Astral

[grand shadow]

either way, @analog berry merged tysm

[robust geyser]

yaaay i can use irq->priv now :D

[analog berry]

thanks, lemme know if something breaks

[grand shadow]

I should change the other drivers to use that 💀

[analog berry]

ngl biggest achievemnt of that pr

[robust geyser]

unironically i was surprised it was missing

i wanted to ask if i can somehow store a context variable per irq

but then i looked at nvme driver

and i saw it uses a fucking hashtable

[analog berry]

lol

[robust geyser]

lost my mind for a second but i just adopted it

[analog berry]

@grand shadow btw is it expected that reset fucks up the console?

[robust geyser]

are you typing reset from xterm?

or the console

[grand shadow]

how fucked are we talking

[robust geyser]

inb4 sleep memes

u should log the syscalls

[grand shadow]

lemme see if reset fucks up xterm

[robust geyser]

[submodule "kernel-src/io/acpi/uacpi"]
    path = kernel-src/io/acpi/uacpi
    url = https://github.com/UltraOS/uACPI

bruh

[analog berry]

yes

[robust geyser]

y no keep things consistent

[grand shadow]

I plan on moving kernel stuff to submodules soon

[analog berry]

^

[robust geyser]

thank god

[grand shadow]

from the jinx regenerate thing

[robust geyser]

good idea

but kinda annoying

[grand shadow]

because then I can git clone and compile the kernel

reset works on xterm

maybe flanterm meme?

[analog berry]

so according to syscall log

it does ppoll like usual

[robust geyser]

reset should work on flanterm

[analog berry]

and gets response

but then doesnt work

like the screen is not updated

[robust geyser]

yeah ur right

thats weird

iirc reset works just fine on flanterm so..

maybe astral console/tty driver does something weird

[analog berry]

btw why does exiting vim not restore the initial console state?

[robust geyser]

i dont think flanterm implements alternate buffer stuff

[analog berry]

oh

[robust geyser]

or secondary buffer

idk what the name is

[grand shadow]

yeah it works fine on xterm

the buffer stuff that is

[analog berry]

i see

[robust geyser]

there is some weird escape sequence that lets you switch to a different buffer

so when you start vim/nano/less

[grand shadow]

which is weird that it works on xterm, /dev/console and /dev/pts/* go through the same driver

[robust geyser]

it doesnt mess everything up

[analog berry]

flanterm bug?

[obsidian pivot]

isn't xterm that one shitty pseudo console for X

terminal emulator

[grand shadow]

its the one with "Abandon all hope, ye who enter here" on the readme

[obsidian pivot]

[analog berry]

Btw why not port whatever managarm is running?

[obsidian pivot]

just like my fat driver

[analog berry]

it feels like a way nicer gui thing

or do u just like this

[grand shadow]

I Am Not Porting Wayland

[grand shadow]

yeah I like it

[analog berry]

why not?

[grand shadow]

needs way too many linuxy apis than I'd like to implement

[analog berry]

unfortunate

[grand shadow]

maybe with uGpu

with your drm stuff

[analog berry]

do u know which specifically?

like drm stuff?

[grand shadow]

drm and evdev and some others

[analog berry]

then yeah ugpu will definitely come in handy

[robust geyser]

reset works on my flanterm terminal emulator

[grand shadow]

huh weird

[analog berry]

astral bug

[analog berry]

are u on latest master

[robust geyser]

yes

ef07a10cc38b34aa003d17be97a9f3542e275069 for flanterm

[analog berry]

hmm

[robust geyser]

which is the commit from 3 months ago or smth

[grand shadow]

inb4 it needs callbacks or something

did the callback stuff even make it out of the limine tty stuff

[analog berry]

callbacks for what?

[robust geyser]

my callbacks dont do anything

they just log stuff

[analog berry]

@craggy tide why not?

