit

[flat dust]

true

[twilit smelt]

I was talking to his defender

I'm not interested in reading whatever retort he wrote

[flat dust]

fair

[twilit smelt]

I lied I looked at it

guess what

it's AI generated

what a loser man

he fed it to the LLM and had it write him a comeback

mannn

[warm pine]

it takes some considerable guts to publish such a project let alone to advertise it as a great advance

TrustOS is the FIRST bare-metal operating system with a built-in real-time kernel introspection laboratory. No debugger attached. No external tools. The OS watches itself run — from inside itself.
what means this?

debuggers built into the kernel have been around since time immemorial

and Solaris lets you install hooks and probes everywhere from the kernel up to your app with DTrace!

[flat dust]

kernel debuggers might just be "add a bunch of prints explaining wtf is happening" at least in my case

[warm mural]

nah

dtrace is great

[craggy spire]

https://www.youtube.com/watch?v=ZrYqgyxrFiY

Bloody hell, what have I just seen

[twilit smelt]

AI psychosis

[warm pine]

the gen z counterpart to gen x and babyboomers hiring black americans to rap in their software advertisements

[twilit smelt]

I clocked this guy as being in his 30s

No gen z kid is this corny

At least not one outside a fandom

[warm pine]

i wonder why he decided to make an anthem

[shadow ridge]

nah man youre dumb af

how many register level bugs have you solved

NONE

this man dedicated his life to solving bugs and securing kernels no other kernel dev could do

and he's doing it FOR THE PUBLIC

[twilit smelt]

Register level bug

What does that even mean

[shadow ridge]

no fucking clue

[twilit smelt]

He doesn't even know what that means

I'm depressed

This poor guy

[shadow ridge]

either he's having a midlife crisis or he's like 5 years old

also the ai he used has probably been trained on your kernels and basically every other osdev here

[twilit smelt]

It doesn't show it

[shadow ridge]

i mean it needs the data to make something that at least works

and there arent exactly a lot of os projects online to train from

[dense vigil]

Linux, all the bsds, Gnu thing, Darwin, etc etc?

[shadow ridge]

ok maybe youre right

NT leak even LOL

[twilit smelt]

Currently the LLMs seem to think that hobby projects are silly toys unlike the ultra serious ToaruOS and various tutorials which are super high signal and robust and should be endlessly copied from

I do believe that if you ask an LLM to do osdev like this it'll primarily pull from the tutorials and toaru

[shadow ridge]

i dont really understand rust but this code seems to have parts that are like

written

but not used anywhere

like it just comments and says shit like "oh yea this thing does this" and it's not even correctly implemented and even worse not used

[dense vigil]

Iirc toaru kernel is relatively simple no?

Its mostly the userspace part that's impressive

Since it has custom everything including a python like language

[twilit smelt]

toarus kernel is pretty trivial and has a lot of bugs

[digital pivot]

iirc its also very buggy

[twilit smelt]

LLM kernel code tends to feel like if toaru was overgrown

if not in style then in spirit

[dense vigil]

Oh ok cuz that comment sounded like he was praising toaru for quality

[twilit smelt]

No I was being a bit sarcastic

[dense vigil]

Ah

[twilit smelt]

The only hobby kernel worth praising for quality is Minoca

We all live in its shadow

Its like will never be replicated

https://github.com/minoca/os

[dense vigil]

Tofita os or whatever

[twilit smelt]

I'm not being sarcastic about minoca

[dense vigil]

I remember that name but I forgor

[twilit smelt]

Minoca is like a gigantic ruined alien spaceship rotting in the middle of our primitive rice fields

It's always been there

Nobody knows who put it there or why

Or how it works

It's beyond us

[dense vigil]

What's so cool about it

[twilit smelt]

I mean just look at it

[dense vigil]

No KDE on screenshots, no Minecraft, no gdt init either

[twilit smelt]

It's probably like 27 hours of monkuous' attention away from running Minecraft

[dense vigil]

Wasn't it written by some nt devs or something like that?

[twilit smelt]

We don't know

It's all very mysterious

But it did iterative IO enqueue before that was even a sparkle in me or fadanoid's eye

So it was people familiar with NT's design regrets

[dense vigil]

Looks very nt like at least

[twilit smelt]

who weren't us

[dense vigil]

So the person who wrote it just disappeared?

