Astral

[grand shadow]

Yeah then I can make a flamegraph with it

[bronze nexus]

flamegraphs are cool

[grand shadow]

unrelated at all to my recent work but I should port an irc client

[grand shadow]

I got the first hex dump out of the profiler :D and its absolute garbage, gotta figure out what is going on

garbage as in, not even in the right format

much less having any useful data

bro wtf the nmi never actually writes to the ringbuffer 💀 where is it getting its data from

what was it cooking

wait fuuuck I forgot to disable the limine serial

[dense valve]

@grand shadow I have question, do you use serial or framebuffer to log? Both now and when you've just started. Thanks in advance, please answer TvT

[grand shadow]

When I started I did e9 logger until I had a pmm and then I added flanterm

I do support serial but I dont log to it

[neon crane]

I don't log much to the screen

Mostly to e9

[prime juniper]

konversation is decent

hexchat too, but unmaintained, deprecated and gtk2 only

assuming you want GUI ones

[night trellis]

@grand shadow have you built an iso bexause i am not building astral's gcc

i desperately wanna contribute to this

[robust geyser]

that's gonna be hard considering you need gcc to build anything for astral

but refer to this, i guess: <#1061407633745125397 message>

i don't know how out-of-date this is btw

[grand shadow]

should be mostly up to date

the one in astral-os.org is a symlink to that lol

But yeah, youll need the astral gcc

[grand shadow]

currently having exams to study for so things will keep being very slow for a bit

but after like the 8th (and after I finish one project... its a stupid one to write some software for dos using x86 assembly and masm) we ball

summer break will be upon me

[proper meteor]

took me a moment to remember you guys are in summer when we are in winter

[prime mulch]

no its actually just him

his hair grew backwards in the womb so he experiences the seasons in reverse

[proper meteor]

bro's heating his house at 35c during winter

[night trellis]

fuuuck

guess like ill have to compile then

[grand shadow]

its not that bad

unless you are like on a cpu thats 20 years old

which then good luck and I am sorry for your soul

#error "Here's a nickel kid. Go buy yourself a real computer."

[fleet osprey]

skullsob

wtf

where does that even come from

[grand shadow]

https://github.com/torvalds/linux/blob/master/include/math-emu/double.h#L29

[fleet osprey]

of course it's Linux lmao

[dense valve]

Lmao, wtf lol

[prime mulch]

it's a reference to a dilbert strip from the 90s which btw the author Scott Adams went down an insane conservative pipeline triggered by an extreme love for Trump and got his comic strip cancelled by saying a bunch of racist and sexist stuff

He's like stonetoss for boomers

[neon crane]

linus torvalds is gen x so kinda sorta checks out

[prime mulch]

I knew that particular strip because it was on the inside of the back cover of a big book about Unix that my dad had that I found when I was little lol

[grand shadow]

I will do a mlibc update either today or tomorrow

this will be a bit funky since I'll need to:

• fix any abi breakages in kernel
• rebuild the world
• solve build issues that arise from the ongoing mlibc namespace pollution changes

[dense valve]

Do it yesterday, jk

[scenic epoch]

do it before astral itself existed

before the gdt was created

before the 8086

well into the stone age

[dense valve]

We'll be Dr. Stone with this one 🔥

No, Dev. Stone. 🔥🪨

[scenic epoch]

make the flintstones know of the mlibc update

[night trellis]

i have an intel core i7 10700K

10th hen

gen*

so not that old ig

from 2020 since its a 10th gen

[dense valve]

me too, but mine are i5

[grand shadow]

we flamin

I might clean up the work I already have and push it so I can work on updating mlibc

no amd pmc but whatevs thats just hooking it up to the infrastructure

theres still some things to fix (like that syscall entry thing being separate

though there I can already see it spends a LOT of time on a munmap while running glxgears

(and thats with one core!)

[grand shadow]

fun thing is you might even be able to run the fucking flamegraph script inside astral

since astral has perl

and python

[marsh sparrow]

very nice

the flamegraph is very cool

I should extend Managarm's profiler to do that as well

Now that Astral has stolen Managarm's PMC code, I can steal Astral's flamegraph gen

[grand shadow]

so true

its a python script that just reads the binary file that is in a

number of pointers (1 byte)
pointers (8 bytes each)
...

format
and shoves all of them into addr2line, puts that into the format flamegraph likes and shoves it into flamegraph

later I can get a longer flamegraph of glxgears and see if I can see anything (as well as fix some of the weird symbol naming) but for now I gotta get the ball rolling on the mlibc update

[scenic epoch]

do it then

#RunTheFlamegraphScriptOnAstral

[grand shadow]

bet

[scenic epoch]

holy crap xml or something

[grand shadow]

damn links doesnt support svg 💔

[scenic epoch]

awh

i feel like running this on a real computer for some reason (i did it with some other os already)

depends on the system requirements i dont want to sacrifice my actual pc for this

[grand shadow]

beware the spooky ps/2 driver that does not work

[scenic epoch]

ohno

what driver does work

[grand shadow]

abandon all hope ye who enter here

[scenic epoch]

none 😔

[grand shadow]

like everything else BUT ps/2 ☠️

[scenic epoch]

why cant yall fix PS/2 then

[grand shadow]

well wanna know a fun thing

[scenic epoch]

interrupts or smth???

[grand shadow]

on my machines it works

its just on LITERALLY EVERYONE ELSE'S THAT IT DOES NOT

so I cant debug this shit

[scenic epoch]

buhr

does it work on qemu

[grand shadow]

yeah

[scenic epoch]

idk what machines you have but i hope i get lucky bcs im gonna do this on a laptop and i assume its gonna emulate ps/2

1.8 gb root image 😔

[grand shadow]

that wont work for booting on real hw

[scenic epoch]

yeah ik

[grand shadow]

you'll need the initrd iso

the root is just the filesystem for it to mount into

I need to start shipping unified images ☠️

[scenic epoch]

yeah i read the qemu command thing its the hard drive crap

[scenic epoch]

possibly

emulation time™

UPDATE YOUR LIMINE

https://tenor.com/view/drunk-tipsy-slip-gif-13181667

[scenic epoch]

just copy paste it

it mightve killed itself a little

the next line didnt screenshot it says timeout

[grand shadow]

u running without the root disk

the initrd minimal thing is minimal

[scenic epoch]

is command in the github evil

[scenic epoch]

of course its minimal its minimal

im not running the initrd thing tho im running the top one i assume its what im supposed to boot

talking about the boot screen

did i not mount smth

[grand shadow]

oh right, I forgot that now I literally only ship the minimal distro

you'll have to install stuff

xbps-install xorg-server etc etc etc

[scenic epoch]

https://tenor.com/view/medic-tf2-team-fortress-2-team-fortress-dance-gif-14842990257268326643

explains it

[grand shadow]

xbps-install -S xorg-server fvwm3 xterm etcetcetcetc

run netd vionet0 btw

to start the networking

[scenic epoch]

ohok

sudo doesnt seem to work by default do i have to install it

it seems to be there its just not owned by "uid 0" (root?)