[robust geyser]

probably didnt think it was needed

also idk waht the proper name for that is

so sorry if im wrong

[analog berry]

kinda annoying to have my vim shit in my console

after i close it

[robust geyser]

just ^L

[analog berry]

i wonder if thats a pun

like taking an L

[robust geyser]

its also broken here, so its not an astral meme :^)

[analog berry]

that works yeah but i lose whatever i had in the history which i often use

or like not in history

just above on my screen

[robust geyser]

the buffer yeah

i get that

kinda annoying

it would be cool if flanterm implemented that tho

might look into it myself

[analog berry]

ye

i have no idea how any of that works as well

just understanding the idea of tty/pty/etc was hard enough

[robust geyser]

you can apparently use \e[?1049h and \e[?1049l to switch to and from the "alternate screen"

[grand shadow]

for me the hard part about a proper tty subsystem wasn't the tty part but rather the process group and session stuff

that was so annoying to get right and I don't even know if my refcounting works 100%

[analog berry]

oh yeah

https://ratatui.rs/concepts/backends/alternate-screen/ found some docs from a random rust thing

[robust geyser]

also apparently its xterm specific :^)

[analog berry]

but everything supports it right

[carmine swallow]

I am pretty sure yeah

[robust geyser]

the crate you sent uses just that```rs
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct EnterAlternateScreen;

impl Command for EnterAlternateScreen {
fn write_ansi(&self, f: &mut impl fmt::Write) -> fmt::Result {
f.write_str(csi!("?1049h"))
}

#[cfg(windows)]
fn execute_winapi(&self) -> io::Result<()> {
    let alternate_screen = ScreenBuffer::create()?;
    alternate_screen.show()?;
    Ok(())
}

}```

so i think its safe to assume most things implement that

[analog berry]

even windows api??

lol

[robust geyser]

maybe, no idea

let me test that actually

[analog berry]

https://learn.microsoft.com/en-us/windows/console/setconsoleactivescreenbuffer

probably this stuff

[robust geyser]

yeah but the escape sequence

[craggy tide]

linux console doesn't either

[analog berry]

but linux console sucks

[craggy tide]

okay

[analog berry]

anyway whats the actual reason

[craggy tide]

literally that

[analog berry]

why limit yourself

[craggy tide]

it's meant to run with TERM=linux with ncurses

[analog berry]

idk what that means but thats alright

[grand shadow]

its neither an astral bug nor a flanterm bug

its a feature

[analog berry]

u mean the terminal hanging or the lack of alternate screen?

[grand shadow]

it tells ncurses to act as if it were running on the linux console

[grand shadow]

alternate screen

[analog berry]

ah

[craggy tide]

if you ever ran nano on the linux console you'd see it behaves identically

[analog berry]

i got that much

[robust geyser]

i just tested it

can confirm

but still, i think it would be nice if flanterm was a more proper terminal emulator

[analog berry]

ye

[craggy tide]

i will accept a pr that implements it

as long as it doesn't break other stuff, directly or indirectly

[robust geyser]

might look into it then, soon

[analog berry]

i mean it probably shouldnt?

u would just store a new char array somewhere

or like the entire state ig

[craggy tide]

i meant break other things when running TERM=linux ncurses

as maybe stuff relies on that escape code doing something else or not doing anything

[robust geyser]

its just two new sequences

[analog berry]

the new windows terminal probably does support it

the old one i doubt it

[robust geyser]

yea conhost is absolute dogshit

[analog berry]

yup

[robust geyser]

tho tbh idk, it might work

it does!

[analog berry]

damn

[robust geyser]

with native nano

also works within wsl

[analog berry]

maybe nano just calls that win32 api?

[robust geyser]

probably

[analog berry]

u could test it by killing it from task manager

[robust geyser]

well it goes back to the original content of the terminal when i do

interesting

[analog berry]

then it does probably...

because the win32 api is manual

u cant arm it to switch back on exit

or maybe u can? idk

[robust geyser]

no idea, but i killed it from task manager and it did switch back

thats what i meant, not sure if that was clear enough

maybe conhost is smart enough

[analog berry]

maybe it has some tricks up its sleeve

[robust geyser]

definitely, its windows after all lmao

[analog berry]

lol

true

[grand shadow]

@analog berry @robust geyser

[robust geyser]

very nice

does it sync? :^)