[twilit smelt]

Seems like it

[warm pine]

it's in some ways an NT-Linux hybrid as i recall

[twilit smelt]

One interesting thing was that IRPs are full objects with permissions and whatever

[dense vigil]

What is iterative io enqueue?

[twilit smelt]

And starting an async IO operation gives you a handle to an IRP object

[twilit smelt]

wrong video

[dense vigil]

💀

[digital pivot]

mintia2 ke seems more advanced already

[warm pine]

https://github.com/minoca/os/blob/6750d31e02d4f6f745a88caa4dd05f2f672b8832/include/minoca/kernel/io.h#L2712 the filesystem for instance is very Linuxy, and the counterpart of struct path here reveals that

[dense vigil]

Oh yeah it even has a handrolled aml interpreter

I remember now

[twilit smelt]

iterative IO enqueue is where you build a request packet and pass it through each layer of a driver stack via an outer loop that calls each one in sequence iteratively

This is as opposed to how NT or XNU do driver stacking

where each level directly calls the next (recursive)

Creating kernel stack scarcity issues for larger driver stacks

[dense vigil]

Ah right

It has its own package manager it seems?

[warm pine]

the whole kernel and core userland is entirely original

[twilit smelt]

minocer is my best friend

[warm pine]

it's really quite remarkable

they don't even have byacc or gnu bison for instance, they have their own yacc-compatible parser generator

[twilit smelt]

it's def the best hobby os project done by someone outside this community

[warm pine]

the authors really were able to turn their effort to just about anything and produce a decent result

[dense vigil]

So it doesn't use gcc or?

[warm pine]

they don't go quite that far

[dense vigil]

Where's the libc located?

[warm pine]

https://github.com/minoca/os/tree/master/apps/libc/dynamic

[dense vigil]

Wait so they wrote their own parser generator and made gcc use it?

[dense vigil]

This seems like insane amount of work

[warm pine]

they use regular gcc and binutils, they just wrote their own yacc for whatever reason, it's used in something called chalk which i think is some custom scripting language of theirs

[twilit smelt]

I wonder sometimes if minoca was written with the intent of selling it back to Microsoft or something

and it didn't pan out so they just opened it

a lot of it seems like it is specifically and conspicuously addressing things about NT that bother the NT team

[dense vigil]

How many years did this project take

[twilit smelt]

There are probably a good number of ideas in minoca that originated as "if only we could..."s inside the NT team

[dense vigil]

I wonder if its better than mlibc

[twilit smelt]

A couple guys seem to have worked on it nearly full time for like half a decade

in the 2010s

[warm pine]

i'm inclined to think it was their get rich scheme

[earnest zenith]

144 FPS in the desktop

[twilit smelt]

The desktop of "TrustOS" having an FPS counter is something I noticed and is kind of a cherry on top of how silly it is

[warm pine]

everyone loves it when their OS, like a video game, redraws the entire desktop 144 times per second

[dense vigil]

oh okay that would explain it

its not really a hobby os I suppose

its an actual product

[twilit smelt]

well

We don't know if they were paid

and ultimately they just gave it to the world and abandoned it

[warm pine]

they got it mentioned at "embedded world": https://www.youtube.com/watch?v=bQ9z_1XtnOw

[twilit smelt]

Minoca is the only reason old mintia got paged pool

[dense vigil]

https://www.minocacorp.com/ ??

what was on there even

[twilit smelt]

Larp corp

[twilit smelt]

Cuz I'd been saying I probably didn't need it and then I saw minoca had it and I felt challenged directly

[warm pine]

i do wonder what became of these guys

[dense vigil]

yeah they're obviously crazy good

[twilit smelt]

something fun about numa support is that if you ever see a global spinlock you can instantly feel smugly superior

[dense vigil]

probably got a normal job

[twilit smelt]

but minoca is by and large much superior to mintia2 and will remain so for many years

I will catch up one day but only by default and after like 10 yrs

[shadow ridge]

at this pace more like 100 years ngl

[twilit smelt]

https://tenor.com/view/roblox-i-pledge-allegience-gif-732767766766283616

[warm pine]

latterly it looks like evan green was working for some startup called 'rivos' which appears to have been acquired by meta

[shadow ridge]

1. randomly get motivated
2. work on project for like 10 hours
3. motivation completely disappears
4. repeat some time in between 1 week and 2 months

[warm pine]

while chris stevens:

Chris Stevens is a Software Engineer at Databricks where he works on Apache Spark, helping to build a unified analytics platform for Big Data, Machine Learning and Artificial Intelligence. He previously co-founded Minoca, where he designed and developed Minoca OS – an open source operating system.