[grand shadow]

wait who is it owned by

[scenic epoch]

i forgot to say it "doesnt have the setuid bit set"

also i ran stat on it it says uid0

eror™

[grand shadow]

well that was supposed to be ran by the init hook

[scenic epoch]

RIP

[grand shadow]

did you by any chance close the vm without syncing the disk?

or running poweroff

[scenic epoch]

yes

[grand shadow]

[scenic epoch]

ok i guess thats the problem

how to fix

[grand shadow]

get a new image

or if u dont want just set it to suid but you might get spooky filesystem errors

[scenic epoch]

bugstral

[grand shadow]

since ext2 is very corruption prone on dirty shutdowns

[scenic epoch]

[scenic epoch]

fair

i redownloaded image and it still dies

[grand shadow]

does it say anything about running the hooks

because like, it works on my machine™

[scenic epoch]

i may be blind but no

ykw im running it from root MUAHAHAHAHAHAAHAHA

logout sharted itself

acpi shutdown time

[grand shadow]

you sure you not running the same image? it is literally supposed to do that on the first boot

[grand shadow]

yeah I dont start login in a loop

I'm downloading the github image to see if that happens on my end

[scenic epoch]

im running the one from november 3rd

idk its on yo github

[willow rapids]

november 3rd is quite a while ago

[grand shadow]

I think I fucked up and have a bad image uploaded (I think I opened up it for testing and forgot to copy a new one)

[scenic epoch]

oh

[grand shadow]

lemme see if theres a good one in astral-os.org

[scenic epoch]

https://tenor.com/view/medic-tf2-team-fortress-2-team-fortress-dance-gif-14842990257268326643

[grand shadow]

I think its the same 🥀

I can upload a new one rq

[scenic epoch]

broski

tfw developer no test thing 😔

(and or dementia)

[grand shadow]

I think I just opened it and closed without syncing

which is funny becaues thats stupid

[scenic epoch]

i was going to say "did you dirty shutdown your own os" 😭

[grand shadow]

yeah

[scenic epoch]

:(

[grand shadow]

https://astral-os.org/releases/

[scenic epoch]

UPDATE!!!(128612789io487128941278918972487129487421142

[analog berry]

how did u implement it?

[grand shadow]

use pmc to get an nmi every set amount of tsc cycles -> record backtrace onto a ringbuffer that is shared between a per-cpu kernel thread and the nmi -> these threads funnel it into a single global ringbuffer which is exposed to userspace in /dev/prof -> the astral-prof program opens it, discards what was there and starts copying data from it into a specified location (be it a file or /dev/com0 or whatever) -> you can now use a python script to generate the flamegraph from that profiler data

[analog berry]

daamn

[scenic epoch]

what???

[grand shadow]

very inspired by the mangarm profiler

[grand shadow]

wtf spooky

the image I uploaded is 1.9gb

[scenic epoch]

evil

ok now it says 1.99

safari crash out

it blackscreens

https://tenor.com/view/drunk-tipsy-slip-gif-13181667

no booting lololol

[grand shadow]

works on my end

[scenic epoch]

https://cdn.discordapp.com/attachments/1406827148068782171/1436584810914254980/how_exquisite.gif

[grand shadow]

I'll add e9 logging to the bootdisk hold on

[scenic epoch]

k

[fleet osprey]

Certified moment

[scenic epoch]

[prime geyser]

yes 😩

[grand shadow]

version up

[scenic epoch]

downloading

e9 is serial?

[grand shadow]

its debugcon

[scenic epoch]

i forgot thats the qemu one

ok so pure silence

literally nothing

ok im gonna go get my linux pc out

the mac gods dont like me rn

[grand shadow]

not even limine?

[scenic epoch]

limine works

everything else sharts itself

[grand shadow]

nothing on e9 is crazy tho

stuff gets printed on e9 like first thing that happens almost

[scenic epoch]

shrug

[grand shadow]

maybe the version of limine I was on didnt play well with it or something

[scenic epoch]

EVIL limine

[prime juniper]

That would be good to do performance shit

[grand shadow]

anyway, mlibc bump and world rebuild ongoing

hopefully not too much shit I have to fix considering the infinity abi breaks and pollution fixes

[grand shadow]

super close to hitting 30k loc

oh fuck I gotta update gcc so I can update mlibc

thats fun

[fleet osprey]

Btw how architecture (and notably bootloader)-agnostic is Astral

I'll be getting some time soon and I might do some work on that PPC port that you hinted at wanting to do sometime Soon™

But if it's still heavily tied to x86_64+Limine then I'll probably hold off until it would be less painful

[grand shadow]

isnt ppc the one with like inverse page tables or some shsit like that

[fleet osprey]

Yes but there's also the BATs that suck less

But you get way less of them

[grand shadow]

as much as I have tried to be arch agnostic astral really likes running with an hhdm and there are probably many x86isms that need to be ironed out

and theres no bootloader abstraction

[fleet osprey]

Oof

I'll probably hold off on it then until some of those are ironed out

[grand shadow]

yeah maybe until I already have it ported to another arch like riscv64 or aarch64

[fleet osprey]

Oh yeah that'll likely entail DT parsing which would be a huge help if it already existed

[grand shadow]

oh god thsi is gonna fucking suuuck

gcc 15 using std23 by default is already causing build issues and I am not even 2% done

🥀

if I keep getting packages broken becauaze of this I am dropping to gcc 14

yeah no I am dropping to gcc 14 I cba

its either adding -std=gnu17 to almost every package or updating almost every package

[grand shadow]

ok, have gotten a minimal distro to build with gcc 14 and latest mlibc

time for the rest of the owl

xorg is down

codeberg and freedesktop websites when it comes to downtime

https://tenor.com/view/aokiji-akainu-battle-two-side-gif-10500515882138008504

[midnight stone]

How long does it take to compiler

[robust geyser]

depends on your computer

on a high end machine it'll probably take a couple hours

of course if we're talking about building a full distro

[night trellis]

building astral

wish me luck

i kinda want to make an astral distro

like a lightweight one that doesnt take a bazillion years to compile

just gcc, kernel and userland

and probably pacman for package management

damn it only took a few minutes for me what

anyways what do i do now>

oh yeah

oop nvm

still needs to compile perl

what?

do i have to shell hop again

[grand shadow]

Oh yeah edit the perl recipe to remove that bash call

I forgot it there 💀

I recommend waiting a bit to do a full world rebuild

I am working on updating mlibc

And so when I do youll have to rebuild everything

[analog berry]