[obsidian pivot]

le cool

[grand shadow]

that'll be a task for init

init will do the echo r > /dev/acpi

but it will receive an event from acpi too

so like poll -> read -> check event -> kill(-1, SIGTERM) -> sleep(5) -> kill(-1, SIGKILL) -> waitpid() until ECHILD -> write(acpifd, "p", 1)

actually instead of a sleep(5) it'd be better to do a waitpid with an alarm

[robust geyser]

ah, i see

that sounds like a good idea

[grand shadow]

probably not even init tbh, I might just leave init only handling commands from a FIFO and have some other process in the background handling listening to acpi

but the shutdown command to init will be like that

[grand shadow]

aaaand there we go

that p was returned by the device

that should be all

for now

[grand shadow]

really simple poweroff

and you can see what init does

(thats why I downloaded gnu hello)

still missing the kill(-1, SIGTERM) etc but thats because I literally dont have that

yet

I'm gonna add a reboot command, the acpi listener and then call it a day

[obsidian pivot]

very cool

I assume shutdown also shuts down drivers?

[grand shadow]

TODO™️

there we go

I'll just look for some weird tty issue that happens on login when you backspace and call it a day

[grand shadow]

bruh

shadow defaulted erase to 0177 (del) instead of 010 (backspace)

[grand shadow]

well, that is all for today

tomorrow I will finally do the auth checks for signals I swear

[analog berry]

Nice work

@grand shadow u sure this pageoff thing is needed in map and unmap? Since we round the address down and round the length up anyway

[stark walrus]

O_O

no

I'll at least wait until shitos9

[analog berry]

yeah i dont think its worth to take up an entire shitos5 for marca if hes going to abandon it after a week 😉

[analog berry]

hmm am i supposed to rebuild somehow to get this to work?

i dont have a poweroff command

[stark walrus]

O_O

Eyy that's not what happens ok :>

I at least work for about a month

And then life gets to me and I have to do other things

And then life gets to me and I have to do other things

...

[grand shadow]

./jinx rebuild init

I think

[grand shadow]

Imagine like, you want to map 10 bytes starting at the last 5 bytes of the page

Wait now that I have a fresh mind I think I did it wrong too 💀 nvm

But yeah you have to account for it

[analog berry]

i think my logic still works?

[grand shadow]

But yeah youd be mapping only one page and get a page fault (best case) or access wrong memory (worst case)

[analog berry]

i cant think of a case where it doesnt

10 bytes would be rounded up to 4096

last 5 byttes would be rounded down to byte 0

i dont see a problem tbh

oh ok i see it i think

e.g.

map 4096 at 2048

[grand shadow]

Yeah

[analog berry]

so u have to like

uhhh

idk

add wahtever u rounded down

then round up that

size += addr - ROUND_DOWN(addr)
size = ROUND_UP(size)

or no wait lmao

there

fixed it

[grand shadow]

Did jinx rebuild init work

[analog berry]

yup

[grand shadow]

Epic

[analog berry]

pretty nice to be able to shutdown like that

also that bug was pretty stupid sorry

[grand shadow]

Lmao thats fine I think it was in my old acpi table mapping code too

[analog berry]

lol

[analog berry]

@grand shadow btw why is your memcpy src non const

that produces lots of warnings

also found a header bug in uacpi, its not a big deal but also produces a warning

[grand shadow]

I forgor

I think I also should get around to implementing the unmounting of filesystems sometime in the future, would be a step in the direction of cleanly powering off

[analog berry]

have u tested real hw yet?

[grand shadow]

Not yet, will do when Im home

[analog berry]

nice

[grand shadow]

@prime juniper does xclock tick on managarm

If so what was.needed to make it work

[prime juniper]

no

[grand shadow]

I thought it was some setitimer meme but it wasnt that

[analog berry]

in other words

what makes it tick

[prime juniper]

(whoops slipped into native dutch)

I'm pretty sure it used to work once upon a time

but last I checked, the clock does not tick

hell, xclock straight says 00:00 iirc

[grand shadow]

Wtf

I remember having.an issue like that and it turning out to be some weird rounding issue

[prime juniper]

I honestly haven't looked into it at all, but I'm 99.99% sure that it used to work

[grand shadow]

-    unsigned long update = (int) (_update * 1000.0 + 0.5);
+    unsigned long update = roundf(_update * 1000.0 + 0.5);

Thats what I have in my patches

[prime juniper]

does it work? does xclock actually work?

cuz if so, I'm going to yoink that

[grand shadow]