[twilit smelt]

we should find his email and send him an invite

[twilit smelt]

calling him a cofounder is a bit overkill

afaik minoca was only ever a garage LLC with two employees

I don't think they got funding

at any point

[dense vigil]

damn must feel bad to work on random stuff like that after building a product like minoca

[warm pine]

in this industry you overkill everything you say or you'll lose

[twilit smelt]

Maybe I should make an LLC

So I can call myself a founder

[shadow ridge]

limnAI startup when

[dense vigil]

why did they even decide to demo such a random project?:

[shadow ridge]

so wait why isnt minoca a bigger deal

[twilit smelt]

This is like standard fare for any Gen Z guy at any major cs university

[shadow ridge]

if it's so good

[dense vigil]

because its abandoned

[twilit smelt]

They literally all have an LLC and call themselves founders on their CV

[shadow ridge]

it's open source someone like john mintia could probably take over development

[twilit smelt]

I heard stories of college girls at like Stanford making fun of guys who didn't even have an LLC

[shadow ridge]

it probably takes like a single call to your accountant and maybe some gov office visits to set up like who cares

[twilit smelt]

It's about the window dressing not the substance

[warm pine]

https://www.youtube.com/watch?v=qp5hxHPlTq8

[shadow ridge]

also it costs money to register a company

[twilit smelt]

The former is much more important in high status fields due to the prevalence of status seekers

[warm pine]

apparently only fools & horses is now very popular in the balkans

[dense vigil]

uhhh lol

how do u imagine that happening

[warm pine]

there's certainly enough serbs in the comments there

[shadow ridge]

john mintia rm -rf's mintia2 and devotes his life to studying minoca idk

[dense vigil]

but its under gpl

u cant make a commerical product with it anymore

[shadow ridge]

oh

[dense vigil]

i mean you can but you will have to open source it

and sell support rather than the product itself

[shadow ridge]

static / dynamic libraries all count as derivative works right

[warm pine]

it's a really interesting generational shift

[shadow ridge]

so you cant just go "oh i built this thing but it's a library so gpl doesnt apply"

[dense vigil]

☠️

[warm pine]

for some previous generations i think wanting to make a lot of money was not well received at all

[twilit smelt]

It's disgusting

[shadow ridge]

says the guy who doesnt even have an LLC

like come on are you even real

[warm pine]

it's maybe a general change

[shadow ridge]

it's probably because of social media culture and influencers / startup bros all becoming more common

[warm pine]

there seems to be a sort of sublimation in reverse going on where the culture was focused on subtler things but now is focused nakedly on sex, money, and steroids

[shadow ridge]

SMS?

[shadow ridge]

like kids actually want to become influencers or some dumb shit like that when they grow up

[twilit smelt]

a lot of it is being propelled intentionally by "role models" like Marc Andreessen

there's a concerted effort to make it cool to be a sociopath?

the most recent volley we all saw was Andreessen claiming that no high value male has ever performed introspection or felt guilt

in history

which is wrong but the gen z hustlers loved

[shadow ridge]

i take pride in not knowing who that guy is

[warm pine]

this one is so rarted

[twilit smelt]

Extremely rich Silicon Valley investor

Major participant in the Palantir death complex

[warm pine]

it's insane hubris to think you are not an emotional being driven by certain irrationalities in the mind that we all have in common

[shadow ridge]

1. make a startup based on a seemingly novel idea / technology that is impossible in practice
2. wait for the VC money to flood in
3. live like a king while pretending youre working
4. declare bankruptcy when they start asking where the returns are

[warm pine]

it makes them extraordinarily vulnerable people since they won't even acknowledge the possibility they could be led astray

[shadow ridge]

optionally you can do like a side thing where you are an influencer tech bro online and sell advice shit

[twilit smelt]

These people complain about the collapse of western civilization. Dude you ARE the collapse of western civilization!