@grand shadow btw did profiling reveal any bottlenecks

[grand shadow]

Havent taken a look yet, Im gonna do that once I finish updating mlibc

But from that first flamegraph u can see that munmap takes a ridiculous amount of time

[night trellis]

oki

[analog berry]

probably unlikely minecraft spams those

[grand shadow]

mesa probably does tho

thats a glxgears flamegraph

[analog berry]

iirc its the glxgears itself spamming it

[grand shadow]

oh bruh

[analog berry]

do u profile across multiple cpus or how does it work btw

like is it a kernel wide profile or process wide

[grand shadow]

kernel-wide all cpus

I could maybe get away with telling it to profile only a process

but for now this works

[sharp carbon]

this is the glxgears loop code

plus some time accounting and the X event handling

[analog berry]

immediate mode opengl is interesting, its kinda impossible to tell what that sequence of calls is going to result in

what does glCallList do? Looks like it references some predefined list

anyway i hope whatever minecraft is doing doesnt result in a mmap/munmap per frame

like even linux cant do more than ~50k unmaps a second

[sharp carbon]

you define a list of vertices/texcoords/etc once and then you can draw it in a single call with glcalllist

[analog berry]

ah so kinda like vbos etc

[sharp carbon]

yea

[sharp carbon]

well apart from the latest few major versions mc mostly uses immediate mode sooo

[analog berry]

wait really?

[sharp carbon]

yeah

[analog berry]

damn

[sharp carbon]

even with the "use vbos" option enabled, it makes a new vbo each frame, draws it once, then destroys it immediately

[analog berry]

how do they manage to use custom shaders and stuff if they use immediate mode?

[grand shadow]

[analog berry]

☠️

[sharp carbon]

iirc with custom shaders they force the "use vbo" behavior i described

but it's been a while since i looked at that part of the old mc codebase so i could be wrong

[analog berry]

thats crazy inefficient

[sharp carbon]

to be clear i'm talking about 1.8 and the like

in proper modern versions they use actual decent rendering techniques

[analog berry]

i guess notch wasnt that great at graphics programming

[marsh sparrow]

another Managarm win

ci-boot.py[63]: stdout: memory mapping, size = 1 MiB
ci-boot.py[63]: stdout:     134676 iterations per second
ci-boot.py[63]: stdout:     134743 iterations per second
ci-boot.py[63]: stdout:     134529 iterations per second
ci-boot.py[63]: stdout:     135607 iterations per second
ci-boot.py[63]: stdout:     135869 iterations per second
ci-boot.py[63]: stdout:     avg: 135084, std: 544

[willow rapids]

[analog berry]

Damn

[night trellis]

whats managarm

[marsh sparrow]

https://github.com/managarm/managarm

[grand shadow]

new mlibc world build is up

cc @night trellis

[night trellis]

tysm

[grand shadow]

flamegraph of running minecraft on astral

why the fuck is the page fault handler taking that much even tho vmm_pagefault is a tiny spec

a better question would be

why is it running after interrupt_set?

ah wait

the stack contains the thing AFTER the call

but still

I should fix that and have it point to the call

ohhh I dont add the rip into the log

https://tenor.com/view/assembly-assembler-programming-despair-gif-21635424

me when I am me when me

[grand shadow]

well shit

I think its time for me to stop being lazy and give astral an actual non shit vmm datastructure

me when linked list vmm

guess sooner rather than later I gotta implement rbtrees

wait why is vmm_getrange under pfisr

think I'm missing a call somewhere

still, that is a looong ass time in getrange

I need a real data structure for this

I could have an avl tree which is piss easy to implement but thats boring and rbtrees are cooler

(plus its what everyone uses so its probably better)

cc @analog berry since u wanted to know what was bottlenecking mc so much

oh yeah I forgot to + 1 the backtrace size

[analog berry]

So basically bad vmm?

[grand shadow]

pretty much

it spends most of the time in getrange()

which just iterates over the linked list

[analog berry]

Yeah thats pretty bad

[grand shadow]

since I'm gonna be on a big break soon I can probably get a rbtree implementation working before the year ends

that should at least double performance maybe triple

[bronze nexus]

I'm guessing this is for finding free ranges?

[grand shadow]

getting the range of an address

[bronze nexus]

ah

[grand shadow]

oh it gets inlined that explains it

[bronze nexus]

So you're saying it might not actually be getrange?

[grand shadow]

no it is getrange

the stack trace just shows pfisr because its still with the vmm_pagefault rbp

[bronze nexus]

Ah, the second graph would be inlining disabled.

[grand shadow]

like if you look here, getrange should be above vmm_pagefault

but its not because it gets inlined

[bronze nexus]

I'm gonna try astral at some point, lmk if you want to see it running off a USB on my framework laptop ;)

[grand shadow]

bet, maybe at one point when I get ps/2 fixed for real

[bronze nexus]

I guess that's a good way to check your trackpad support

[bronze nexus]

IIRC framework has a USB trackpad

[grand shadow]

oh

that will take a while

I do not have anything usb at all

[bronze nexus]

lol

I can do you one of my old PCs then, will take me a bit to find PS2 mouse and keeb though

I'd try running it on my main if I didn't have the principle of never running a hobby OS on something I don't want filesystems corrupted on :P

I'm off now. Keep up the good work.

[gritty shore]

AVL is faster in practice for lookup-intensive workloads iirc

But rbtrees are better in general

[grand shadow]

yeah

[grand shadow]

I need to impelemt inline unwinding in the addr2line call in the flamegraph script

[deft oak]

you should try caching the result of getrange

[robust geyser]

or… try using a data structure better suited for frequent lookups

better than caching linked list lookup results

[marsh sparrow]

AVL vs rb is not that important

[robust geyser]

anything but a linked list tho

[flat copper]

left-leaning rb trees are much simpler than avl

[deft oak]

well yea do that too ofc

but caching should still help

since its quite common to lookup the same range over and over

[dense valve]

What do you use for that rn?

[grand shadow]

linked list, which is not really good for this

I plan on writing a red-black tree implementation

[dense valve]

Ahh, well yeah that's not good

[dense valve]

🆗👍

[wraith surge]

how did you get this flamegraph?

[grand shadow]

use PMC to generate a nmi every X tsc cycles
that nmi takes a backtrace and stores it in a per cpu ringbuffer
a per cpu thread copies the data from that per cpu ringbuffer into a global ringbuffer
that global ringbuffer is exposed to userspace in /dev/prof
there is a userspace utility that is used to copy the data, discard old data etc etc and copy it into a file (could be a file or say /dev/com0)
then there is a python script that takes that file, puts it in a format flamegraph wants and generates a flamegraph out of it

[wraith surge]

oh ok

thx

[neon crane]

why did you decide on a linkedlist

[wraith surge]

i'm using a linked list too for vmm

lol

[grand shadow]

That was like 3 years ago and I was lazy

[opal lotus]

i use a bump allocator

[neon crane]

for what

VA allocation I guess?