It doesnt tick but the clock does show the current time

[prime juniper]

let's see if we can test that in class

booting vm

ah I get a divide by zero in xclock

time to gdb that bitch

0x0000000000405041 in round_time (tv=0x7ffffffff320, tm=0x7ffffffff330, _update=0) at /var/lib/managarm-buildenv/src/ports/xclock/Clock.c:1306
1306        unsigned long    new_millis = intervals * update;
(gdb)

interesting

[winged basin]

i wish i knew how to make xclock tick

i also assumed it was an iTimer thing

[prime juniper]

/*
 * Round the current time to the nearest update interval using
 * milliseconds since midnight
 */
static void
round_time(float _update, struct tm *tm, struct timeval *tv)
{
    /* interval in milliseconds */
    unsigned long       update = (int) (_update * 1000.0 + 0.5);

    /* compute milliseconds since midnight */
    unsigned long       old_secs = time_seconds(tm);
    unsigned long       old_millis = time_millis(tm, tv);

    /* Nearest number of intervals since midnight */
    unsigned long       intervals = time_intervals(old_millis, update);

    /* The number of milliseconds for that number of intervals */
    unsigned long       new_millis = intervals * update;

[grand shadow]

Very interesting division happening there

[prime juniper]

yes

but t = tv->tv_sec - old_secs + new_millis / 1000; and update is null

so new_millis is something times 0

and therefore 0

and something plus 0 / 1000 is UB

[wild marsh]

?

[prime juniper]

let's try your patch

[analog berry]

why is 0 / 1000 ub

its just 0

[prime juniper]

wait

you're right

[wild marsh]

zero divided by anything (!=0) is just 0

[prime juniper]

yeah I'm stupid

then idk

but I'm gonna yeet that patch

and test that

[wild marsh]

i guess _update = 0.0?

[prime juniper]

yes

gdb says it's 0

just 0 cuz int

oh no _update is float

hmm

still 0 tho

testing again

@grand shadow update on xclock, your patch unfucks the divide thing, but no ticking

[grand shadow]

Yeah I imagined

I do wonder why it doesnt tick

[prime juniper]

yeah

I'm pretty sure that used to work

but idk

almost like the time update function isn't called

no weird shit in e9 log either

[grand shadow]

Guess its out of battery or something

[prime juniper]

if you figure it out please let me know

I might take a stab at it later this week if all other items are blocked (something something multi user / cooked mode / systemd)

[grand shadow]

I guess one would have to dig into the xclock source to figure it out

Hm I do have nothing to do at work atm

[prime juniper]

yeah that's gonna be source digging

[grand shadow]

Seems like it adds the timeout in the clock_tic function

if (w->clock.update && (id || !w->clock.interval_id))
    XtAppAddTimeOut(XtWidgetToApplicationContext((Widget) w), waittime(w->clock.update, &tv, &tm), clock_tic, (XtPointer) w);

In SetValues too it seems

Redisplay is exposed as the expose function which I assume is what Xorg wants when redisplaying it and it calls the clock_tic function

So Id have to check the timeout stuff there

[prime juniper]

probably

[grand shadow]

I wonder if building another version of gcc makes the mlibc meme go away.....

@analog berry my main pc (got it 2021)

It says it has 16 processors but its 6 core, does that mean the cpu originally had more and theyre just defective?

[obsidian pivot]

maybe hyper threading?

[grand shadow]

Nah it doesnt have that

Its a i5-9400f

Gonna get a full distro and try it on my laptop cuz ik there the ps2 kbd and mouse work

[obsidian pivot]

maybe the madt table is schizophrenic

[errant zephyr]

this is how cpu manufacturing works

they make a 16 core part

and if a lot of cores dont turn out right they sell a 6 core part

or rather

its an 8 core part

with 16 hyperthreads

and two cores went bad

[grand shadow]

I imagined

[errant zephyr]

or alternatively it's just that the controllers are the same and dont change

even though the cores dont really exist

[grand shadow]

Thatd explain why the default lapic ids are how they are

[obsidian pivot]

I assume they are out of order

i.e., instead of 0,1,2,3,4,5

it's 1,4,7,4,8,9

[grand shadow]

it was like 0 2 4 ...