My favorite genre of radical is the one who knows their audience is full of morons who will salivate over tradition and so they just call their new psychotic radical ideas "traditional" and the morons fall for it hook line and sinker because they never knew what the tradition really was in the first place due to being uneducated

[warm pine]

incidentally andreessen hasn't made anything useful to the world since over 30 years ago when he last worked on the mosaic browser

[twilit smelt]

This is certainly an example of that

[warm pine]

since then he has made nothing but scams and junk

[shadow ridge]

arent they all like that tho

selling people lies and scams to fund a rich life style

[warm pine]

some believe it

[shadow ridge]

why live knowing youre doing bad for the world when you can live thinking you're doing good for everyone

[twilit smelt]

In this case he's pretending that sociopathy and a lack of empathy is trad and based

[shadow ridge]

like if you think about it, this really is just optimal

theyre basically living in their own world where theyre tony stark or some shit

it's like a sort of feel-happy matrix

[twilit smelt]

When in reality any of the great men he's claiming had minds that worked this way would likely have considered him a tragic freak

[warm pine]

he has to generate some narrative to be able to make sense of why he has such a privileged position in the world

it's necessary for his wellbeing

if he were to start being reflective about his life he would regret a lot of things and it would be a very painful process for him to get through it

[shadow ridge]

it's also kinda necessary for him to get money so it's better for him if he just thinks this is all great and serves some purpose

[twilit smelt]

I really liked this persons take on it

[shadow ridge]

the moon isnt made of cheese?

[warm pine]

so he is demonstrating this human need in a spectacular fashion by inventing lunatic conspiracy theories about strangers from '1910s-1920s vienna' (what might that mean?)

[twilit smelt]

Every right wing tpot freak immediately fell in line with his take and started e-warrioring in favor of it, they're the ministers in the analogy

[shadow ridge]

also are we sure this tweet isnt ironic

it reads like satire to me

especially with the "I regret nothing." at the end

[twilit smelt]

It's unfortunately definitely not satire

[warm mural]

ah andreessen is that the browser guy

[twilit smelt]

He's saying he regrets nothing because that was a follow up tweet to the controversy generated by him saying it before

[warm mural]

most browser guys seem to be assholes, like brendan eich

wait now there's andreas kling

there's a correlation I'm sure

[shadow ridge]

"It is 100% true that provably wrong thing. I regret nothing." sounds like something I'd say in a joke like

[twilit smelt]

Kling isn't really an asshole he's just lame

[warm mural]

yeah but ever since he's focused on the browser he's been getting more and more annoying and techbro-y

idk there's something with browsers that's messing people up lol

[shadow ridge]

browsermania

[ocean valley]

I keep in mind to never write a web browser then

[dense vigil]

I wonder what minoca devs would say about serenityos

Because it's a few orders of magnitude more popular

Btw there were corporations dming kling about running serenity os as like the server os

In production

Completely misunderstanding what it is and its state

[shadow ridge]

LOL

[twilit smelt]

source.

[warm pine]

somehow i am unsurpsied

the unikernel concept still has a following too even though it's absurd

kernels do a lot of work to optimise things which any unikernel has to redo from first principles

[mortal thunder]

This one's a hobby OS?

oh my god this thing has an arm32 port

[ocean valley]

Is it worth keeping in mind as a reference OS?

[twilit smelt]

Yes

It's quite high quality

[ocean valley]

coolio :3

[mortal thunder]

tbh minoca makes my own OS look like a toy

[dense vigil]

Either a podcast with him or his driving videos

[dense vigil]

Yep

[night needle]

what the hell

[twilit smelt]

"We are Charlie Kirk" coded

[night needle]

omfg yes

also

wtf does that mean

does he jerk off with the other hand

[mortal thunder]

maybe he has only one hand

[warm pine]

the three sources and three component parts of osdev:

1. debugging PCI registers
2. deciding the architecture
3. pushing through when nothing boots

[warm mural]

this is AI too

pathetic

[night needle]

"come up with a counter to this guy"

does he think he'll become bill gates by vibeslopping an os

does he think apple and microsoft don't know how their systems work

[blissful smelt]

will xrarch intend to support cluster-like features such as things like VAXcluster?

would be pretty cool to see mintia2 on cluster computers

was reading the OG lustreFS paper (I think it is the original) and came across them explicitly referencing VAXcluster as a source of their DLM design

[blissful smelt]

yeah they do have NT leaks in their training

[twilit smelt]

The DLM was designed and implemented by a woman. Just thought I would mention that for girl power factor

[blissful smelt]

I was curious on how much chat gippity knew about NT mutexes and it just caved under zero pressure and spit out the structure definition (early NT not push locks)

[twilit smelt]