[opal lotus]

yes

[neon crane]

i see

if you have LA57 support thats probably fine

[winged basin]

Its surprisingly common to use a linked list for map entries

Mach vmm did so and it was inherited by its descendents

With at least a last map entry faulted on cache however

So for the typical case just test if that's the one you faulted on. This approach worked well for a long time

[grand shadow]

interesting

[opal lotus]

"worked" implies it doesn't anymore?

[grand shadow]

I assume address spaces were smaller + less fragmented back then

[proper meteor]

I use a rbtree for vmm

[hearty fulcrum]

same

an interval tree

[worthy mica]

@grand shadow how does astrals init route work? I'm thinking about making one as im kind of tired of having 10k init calls in my main func

[grand shadow]

Macro soup that takes a name, flags, function and a list of dependencies
It loops over a special section that has pointers to init routines, and goes calling them and their dependencies

[grand shadow]

currently trying out os-test stuff on astral

it git cloned from gitlab and is compiling now

[grand shadow]

compiled, getting it to run

so far only one crash (syscall_accept() code did not check for an implemented socket op and thus call to zero)

and a hang on "udp/bind-socket-shutdown-r-sendto-first-poll"

[dense atlas]

ohhh woa I like it

I might wanna adapt it to support modules depending on modules, although I don't need that yet because modules can't link to other modules currently

https://media.discordapp.net/attachments/1177852959820890224/1184984343978975262/ezgif.com-video-to-gif_2.gif

[opal lotus]

i have that

but i don't use extra code to encode those dependencies

i just have the dependencies in the dynamic section of the modules

the compiler/linker creates them for me

[dense atlas]

where do you call them?

[opal lotus]

wdym where

[dense atlas]

do you not have dynamically loadable modules?

that's what I meant

[opal lotus]

i do

i specify the modules i want to link against in the build system, and the dependencies get saved like regular shared objects in userspace

the kernel doesn't do automatic dependency resolving though, it's up to the user to load the modules in the correct order

the kernel will simply refuse to load a module if a dependency is not loaded

[dense atlas]

omg sorry I confused you with mathew 😭😭

I'm sleepy I see

[opal lotus]

💀

[dense atlas]

It would be hard for the kernel to resolve the dependencies anyway

[opal lotus]

yeah

[dense atlas]

since it doesn't know where they are

[dense atlas]

I want to give the user the choice to make e.g. a driver modular or not, similar to how Linux does it

in fact I am using Kconfig :3

[opal lotus]

i did that before

but rust is kinda annoying in this regard

i can't use a build system of my choice

[dense atlas]

oh yeah, rust really does want you to use their stuff

I had to deal with it with my Makefile mess, as I have support for rust

[opal lotus]

i want to use meson but i'm being held up by stupid issues

[dense atlas]

I actually quite like my makefile mess, it's a mini kbuild

[grand shadow]

I had done an oopsie when compiling the os-test and now after recompiling it I run into more issues, so I will do more debugging of it later

[sharp carbon]

I don't really see the point of that as long as you have a way for the user to pass boot modules

[dense valve]

same

[dense valve]

[dense atlas]

it really depends on the platform, you may just simply not want the overhead of loading the extra modules instead of linking them right into the kernel

or there's simply no option for the bootloader to pass any modules that way

sometime my ESP8266 port will get somewhere..

[grand shadow]

okay so schedule is:
finish uni project
do some os-test work
then write out an rbtree implementation

[grand shadow]

negative refcount

happened because I forgot to check if the fifo vnode was already bound to a pipe before unbinding it

woo that seems to be the only remaining hard issue on os-test

time to shove those results into linux and see how astral did

astral better than macos real

(this is missing the locale stuff by leo)

cc @timid jetty

wrong leo ☠️

cc @formal patio

hm most of the udp tests fail because connect return ENOTSUP on udp

that sounds like a easy way of bumping up the score

0% on pty is crazy tho

all ENOSYS

tf am I missing

oh wait

wasnt openpty turned into a sysdep or something

well thats another place that probably can get more % on

seems like a lot of the sys_socket code did not work because ENETUNREACH

huh it does seem like I have most of the infrastructure needed in udp to support udp connect() its just that I dont have udp_connect() proper

[bronze nexus]

UDP connect is kinda an oxymoron if you think about it

What with a protocol that doesn't handshake at all

[grand shadow]

might be because of unlockpt

[grand shadow]

well it serves the "set the address to send datagrams to" purpose ig

[bronze nexus]

Sure

[grand shadow]

#define TIOCSPTLCK 0x13376167

actually hm, I'm seeing that unlockpt() is a noop on like bsds

ah whtaver I can make it a no-op ioctl kernel side

or just check that its the master pty

[grand shadow]

<---- forgot to add POLLHUP to his ptys

[dense valve]

How do you get those?

[worthy mica]

os test

[dense valve]

ok

ty

[worthy mica]

https://sortix.org/os-test/

[opal lotus]

leet 61 67

what about _IOC

[proper meteor]

I feel like if I try os-test on maestro it would be a disaster

[hearty fulcrum]

https://tenor.com/view/doja-cat-star-wars-gif-25078126

[proper meteor]

maybe

[viscid flame]

how to do os-test

[proper meteor]

I guess you need to download it from https://gitlab.com/sortix/os-test but gitlab is dead apparently

[opal lotus]

don't you need sshd

[proper meteor]

idk. If that's the case then I cannot run it

[opal lotus]

at least it says on the website

@formal patio confirm

[formal patio]

not necessarily

ssh is only preferred if you want sortie to update os-test results on the website

you can run it without ssh locally

or you can do the managarm/redox strategy of allowing sortie to download os-test result tarballs from CI

[opal lotus]

well i'd just use os-test for my own purposes

is it hard to run locally?

i remember you talking about some memes with html or something

[grand shadow]

Well the tests are fine to compile and run

The big issue is generating the html

[opal lotus]

why?

[grand shadow]

You have to do that on the host

On astral with 12gbs of ram it still SIGBUS'd the html process

[opal lotus]

wth?

[grand shadow]

Too much memory

[opal lotus]

what is it doing that needs that much memory?

surely this can be fixed

[grand shadow]

No idea

inb4 every character is in a linked list or something

[formal patio]

html generation is fine on managarm tho

there's just no point to doing it on managarm when I can do it on my host, where I also have other results available, and where I can view the results, too

[opal lotus]

how does one do that? does it like print intermediate results to a file?