[obsidian pivot]

yeah these were just random numbers

point is they looked random

[analog berry]

i mean this is just the number of devices in the AML namespace of type Processor

which are deprecated btw

u would match them against MADT to know which one is which and which one is online

there can be 0 or tehre can be 99999

its not indicative of anything in particular

[analog berry]

nice that it shuts down

does the power button work?

[grand shadow]

yeah I pressed the power button

[analog berry]

niice

[grand shadow]

which put some data in /dev/acpi which was read by acpid which passed an event to init which synced the disks and wrote into /dev/acpi to shut down

[analog berry]

💀

[grand shadow]

"""synced the disks"""

[obsidian pivot]

real chads just throw that in the kernel

[grand shadow]

(it was an initrd)

[obsidian pivot]

register gpe event handler

shut down

[grand shadow]

you gotta put the running processes in their place first smh

(I dont even do that I don't have kill(-1, *) yet)

testing on my laptop now

it has a key power button idk if that will matter a lot

@analog berry seems like it worked on my laptop too

[analog berry]

thats amazing

power button too?

[grand shadow]

Gonna first try out some things in xorg

[analog berry]

that doesnt really matter, what matters is how they routed it

it could be the "simple way"

aka fixed event u have right now

or the "complex way" via the EC

[neon crane]

the numbers are unique but some numbers are missing

[obsidian pivot]

it's a 6-core cpu

so 6 lapic ids

reported in the madt

[neon crane]

well, 12 due to hyperthreading?

[grand shadow]

Neat glxgears work and is a bit faster too

[obsidian pivot]

I mean it doens't matter

[analog berry]

damn thats pretty cool

[obsidian pivot]

le very cool

[grand shadow]

cpu doesnt have hyperthreading

so I really just assume it had defective cores

[grand shadow]

same thing for my laptop, its a 4 core 8 threads cpu iirc

[neon crane]

thats weird

[analog berry]

now i want to test on my laptops as well

maybe it will boot into gui at least unlike managarm

[obsidian pivot]

[neon crane]

never seen that before

[grand shadow]

probably cuz initrd

if it doesnt hang at ps2 init

I hate ps2! I hate ps2!

[analog berry]

managarm was failing mostly because drm bugs

[grand shadow]

why is it that the only two damn fucking computers that it works are the only ones that I can test

[analog berry]

sometimes bad xhci, sometimes bad ahci, sometimes ps2 as well

its always a different thing

[analog berry]

lol ikr

uacpi has been tested on probably like 50ish computers at this point

[grand shadow]

Oh quake is fucking SMOOTH in real hw

[analog berry]

daamn

nice

wait until u see it running on an igpu

[obsidian pivot]

astral is definitely one of the cooler oses on this server (idk about the kernel quality, but the user space is very cool)

[neon crane]

you sure it wasnt before the uacpi stuff?

[analog berry]

i think its unrelated

he just never tested before

[analog berry]

yeah

[obsidian pivot]

does astral support swap of some kind

[grand shadow]

It ran a bit slower but uh its more due to me unshittifying my ringbuffer and memcpy

[obsidian pivot]

or no

[grand shadow]

Only page cache but I assume itd not be too hard since I have most of the facilities in place

[analog berry]

it has page cache

[grand shadow]

And Id say the page cache is battle tested at this point since it compiled like half of gcc

[obsidian pivot]

well definitely read the denning working set paper

when (and if) you implement swap

it's pinned in #virtual-memory

[grand shadow]

I was planning on doing that at one point in the future

Right now if a page is paged in it stays there forever

Until the process unmaps or quits

[obsidian pivot]

maybe I could do that sometime soon

but since school for me starts tomorrow I won't have enough time to work on my own kernel

[grand shadow]

Gonna try some xscreensaver demos now

[grand shadow]

Im in uni break now even tho I still have a job
I will try to do as much as I can until late september

Note to self: mount tmpfs root as 755 instead of 000 💀

Ah shit xscreensaver is stuck I think I ran the wrong command 💀

Oh well time to test the power button

@analog berry yeah here it didnt work

[analog berry]

yeye for laptops its usually the "hard" way

[grand shadow]

lemme see if the acpi shutdown will work tho

with shutdown

[analog berry]

u mean poweroff

[grand shadow]

yeah poweroff

good to see that astral at least still works on (my) real hardware

[analog berry]

did it power off?