Kathleen Morse

[warm mural]

even cooler: there's a quebecoise in the original NT team

😎

[twilit smelt]

VMS was the girl power kernel

[warm mural]

she didnt do very cool stuff tho

i think she worked on the login prompt

she's mentioned in showstopper

[twilit smelt]

DLM: Kathleen Morse
VMS for Alpha, director of kernel engineering: Nancy Kronenberg

And there was also uhhhhhh

There were like several other women on the VMS kernel team who I'm forgetting

It had the largest proportion of women of any kernel team I'm aware of

It was like 30% women

And led by a woman for a long time

VMS is woke

[balmy zealot]

Was Windows actually technically behind back then?

[digital pivot]

who

[warm mural]

idk idr her name

[twilit smelt]

It had/has some unfortunate flaws like any other design

Idk about "behind"

Depends on what axis

[warm mural]

she's mentioned in showstopper

[twilit smelt]

Are you talking about like Johanne Caron or whatever

[warm mural]

yes

thats it

https://tenor.com/view/scream-if-you-love-quebec-scream-quebec-crie-si-tu-aime-le-québec-québec-crie-gif-6243316274443352922

[twilit smelt]

You can tell I'm not obsessed because I was able to remember the name of a person that I haven't thought about in years from a book I read 5 years ago

[digital pivot]

isnt it published by msoft and u dont need leaks for that

in wdm even

[warm mural]

yeah

there's a website with vista struct definitions

[balmy zealot]

Was it so bad that it would make sense to switch to the MINOCA Corp. kernel?

[twilit smelt]

Probably not

It was just wild speculation on my part that they might have wanted to sell their new kernel back to MS

It's a cool thought but very likely not at all their plan

It's of the quality that I wouldn't be surprised if they genuinely thought of it as a possible replacement kernel for a major desktop OS though

It could be

with a few years of corporate scale polishing first

[mortal thunder]

tbf most kernel versions have debug symbols

[balmy zealot]

Considering that most of the kernel was written by a single person, that’s particularly impressive.

[tame phoenix]

I absolutely agree with this comment in isolation. you can't vibe code an os but you can use an LLM to increase your productivity by hammering out boilerplate, analysing large tracts of documentation etc but yeah it won't debug for you, it could decide the architecture but imho never ever let it. keep it on a short leash.

now I've stuck my 2p worth in, I'll go back and read the rest of the convo

[shadow ridge]

that guy's also using his disability as an excuse

as if AI is an accessibility tool

which is just bullshit and lame

[blissful smelt]

chat gippity stated that it came from NT leaks but ya I guess if it didn't say that, that would be reasonable

[twilit smelt]

i am interested in the xnu zone allocator

i will probably steal it

because its much better than mine

[digital pivot]

iirc its basically a slab allocator?

but they call slabs zones

[twilit smelt]

yeah but the implementation is cool

i disagree

there are obvious specific things you can pick out that were basically one to one borrowed in the first releases of NT

[warm pine]

opinions about lineages in software in general seem to have a habit of bouncing between two poles of "everything was ripped, it's as good as a copy" and "there was no similarity whatsoever"

[twilit smelt]

"nicholas wilt" was apparently (according to him) a hot shot in microsoft's OS division in the 90s and "would know"

ive never heard of him before

[warm pine]

maybe he once had to deal with people claiming nt was a direct reprint of the vms sources in c

[twilit smelt]

its funny because thats definitely false but theres a directly adjacent hypothesis which might just be true

which is that its a direct reprint of the mica sources in c

according to a (court appointed?) "MIT expert", it is

https://www.techmonitor.ai/technology/dec_forced_microsoft_into_alliance_with_legal_threat

Last year, somone from the Massachusetts Institute of Technology apparently found whole chunks of Mica comment for comment, note for note still there in Windows NT.

an article from 1996

[shadow ridge]

today people use AI to steal others' code and write boilerplate

in 1996 they just stole the boilerplate directly

is this really that much of an improvement

[twilit smelt]

he started crashing out when i replied mentioning some of the similarities

"carefully designed HAL to do cpu-specific things like edit page tables"

no...