[blazing atlas]

imagine acpi microwave turn on support

[grand shadow]

oh skull

I have a leak somewhere then

or it does some allocation on loop because it never seems to print anything

[formal patio]

os-test writes the results to files, and I can just shut down managarm, mount the image and copy that out

[opal lotus]

ah that's easy enough

i'll add ext2 support very soon anyways

[proper meteor]

‼️

[grand shadow]

considering doing a switch over to xbstrap at some point

since newer jinx does NOT work on my environment for some reason

theres a bug on jinx that has been happening for the past 3 months

[opal lotus]

one thing i'd miss from jinx would be functions

just being able to do meson_configure is pretty neat

[grand shadow]

I dont use that so I'd not miss out on much

[opal lotus]

it saves about 10-20 lines per recipe

[gritty shore]

you could do your own stuff as well

shouldnt be that hard right?

[grand shadow]

gonna use ebuilds

I do wonder how difficult it would be to use ebuilds

it would be very fucking funny if I did that

switch out xbps for portage

port openrc

and become gentoo 2.0

and in like 5 years when astral is actually fast I can mog on the haters by using portage without binary packages

but for now, I'm gonna finish the linux-mlibc-os-test project using xbstrap to get a feel for it and later I can see if I can do an ebuild mlibc linux thing to also get a feel for that

[marsh sparrow]

Using ebuilds would probably not be that difficult

Using nix may also be an interesting option

iirc @tribal dust was doing that basically

[sage locust]

have you considered trying to figure out the bug instead?

i do not have the time to effectively juggle all the projects i work on, sorry

all you did was ping me with an error message that i have never seen, without instructions for me to be able to reproduce it

idk wtf you wanted me to do about it lol

not even so much as to tell me basics about your environment like what your distro is, etc.

i know you're way smarter than this Mathew, so if this is just about looking for an excuse to fully bite into the managarm ecosystem, for whatever reason, do so

you don't really need to give me an excuse, unless you're trying to give yourself an excuse

[opal lotus]

use rpm spec files

[prime mulch]

I think he should be banned immediately.

[grand shadow]

it literally is just running ../jinx rebuild * on wsl ubuntu 24, it doesnt even get to actually building any packages (but it is after the debian container is set up)
I remember adding a mkdir or something got jinx further #1061407633745125397 message but it still died somewhere else

I'll try actually debugging it then

I still prefer jinx to xbstrap because yesterday I realized writing yaml sucks 💀

[opal lotus]

lmaooooooo

[marsh sparrow]

because it's yaml or because of something else?

[grand shadow]

it has the same issues with python wrt identation and like shell scripts are just more convenient

[opal lotus]

build: |

[grand shadow]

which is why I was gonna start looking into ebuilds but thats probably another kanker into itself

the last known good commit is 0.6.7

https://tenor.com/view/cat-67-scp-67-funyuns-funny-gif-75413186200073602

I dont see any big changes other than how debootstrap is called so hm

dawg I read my own message wrong

jinx 6.7 is not the good commit its the first bad commit

running on 3 hours of sleep 🔥

[prime mulch]

https://tenor.com/view/rigby-cat-rigby-cat-cat-stare-rigby-tongue-gif-12118654088387148905

will astral ever tackle numa

[carmine swallow]

I can build astral using jinx 6.8 just fine (after changing build-if-needed + the needed jinx version + setting debian snapshot version)

or well apparently I can't because of some m4 meme (might need a bit older snapshot idk) but like at least it started building

[grand shadow]

I think its some wsl meme

since on anything else it works

[grand shadow]

maybe

idk

[carmine swallow]

also what would be nice would be having some kind of minimal desktop package set

[grand shadow]

yeah I need to do that

probably just figure out some minimal set that can boot into fvwm with xterm

or maybe that has all the xorg shit + fonts + some other packages

yeah debootstrap is not creating /usr/local

[gritty shore]

lowkenuinely a great idea

[grand shadow]

bro FUCK ME I'm trying to open the debian wiki about deboostrap but wiki.debian.org is down

🥀

[proper meteor]

skill issue

[elfin turtle]

time and place

[robust geyser]

unless you're running wsl1, wsl is just a normal linux vm with some fancy stuff to make the user experience better

so i highly doubt it's anything to do with that, especially considering it works for me™

but i never used the weird ass ubuntu distro

[elfin turtle]

are u running it on a mounted ntfs

[robust geyser]

no

[elfin turtle]

cuz for me those have been fucky asf

[grand shadow]

I'm not that insane

[robust geyser]

you can see the path is ~/...

[elfin turtle]

good

[robust geyser]

and the errors are /home/mathew/...

so yeah

[grand shadow]

ok I have a clue

somehow the second stage bootstrap is not being run

theres no errors logged which sucks

[robust geyser]

you can try using set -x

to see all the commands being ran

but it might be a little verbose

[grand shadow]

can just set -x before the run_in_container1 "$JINX_CACHE_DIR/sets/.image" /debootstrap/debootstrap --second-stage and set +x after

[robust geyser]

that's not gonna do much considering it unshares and executes a new shell instance

can you probably try doing sh -c 'set -x; /debootstrap/debootstrap --second-stage'

[grand shadow]

            set -e
            set -x

            ld_preload=/nochown.so             home=/root             lang=en_us.utf-8             lc_collate=c             path=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin             ld_library_path=/usr/local/lib64:/usr/local/lib:/usr/lib64:/usr/lib             debian_frontend=noninteractive             "/debootstrap/debootstrap" \
"--second-stage" \
            # keep this whitespace
        ' > dbstraplog &> dbstraperr```
this runs but dbstraplog and dbstraperr are empty

env -i path= chroot
chroot not found

fym

that path= does not look like it should be empty

tho nothing in jinx sets ${path}

[prime mulch]

hope you do, its fun

[grand shadow]

wait why is everything inside this string lowercase

what the fuck is this some curl bullshit

@sage locust sorry for bothering you again but I think found what was the issue: I forgot to pass --raw to curl I am actually stupid ☠️

TIL that curl does http shit if you dont disable that

[viscid flame]

https://tenor.com/view/whatsapp-gif-21302024

[grand shadow]

wait wtf that was not it??? what

[viscid flame]

nothing ever works

[prime juniper]

y4 macro’s?

[robust geyser]

nice green role bro

[marsh sparrow]

why would jinx care about what flags you pass to curl?