[analog berry]

yeah thats p cool

[grand shadow]

nah its rebooting I had turned it off from holding the key too much

[analog berry]

ah

yeah holding the key too long bypasses acpi

it does send the emergency shutdown event tho

[grand shadow]

I need to read a bit on what acpi than do for when I delve deeper into that

[analog berry]

wait uhh

[grand shadow]

poweroff does work tho

[grand shadow]

how

[analog berry]

how did u follow yourself? idk

[grand shadow]

it sends the event and everything and then powers off

[analog berry]

nice!

yeah if u want the power button to work there thats gonna take a bit of work (not too much)

https://github.com/managarm/managarm/blob/master/kernel/thor/system/acpi/ec.cpp

this basically

[grand shadow]

hmmm 400 lines of phd code

[analog berry]

if u dump the tables from that laptop it probably has a PNP0C0C device

which is the power button

[analog berry]

its similar to ioapic

write one reg read the other

or write both

etc

https://uefi.org/specs/ACPI/6.5/12_Embedded_Controller_Interface_Specification.html

[grand shadow]

it'll be a TODO™️ for now

[obsidian pivot]

how do you hook it to uacpi though

[analog berry]

i sent an example above

from managarm

uacpi_install_address_space_handler(
  ecDevice->node, UACPI_ADDRESS_SPACE_EMBEDDED_CONTROLLER,
  handleEcRegion, ecDevice.get()
);

and for ec events theres uacpi_install_gpe_handler

[prime juniper]

DRM bugs?

[grand shadow]

wrong reply

[analog berry]

lol

[grand shadow]

@analog berry

[prime juniper]

Yeah fuck

[analog berry]

yeah drm bugs

[prime juniper]

Fat fingers

[analog berry]

screen goes black after wayland is brought up

and never goes back

[prime juniper]

@formal patio DRM issues?

[analog berry]

either ps/2 code dies or the resoultion it tries to modeset is bogus

no idea after that

i think qookie had some ideas

[prime juniper]

Sounds like something might not be starting properly / application crashing / driver crashing

[formal patio]

context?

[prime juniper]

Boot to weston and keeps black screen

[analog berry]

managarm no work

[formal patio]

I hate to tell you but glxgears really isn't a good (graphics) benchmark, it spend relatively little time drawing

[analog berry]

well his quake is pretty good as well sooo

[formal patio]

all the context switching, IPC et al is very likely more significant

so depends

[obsidian pivot]

I think the point matthew was trying to make is that running glxgears is fucking cool and it runs faster

[formal patio]

yeah I'm not saying it isn't, just pointing out that glxgears fps are to be taken with some caution

[winged basin]

digital rights management

[obsidian pivot]

dog ritual manager

[grand shadow]

yeah

for what oberrow said

plus if quake is smooth then astral is good enough I dont need to develop it further smh

[obsidian pivot]

if it can run:

while(1)
  fork();

[analog berry]

i wanna play gta sa on astral

[obsidian pivot]

then it's good

[winged basin]

will Astral get a DRM subsystem?

[analog berry]

for that we need wine and for me to implement igpu

[obsidian pivot]

I plan on supporting this in obos

[analog berry]

via uGPU

[flat copper]

port qemu

:^)

[obsidian pivot]

port qemu

then

[winged basin]

this uGPU, what will it need?

[analog berry]

probably same complexity

[obsidian pivot]

run astral

[grand shadow]

RLIMIT_NPROC

[obsidian pivot]

then get qemu on astral qemu on host qemu

[analog berry]

like from the kernel?

[winged basin]

yeah

[grand shadow]

(I dont have rlimit)

[obsidian pivot]

then run astral

then get qemu

[grand shadow]

I probably have everything for a qemu port just not the recipes

[obsidian pivot]

then get windows

then run qemu

then run astral

then run glxgears

[grand shadow]

1 year per frame

[obsidian pivot]

so like me running mc with shaders?