that is not what that did

lol

[shadow ridge]

say it in the thread

also wait why does it say tweet

i thought they made it x or whatever

did they change it back

[twilit smelt]

i have an extension that de-naXifies it

[shadow ridge]

ah

for a second i thought i went back in time to a year where there was less stupid

[warm pine]

seems to be a quite common misconception

but rather nicely casts complete doubt on wilt's authority

[twilit smelt]

its annoying to get qrt'd by a much larger account basically implying im stupid and then proceeding to prove he knows less than i do

in a way that i have no way to point out

i could reply but nobody will see that

qrts are an evil feature

[shadow ridge]

what is a qrt

[twilit smelt]

quote retweet

basically he quotes my reply as a whole new full tweet all his followers can see

[shadow ridge]

just reply anyways who cares

alternatively you can delete twitter for a 61% mental health boost

[shadow ridge]

so he can cherry pick which responses he wants to show huh

[twilit smelt]

yes lol

i wont mention that there are some grounds that the NT team actually did do IP theft (of Mica)

theres at least one chapter of the NT workbook that was extremely blatantly written with the analogous Mica chapter open to the side

its like the exact same structure and sentence topics and just restated

that is, the object manager chapter

specifically the introduction

[dense vigil]

can u send a link

to the tweet

[digital pivot]

what would u like to borrow from it?

[blissful smelt]

https://x.com/CUDAHandbook/status/2035663645716230176?s=20

Nicholas Wilt (@CUDAHandbook)

This claim of IP theft by Microsoft has come up a few times in this thread. What the hell theory of harm does anyone think Digital could assert, here? Patent violations? There isn’t a hint of a sign that NT was anything but a clean room design. In fact the VMS kernel 1/x

X · Mar 22, 2026, 10:23 AM

[twilit smelt]

dont chime in lol

[dense vigil]

i just wanted to see u destroy the guy

[twilit smelt]

i dont really want to destroy him nor can i

twitter isnt where the more right person wins

i will lose and get ridiculed if i get in a catfight with this guy cuz he has like 100x as many followers as me

this is basic twitter self preservation

i dont want to have a twitter account linked to my irl identity where i just constantly get ratioed and made fun of by people with 100x-1000x as many followers as i have lmfao

bad look

thats what it would end up as if i corrected every wrong high follower account

[dense vigil]

if u have sources to back it up i dont see that happening

[twilit smelt]

it would

trust me ive been down that road

they would just qrt me like "You have 'sources'? Well I WORKED at microsoft. Kid." or something and then theyd have like 100 boomers and zoomer cs majors reply to them like "Wow you really owned that kid." and he'd go like "Yep sometimes just need to teach kids on here a lesson."

these things are highly predictable

ive gotten in arguments with people who were very proud of "being there" before where i had a primary source document from over their head at the time that proved their hazy memory and/or narrow viewpoint wrong and they basically never admit defeat

even on discord where the most right person usually does win

[shadow ridge]

get dave cutler himself to reply saying hes wrong or smth

[twilit smelt]

and especially not on twitter where they can easily own you by siccing their 10 gorillion followers on you

[twilit smelt]

another example was a guy who was at MS in the late '00s who got in an argument with me about whether Mach was any kind of inspiration for the NT team

he was absolutely insistent that the NT team had never even heard of Mach when they designed it

i showed him the NT workbook from the time which literally mentions Mach by name as an inspiration and he was like "this is an artifact of marketing dude. The marketers put this in here because Mach was hyped up at the time. I've had lunch with these designers. You're wrong kid"

which was just factually wrong considering the NT workbook was internal to the engineering group and was never a marketing document, but he wouldnt admit it

cuz he was too proud of having worked at MS and having worked with a few of these guys irl

even though it was like 20 years after the time period in question lol

along with conceptual inspirations like the whole concept of personalities in NT, it also has possible direct implementational inspirations

like the way the cache manager works in NT

is a mapped file cache

which was also used in... mach

and was a unique feature of both at the time

theres also a dave cutler interview where he talks about studying mach in the 80s and this guy still wouldnt admit he was wrong

one of the more frustrating internet arguments ive ever been in because everybody in that discord sided with him automatically cuz they knew him better, and he took advantage of that out of ego to avoid engaging with the argument and consigned me to public tarring and feathering for daring to challenge him instead

which is why im still slightly salty like 4 years later

[warm pine]

the scene illustrated

[twilit smelt]

lesson: never challenge an I Was There guy with a primary source document that only his bosses ever saw, he will kill you

then probably read the document and think "huh i never knew that" and update his perspective but only after tarring and feathering you

[hollow musk]

hi, is it known what part of the code that was?