[prime juniper]

It’s an upgrade over white for sure

[robust geyser]

i think mathewnd meant that when you download jinx using curl it might have screwed with the actual script

[grand shadow]

for some reason downloading jinx using curl made everything inside the strings do lowercase because it does some http processing bullshit

[grand shadow]

dont diss green bro 💔 I read that wrong

[marsh sparrow]

lol

[robust geyser]

http does not do any processing btw

[grand shadow]

then idfk why curl made everything lowercase but it did

[marsh sparrow]

Maybe it's doing processing based on the mime type

[robust geyser]

iirc jinx distros now switched to git cloning the jinx repo and copying out the script

[prime juniper]

Skill issue bro

[grand shadow]

I'm running on 3 hours of sleep and a few peanuts

smh

[grand shadow]

sounds wasteful

[gritty shore]

just thought about this

port system using lua when

[robust geyser]

you either do that or ship a copy of the jinx script in your repo

[prime juniper]

Bro I’ve been working since 6am (now 13:47)

[robust geyser]

i think that's what gloire does

[gritty shore]

nah bash is pretty much ideal for this but whatever lua is cooler

[robust geyser]

yeah https://codeberg.org/Ironclad/Gloire/src/branch/main/jinx

[gritty shore]

or python

[marsh sparrow]

Just gotta write a yaml generator in lua, then you can use xbstrap with it

[gritty shore]

my biggest gripe with xbstrap is yaml

i hate it

i think ebuilds may be a good alternative though

and I will try them in the future

• I get free gentoo recipes

[prime mulch]

xrbuildtool

[marsh sparrow]

Since xbstrap now supports preprocessing you can in principle use any language that can emit json as a recipe lang for xbstrap :^)

[robust geyser]

wait until you realize the amount of patching you'll have to do to have them build for non-linux non-glibc platforms

[gritty shore]

wait until you realize most gentoo packages run on freebsd too

[robust geyser]

Why the fuck

but still

[gritty shore]

gentoo freebsd is a thing

[robust geyser]

again

Why the Fuck

[gritty shore]

ah its not anymore

anyway

[robust geyser]

thank god.

[gritty shore]

but that shows its portable

[robust geyser]

yeah, i guess, overlays are a thing

i think arsen made an mlibc overlay or something lol

[gritty shore]

overlays are repos

[robust geyser]

i wouldn't know

sounds fancy

[gritty shore]

no it's literally just a name for a repo

[robust geyser]

yeah i know, but for the people who dont know it sounds like it does more than just provide packages

[gritty shore]

but anyway if the portage repos were ported to freebsd I'm pretty sure its doable to do it for your own OS

[grand shadow]

cant u remove the gentoo main overlay and use your own instead or something

[gritty shore]

yes

[robust geyser]

but then it's not gentoo anymore

it's just portage

[gritty shore]

yes

but you can still steal ebuilds from the gentoo repos

[robust geyser]

boooring

go big or go home

[gritty shore]

you literally do that with managarm tho

[marsh sparrow]

It's not surprising that ebuilds run on FreeBSD, is it?

It's just shell scripts

[gritty shore]

yea I know its not

[gritty shore]

but like this isnt really true

[marsh sparrow]

Whether the recipes actually work is a different story:^)

[grand shadow]

1. xbstrap and steal from managarm
2. jinx and steal from gloire
3. ebuild and steal from gentoo
4. pkgbuild and steal from arch

[gritty shore]

unless you wanna run linux specific software

[marsh sparrow]

nix

[gritty shore]

xbps and steal from void

[marsh sparrow]

I wanna see if nix is feasible for a hobby OS lol

[robust geyser]

xbps-src is dogshit

[viscid flame]

am i only who steals some stuff from arch pkgbuild

[robust geyser]

it's like worse pkgbuild

and pkgbuild is already questionable

[gritty shore]

afaik ebuilds are pretty much the best

for that kinda stuff

another option would be to use netbsd's port tree

[marsh sparrow]

ebuilds still have the disadvantage that they are shell scripts

[gritty shore]

how is that a disadvantage?

[robust geyser]

how is that a disadvantage

[marsh sparrow]

nix is side effect free

[robust geyser]

just run it in a container bro

[marsh sparrow]

xbstrap is also side effect free

[robust geyser]

then it's side effect free*

[gritty shore]

nix is an overengineered way to do shell + container

[gritty shore]

how? it still runs shell commands in the end

[marsh sparrow]

It's still not deterministic though

[marsh sparrow]

It knows what it will run in advance though

[gritty shore]

I don't understand the issue

like, jinx is pretty much deterministic

and it's a glorified shell script in a container

[marsh sparrow]

That's why it's really easy to support distributed builds, for example

which is what xbbs does

[gritty shore]

phd level script

I still don't get why shell scripts dont work

is it because you have finer control over when commands get executed?

[grand shadow]

brother

    unshare --user --map-root-user --mount --pid --fork -- sh -c "
        set -e

        mount --rbind --make-rslave /dev '${run_in_cont1_root}/dev'
        mount -t proc proc '${run_in_cont1_root}/proc'
        mount --rbind --make-rslave /sys '${run_in_cont1_root}/sys'
        mount -t tmpfs run '${run_in_cont1_root}/run'
        mount -t tmpfs tmpfs '${run_in_cont1_root}/tmp'
        mount -t tmpfs tmpfs '${run_in_cont1_root}/var/tmp'

        mount --bind '${apt_cache}' '${run_in_cont1_root}/var/cache/apt/archives'

        env -i PATH="${PATH}" chroot '${run_in_cont1_root}' /bin/sh -c '
            set -e

            LD_PRELOAD=/nochown.so \
            HOME=/root \
            LANG=en_US.UTF-8 \
            LC_COLLATE=C \
            PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin \
            LD_LIBRARY_PATH=/usr/local/lib64:/usr/local/lib:/usr/lib64:/usr/lib \
            DEBIAN_FRONTEND=noninteractive \
            $(for arg in "$@"; do echo '"'"$(echo "${arg}" | sed 's/"/\\"/g')"'" \'; done)
            # keep this whitespace
        '
    "

see that "${PATH}" there?

[marsh sparrow]

Shell scripts work in principle if you have the discipline that everything is 100% deterministic

[grand shadow]

see those double quotes?

${PATH} is being expanded outside that string

thats fucking everything up

if thats the issue its gonna be so funny

[marsh sparrow]

do you have spaces in PATH or in cwd?

[grand shadow]

yes wsl does

[robust geyser]

wsl path includes windows path too

lmao

L

fix your windows PATH bro 🙏

[grand shadow]

[marsh sparrow]

lol

[grand shadow]

mfw "Program Files"

[robust geyser]

oh, mine also does that

/mnt/c/Program Files/NVIDIA Corporation/NVIDIA App/NvDLISR

crazy

hm then that shouldn't be an issue i guess?