[analog berry]

it will work as a drop-in drm subsystem or work with an already existing one. very similar api to uacpi with probably a few extras on top as it needs more kernel integration to work, also stuff like registering an on-disk file/directory etc to make it work

well and something in the userspace to submit commands, e.g. Mesa

[obsidian pivot]

might port that to obos in a million years

or I could speedrun it and become the first adopter+tester on real hw

[grand shadow]

one thing I wanna do is try to passthrough my laptop's ethernet card and see if I can write a driver for it

[winged basin]

i am thinking about e.g. with DRM-PRIME then it will need some sort of VMM integration

[analog berry]

yeah definitely

ill need to think of how to design that api well

[analog berry]

this stuff is very very far off

[grand shadow]

by ugpu my kernel will hopefully not suck as much™️

[analog berry]

ill probably use your kernel as a test bed

[obsidian pivot]

so two million years?

[analog berry]

i want to get the very basic proof of concept running

e.g. glquake with actual acceleration

[grand shadow]

glquake would be very neat

[analog berry]

that will take probably a few months

to work on my one latpop

[obsidian pivot]

I wonder if anyone's ported openjdk

to their os

[grand shadow]

mamogram

[obsidian pivot]

and if it has mesa

[grand shadow]

mamogram

[obsidian pivot]

wonder what mc needs to run...

[grand shadow]

web browser for launcher

[obsidian pivot]

without launcher

[analog berry]

i will need u to port/implement whatever is needed besides DRM for Weston

or well as long as i can get mesa to work i guess

whatever

[grand shadow]

:(

I mean I think drm is the worst thing?

[analog berry]

definitely is

everything else is smaller

[grand shadow]

I have no idea what all the linux specfic apis weston uses are

[winged basin]

westron needs drm, epoll, evdev i think

that's what netbsd needed for it

[analog berry]

ye

u can probably do epoll easily

evdev idk, probably also not that complex

[grand shadow]

maybe when ugpu gets closer to being real I can look into that

[analog berry]

i can run mesa in astral right now right?

[grand shadow]

then ig I could also move xorg to use xf86-input-libinput instead of keyboard/mouse

[grand shadow]

yeah

it runs with llvmpipe rn

[analog berry]

glquake works with it rn?

[winged basin]

epoll you can do a shite implementation of relatively easily if you don't care about the case of thread A waits on an epollfd, thread B adds some new epollfd to the set, new fd has an event

[grand shadow]

yeah but has some graphical bugs

[analog berry]

why is that?

[grand shadow]

I have no idea

[analog berry]

with llvmpipe?

[winged basin]

and you have to be careful with epoll to not do the sensible thing too much as linux certianly doesn't

[analog berry]

lol

[grand shadow]

yeah

[winged basin]

https://idea.popcount.org/2017-02-20-epoll-is-fundamentally-broken-12/

[analog berry]

bad fpu state?

[grand shadow]

tyr-quake (fully software) works but tyr-glquake has some glitches

[grand shadow]

honestly idk

I imagine everything is fine?

[analog berry]

well if it works on linux with llvmpipe

then..

either mlibc bug or astral fpu init bug

[grand shadow]

I could see if I can adapt my current poll system to do that

my current polldesc_t only accepts one thread on the sleeping list but it could maybe be bigger?

I'd have to think about that one

[winged basin]

i have been thinking about epoll literally right now before coming into this conversation and i haven't decided yet how it should best be done

[grand shadow]

my fpu init happens in two places, per processor and on context init

[winged basin]

i'm looking at what solaris does right now (don't worry, i don't understand it either)

[analog berry]

im betting on the former in that or btw

its easy to get some float thing wrong

[grand shadow]

https://github.com/Mathewnd/Astral/blob/d81c2fdc9d4d931d383e9b7e7b470f07da705502/kernel-src/arch/x86-64/cpu.c#L65
https://github.com/Mathewnd/Astral/blob/d81c2fdc9d4d931d383e9b7e7b470f07da705502/kernel-src/include/x86-64/arch/context.h#L59

[grand shadow]

/dev/poll wtf

[grand shadow]

tbh there was some weird rounding issue in xclock

[winged basin]

one of the prior arts epoll didn't copy

but with a little work it was made to almost emulate epoll

it doesn't emulate the thing where files that you've closed still give epoll notifications with the wrong FD number

[analog berry]

wtf

[winged basin]

it requires you to do certain things when you fork() to trigger the wild behaviour

[grand shadow]

I should see what it disassembles to

[carmine swallow]

btw @grand shadow are you planning on doing a hda driver yourself at some point? because if not (and you are open for contributions) I could try making one for astral (and also maybe look into somehow making the nitpicky lib (aka alsa) work with it)

[analog berry]

yeees pleasse do that

[grand shadow]

I mean it was on my radar to do but I never began writing it too much (I was mostly just reading the spec and trying to understand how it works) but if you want to I won't complain

[analog berry]

that would be very neat

then we could have the full stack on astral

usb+graphics+sound

[grand shadow]

plus you probably know better how it works than me

[analog berry]

@grand shadow can u reboot via poweroff somehow?