[twilit smelt]

Ke, Io, and Mm are the most likely places to have originally contained direct rewrites of mica code because they are the components that changed the least from the original mica design, and were owned by the same people who owned them in mica

especially Ke

its known that Ke was started in early march 1989 and finished by april 1989

which means cutler would have had to write and debug like 15,000+ novel lines of code in a month, with the world's most asspain toolchain and the slowest simulator ever

i have always had a feeling that this isnt very realistic and he may in fact have rewritten large sections of mica's equivalent component

that were actually written over the previous 3 years at DEC, a much more realistic timeframe

using the world's second most asspain toolchain (the PILLAR/SIL toolchain) and second slowest simulator ever (the PRISM simulator, on MicroVAX workstations)

a lot of this is speculation but when new information comes out about those days it usually proves me right so.

it also would have:

1. been extremely tempting,
2. felt morally correct (i agree, for the record), and
3. been extremely easy
   to just reference binders of mica code they probably already had laying around their belongings at home because that was typical for DEC OS guys to bring home so that they could look at it in their off time and notice bugs or improvements or etc

as an aside i dont think NT code was ever printed in binder form like that

that was a thing DEC did since their earliest osdev days in the 60s and they did it up through mica's cancellation but i dont think the NT team ever did that

it sounds like all the VMS kernel guys had binders of the code that were regularly updated, with the intent they would take it home

there are many anecdotes involving these binders

i always thought that sounded cool to have so i thought about hiring a print shop to do that with mintia2's code when its done lol

by NT days the binder format would have been an unnecessary expense because a floppy disk would be much cheaper and sufficient to take home whatever code you wanted to look at at home

that was also likely true during mica but theyd maybe just keep doing it through inertia and that being something the engineers were used to

the hard transition to the microsoft work environment would have killed off the gigantic binders of OS code probably

[warm mural]

You should look into the freebsd one, it integrates with their safe memory reclamation mechanism and it's really nice

In the end it's all just slab allocators anyway

[twilit smelt]

doesnt xnu's do the same

[mortal thunder]

sounds like a waste of paper unless you reprint like once every 3 months or something

[twilit smelt]

they would replace only the parts that changed

[mortal thunder]

ah

i should do a paper print of my OS code at some point... but probably 3 columns of tiny text and only when i hit a major version bump like v1.0.0 or v2.0.0

[warm mural]

I don't know, but the freebsd one is more documented at least

Also idk if you've seen but they updated ULE quite a bit, they don't have multiple queues per cpu anymore

[twilit smelt]

wdym

[warm mural]

They used to have one queue for real-time, one for batch and one for idle (each has an array of PRIO queues)

Now they have one array of 256 queues from which they index

[twilit smelt]

@dense vigil i have sufficient toolchain features for doing percpu stuff as a duplicated section, it would just require bootloader support and for the percpu structs to be broken apart into "globals" placed in that section. i will pay you 1.23 chinese yuan if you do this for me (not usd that is worthless now that the petrodollar has been destroyed forever)

[dense vigil]

Tbh im not even sure what you're asking

[twilit smelt]

u guys were talking about doing per-cpu variables as a section of the kernel executable weren tyou

which is duplicated

during initialization time

[dense vigil]

Yep

Do u also want to get rid of aggregated structs for percpu with indirection?

[blissful smelt]

freeBSD UMA is really interesting

[warm mural]

Yes tho the names are confusing my bonwick brain

But I think buckets = magazines, zones = caches

[blissful smelt]

computer scientists are generationally bad at naming things

just part of the culture

[sterile frost]

isn't the per-cpu part called magazines? in which case the magazines would be caches

but tbf i did not read the paper so i have no idea what refers to what :^) i'm just going off vibes

basically a human llm, if you will

[warm mural]

basically in bonwick speak cache = the entity from which you allocate slabs

you create a kmem_cache for every object

it's called a cache because it's an object cache

magazines were added later on and they are per-cpu caches that are present in every kmem_cache

[sterile frost]

ahh

yeah that would make sense then

[warm mural]

I guess thats why they changed the names in freebsd because it is pretty confusing lol

I think zone makes more sense, though i like the magazine naming

[shadow ridge]

did you implement it

[twilit smelt]

oh yeah I forgot about that

[shadow ridge]

do you even work on this project

[twilit smelt]

stop being annoying about the fact im busy with life stuff