[grand shadow]

running jinx for real now lets see

DEBOOTSTRAP SECOND STAGE RAN

https://tenor.com/view/rivulet-rain-world-slugcat-dunk-ballin-gif-7684234335110226051

guess I gotta create a codeberg account now

[hearty fulcrum]

https://tenor.com/view/but-heres-the-baller-niko-the-kicker-oneshot-dunk-gif-26469380

[feral depot]

i like to look at this channel at random at times and just be surprised at how random it is

true random

[gritty shore]

pixel art genuinely impressive

[sage locust]

you can just send me a git format-patch formatted patch

thanks for looking into it <3

[grand shadow]

np lol

I already made an account so I'm gonna open the pr when I'm done testing it

ya ok its working

[grand shadow]

https://codeberg.org/Mintsuki/jinx/pulls/13 :3

[sage locust]

merged it, tyvm

space in $PATH is evil 💀

[flat copper]

yeah it's a wsl2 issu

[marsh sparrow]

I'm surprised that this doesn't cause even more issues

[flat copper]

i've worked around it a while back by disabling the feature

because i've had other issues with it in the past

i think with pip or dune

[marsh sparrow]

I'd expect many of the make based build systems to break

[prime juniper]

Real men steal and rewrite in yaml

[prime juniper]

No I look at them too if needed

[grand shadow]

is there any way to disable it redownloading the whole tarball on a jinx rebuild? I am spamming the gnu mirrors downloading gcc over and over again ☠️

[sage locust]

it doesn't do that after it succeeds installation once

i make it succeed once by stubbing everything

aka comment out all steps

then it should not do that anymore

[grand shadow]

I see, thanks

[sage locust]

np

[grand shadow]

cooking up an rbtree insert implementation while running on 3 hours of sleep at 11 pm, what could go wrong

[gritty shore]

wdym cooking up

you guys arent translating the "introduction to algorithms" pseudocode to C ?

[grand shadow]

Tfw you actually want to understand the code you wrote 🥀

My dsa class did not give any pseudocode for rbtrees

At most we got the trees

And even then the slides arent complete and fail to mention some stuff ☠️

Plus it makes me feel smarter knowing I was able to actually implement that

[gritty shore]

well the pseudocode is far from complete

[grand shadow]

In my class my professor always gave out implementations for shit like linked lists in the labs 💀

Like if u cannot write a linked list u not built for this

[gritty shore]

stuff like rbtrees aren't conceptually complicated

So they're not hard to understand

[grand shadow]

Yeah

[gritty shore]

The pseudocode just makes it easier to implement

imo

[grand shadow]

Once u understand them doing it without any pseudocode is also easy

[gritty shore]

yeah that's how I did AVL

rbtree in particular I implemented a while back and didn't bother

[grand shadow]

Do u use both in your os or did you swing towards one over the other

[gritty shore]

both

The funniest data structure I have is HAMT

That's conceptually very simple but the implementation is a bit tricky to get right

[grand shadow]

I think I remember you mentioning that before

[gritty shore]

I haven't gotten to the point of using it yet

Nor AVL or rbtree

[grand shadow]

Inb4 30k line hello world like infy

[gritty shore]

nah it's around like 8k total atm and it has a scheduler

[grand shadow]

Oh nice

Ule I assume?

[gritty shore]

yeah it's a fixed version of my fancy scheduler

Seems to work pretty well

now I haven't worked on stuff for like two weeks because of finals but I'll get back to it soon (I was in the middle of implementing turnstiles)

[grand shadow]

still working on that rbtree

had a busy weekend but now I'll have a bunch of free time

wonder how big of a speedup will rbtree + caching the last accessed range per thread would be

especially on mc where like 70% of the time is spent on range search

[limpid trail]

When you say caching the last accessed range, do you mean that you just store it as a pointer in your vmm context? I assume that you could just check that first during a pagefaukt before iterating through all ranges?

Because that’s hella smart + simple

[grand shadow]

probably storing it in a per thread variable

since each thread will be faulting different addresses

[limpid trail]

Oh true

But same idea

Can I yoink that idea for my vmm

I still use a linked list for my ranges and I don’t feel like switching to something better rn so I kind of need the free speed where I can get it

[grand shadow]

its not like I am the owner of ideas or anything

[limpid trail]

Bet

Also I’ve said this before I think but your project is so sick

2030 year of the Astral desktop?

[grand shadow]

thansk

more like 2100 :p

[grand shadow]

seems like I will have to add a container_of macro to astral finally

[grand shadow]

first test run of the rbtree (with 100 nodes):

I triple checked the insert function to make sure everything is ok so maybe I made an error in the check function where I did not triple check it

thats an issue for tomorrow me

its 3 am

[grand shadow]

looking at it its only after 2 inserts which is not enough for a black height violation

so my check is broken

[grand shadow]

ok, that was a check bug

now I think I actually hit a bug in the insert

I think its a bug with the left rotation

something is causing a right pointer to become null

when it should not

[grand shadow]

update: rbtree insertion works

can insert 100 values and look them up and they all pass the check

[prime mulch]

the 101st value is what gets you

[analog berry]

Yeah id add test cases for every rebalance case at least

[marsh sparrow]

If you insert a few few 1000 random elements, you'll hit each case anyway

[analog berry]

also removal is easy to fuck up

[proper meteor]

Don't remove stuff. Reboot machine when out of memory

[wild marsh]

replace removed element with tombstone :^)

[elfin turtle]

the discord way ™

[pine ledge]

Reboot the human using it, the design is very human

[scenic epoch]

release mustard gas (resurrect the body afterwards)

[dense atlas]

@grand shadowI am currently taking "inspiration" from your VFS code to maybe possibly finally come up with something somewhat sane for my kernel and I wonder if you can break the devfs

unfortunately I'm at school and the wifi is too slow to download the astral image, so I can't test it rn

what happens if you

• mount a devfs on at /dev
• mount devfs again (which is the points to the same files as the first one right?) at e.g. /mnt
• mount a tmpfs on /mnt (on top of the devfs)

Now if you look at /dev you should see the tmpfs instead I think???? until you unmount /mnt again

Just guessing, I may also just be completely wrong as all this is from me just studying the code

[grand shadow]

Yeah that is kind of expected, as it mounts the same vnode

I dont really support bind mounts and traditionally you cant mount the same filesystem twice

So this is more so a hacky way of supporting mounting devfs to multiple places

[dense atlas]

I thought so, okay

[dense atlas]

hehe

silly :3

[grand shadow]

2025 is almost over and so I need a proper 2026 goals list for astral
hmmmm

• compiling a distro inside itself without crashing, even if a minimal distro
• procfs, maybe
• audio stack
• finish the installation guide or write a install program

maybe throw in a webkitgtk there, but I feel like thats way too ambitious

[fleet osprey]

Debian GNU/Astral when

it works on Hurd and the BSDs....

(or at least, did, at various points in the past)

honestly not completely far fetched

[grand shadow]

💀

anyhow, havent worked on this these past few days

[prime mulch]

numa

astral numa 2026

its calling you

[grand shadow]

https://tenor.com/view/squidward-squidward-shocked-squidward-tonight-tonight-squidward-shock-gif-23057510

stop summoning demons smh

[proper meteor]

wtf is that

[prime mulch]

numa

[proper meteor]

https://www.youtube.com/watch?v=YnopHCL1Jk8

[prime mulch]

each vertex on those hypercubes is 2 numa nodes

4 processors

in an SGI Origin 2000 from 1998

thats the network topology between them for cache coherency messages and so on

1 computer

[neon crane]

i always remember this song whenever i hear talk about NUMA

its so funny

[opal lotus]

https://tenor.com/view/headphones-gif-4714741

[dense valve]

WOW

[dense valve]

NUMA?! Pretty Nice reading SRAT and SLIT

[dense valve]

How lol, I don't get the reference

[neon crane]

One of the lyrics of the song go "Nu mă, nu mă iei", which sounds like "Numa Numa Yay!" in English

thus, the "Numa Numa" song

[dense valve]

LOL

[neon crane]

and thus the connection I keep drawing unintentionally with NUMA (non-uniform memory access)

[dense valve]

I only connect it with MY YAHOO

Lmao

MY YAHOO HO (Ma-ia-hoo, ma-ia-haa)

[night trellis]

honestly i wanna make an astral distro called Tidal Axis with its own package management system called axpkg

and make wayland support for the astral kernel

[grand shadow]

If u want I can make the kernel build standalone from the rest

[grand shadow]

gyatt rizzmas or however the cool kids say nowadays

[willow rapids]

i can't do that yet :(

[dense valve]

Software Framebuffer or GPU (even virtio)

[viscid flame]

btw what is this wm

[sage locust]

fvwm

[elfin turtle]

istg u can take any random 1-3 letters and postfix wm and it probably exists

[grand shadow]

67wm

[elfin turtle]

kill me

rizzwm

[scenic epoch]

or de

wmde

dewm

except it would be like a desktop environment instead of a window manager

[grand shadow]

rbtree update: inserting 1000 nodes and removing them does not break the rbtree properties (tested after each operation)

I will do more specific tests like checking if succession is ok and then maybe do a test with 10000 or 100'000 nodes

[grand shadow]

ok predecessor and successor checks seem to be working

iterative rbtree is almost done :D just gotta tidy up the code

[grand shadow]

50k worked, 100k didnt work because rand is not random enough to not repeat stuff :P

[prime mulch]

even if it were perfectly random it wouldnt be that shocking to get a duplicate in 100k

if its a 32 bit rng

its not LIKELY but if you saw it every once in a long while it shouldnt be surprising

[grand shadow]

time to begin converting the vmm to using the new rbtree

[grand shadow]

one last function remaining, changemap

[night trellis]

ya :3

[grand shadow]

sure let me finish integrating the rbtree into the vmm and I'll do it

(and rewriting the worst function there while at it)

[grand shadow]

oh thats a bug I never noticed

[night trellis]

ohh

lmao

[grand shadow]

when splitting a mapping and in the remap case I dont set the middle file offset properly

[night trellis]

do you mind making the package manager my own instead of xbps

axpkg and axports

[grand shadow]

should've been new_offset = range->offset + range->size

https://tenor.com/b0czE.gif

[night trellis]

astral is unix-like iirc, no?

[grand shadow]

the thing about xbps is that it is super convenient to use with jinx, as jinx already creates xbps packages for you

[grand shadow]

it is

[night trellis]

im not gonna use jinx

in tidal axis development

[grand shadow]

ah then you can just use whichever package manager you would want

[night trellis]

or at least for axpkg

[grand shadow]

@sharp carbon when you added the arch_extracontext_detect() function, did you try it on tcg as well? it seems like tcg really does not like the mov to cr4

[sharp carbon]

I don't think I did, and looking at the code again now, the xsave presence check is indeed incorrect: it should check if CPUID[1].ECX[26] is set, not if CPUID[0].EAX >= 0x0D

[grand shadow]

I'll check if that fixes it in a bit

currently working on the rbtree stuff

got to a bash prompt using it

went thru all the whole init and first boot process too

so it seems to at least be working for most cases

fvwm boots

time for the trvth nvke

oops, didnt work but thats because of a lazy assert I added for a very specific case (unmap on an empty space)

damn java doesnt work

🥀

spooky backtrace

[grand shadow]

meinkraft got into the main menu, creating a world to see if it works

too many debugging logs to get an idea of improvement tho

ran into the font meme I dont have the patch for

https://tenor.com/view/spongebob-screaming-spunch-bob-spunch-bop-spongebob-sad-spongebob-gif-26152448

actually I forgot if thats a patch I need or just an unclean shutdown after handling fonts which causes a failure on the next shutdown

[grand shadow]

new minecraft flamegraph

the lookup funkyness has been eliminated

didnt notice a lot of perf improvement in mc though

I also need to figure out that isr_14

because it only happens in the brief time the very start of the page fault isr is being executed

but its taking up so much time that idk if its the cpu taking too many cycles to handle a page fault because kvm in wsl or what

I still gotta do the range caching thing

oh wait my laptop also isnt connected to power

💀

[robust geyser]

building astral rn, i wanna see if i can write an xhci driver

[grand shadow]

Yooooooo

[robust geyser]

i don't promise anything but if it works out it'd be pretty sick :^)

i hope you have msi(-x) support, otherwise i'll be very sad

[grand shadow]

I do

I dont have any support for the acpi pci interrupt pin thing

But I do have uacpi :P

[robust geyser]

that's fine, i'll use msi(-x) since that's what xhci spec requires iirc

or i might be wrong, who knows, not me since i haven't looked at the spec in god knows how long

but even then, having a shared interrupt handler is probably fine

[opal lotus]

[robust geyser]

bashline.c:65:10: fatal error: builtins/builtext.h: No such file or directory
   65 | #include "builtins/builtext.h"          /* for read_builtin */
      |          ^~~~~~~~~~~~~~~~~~~~~
compilation terminated.```

lol what i just reran make and bash built just fine, wild

[grand shadow]

Wtf

Bro got bit flipped

[robust geyser]

yeah i'm not sure what that was about lol

but it's all good now

i wrote a qucik gen_compile_db.py script to create a compile_commands.json, should help writing any code considering i don't know the codebase at all

i'll start working as soon as everything is built, i think i'm building a full distro so that might take a second

[formal patio]

wait astral doesn't use meson/cmake?

I don't know enough about jinx to figure out how astral gets built

[grand shadow]

Just a plaon makefile

[grand shadow]

And yeah by default it builds a full distro, you can change it in the makefile using the DISTROTYPE variable (set it to minimal)

[formal patio]

bruh

[formal patio]

how would you like a PR that changes that to meson

[robust geyser]

if it aint broke...

[formal patio]

handwritten Makefiles are always broken

at best, it's just (yet) undiscovered brokenness

[robust geyser]

meson has it's own set of issues

[robust geyser]

not a problem, i can wait, a wm will be useful to test usb mouse anyway :)