[grand shadow]

reboot

[analog berry]

oh u added that as well?

lol

[grand shadow]

yeah

[analog berry]

nice

[grand shadow]

almost the same code except it sends 'r' to the init pipe rather than 'p'

[analog berry]

big api

[carmine swallow]

well I have written one driver for my own kernel that I ported to vinix :p I do know in general how it works but the alsa stuff I have no idea about (as it has some weird parameter refining stuff and then if you don't get that right alsalib refuses to work)

[analog berry]

yes qwinci do htat pls

[grand shadow]

that'd be neat

[analog berry]

u now have the isr->priv field

what else might u want

[carmine swallow]

ig I can take a look then, first have to compile astral tho

[grand shadow]

I also dont complain about people wanting to write drivers cuz it helps me unshittify the kernel and make the apis a bit better (thanks infy)

[analog berry]

lol thanks

[analog berry]

u can make a lightweight build

that takes a little less than 10 years to compile

[grand shadow]

you only have to compile gcc once, thats one time less than twice and you dont need to compile llvm

[analog berry]

we should make a script like ./just-build-the-goddamn-thing.sh

that would do all the steps

[grand shadow]

or I could y'know
make the build system actually good

maybe I can look into that

[analog berry]

or that

[grand shadow]

I'll first make it so that DISTROTYPE=minimal doesn't build everything and push that

[carmine swallow]

to build a minimal build I just need to do make jinx + update submodules + make kernel + make minimal?

(and then the run the iso rule after that to create the iso)

[analog berry]

make jinx
./jinx build mlibc bash coreutils init distro-files vim nano mount netd
make DISTROTYPE=minimal kernel initrd astral.iso run-kvm

iirc something like this

[grand shadow]

welp that was easy

no idea why I didnt do that before

[analog berry]

btw why do u have multiple accounts

on gh

[grand shadow]

huh

[analog berry]

oh its a github bug lol

it tells me mathewnd followed matheus

[grand shadow]

I sure do wonder how I followed myself

[analog berry]

thats why i thought maybe u had an alt

[grand shadow]

anyways I pushed a change to the makefile that should hopefully make it that DISTROTYPE=minimal only builds the MINIMALPACKAGES

brb

[carmine swallow]

should cd minimalsysroot be cd mlibc? because no files ever get put to minimalsysroot even if I do ./jinx install minimalsysroot

in any case the shell is very snappy when compared to eg. managarm lol

[grand shadow]

wait

ohhhhhh

ohhhhhhhhhhhhhhhhhhhhh

I forgot to add minimalsysroot to the ./jinx install

I have dumb

[carmine swallow]

I tried running that manually but it didn't put anything to there

[prime juniper]

@grand shadow there’s a non zero chance that you’ll need to expose some shit that udev usually does / dependencies on udev and friends

If, and that’s a massive if, that’s the case

[grand shadow]

@carmine swallow pull again

[prime juniper]

Enjoy netlink (not all of it tho)

[grand shadow]

what even is netlink

[prime juniper]

Linux specific API to broadcast events. But used for all kinds of shit

if I recall correctly it was made for SELINUX originally

But now a metric shitload of shit goes over it

[grand shadow]

aghhh they need better names I thought it was network related

[prime juniper]

Including some network route BS

[prime juniper]

Oh it’s used for it

But not originally if I recall correctly

It’s a Linux kernel interface for IPC if I trust Wikipedia

[grand shadow]

linux picking non ambiguous names challenge

[prime juniper]

https://datatracker.ietf.org/doc/html/rfc3549

[grand shadow]

if I told you "I'm implementing netlink" and you never heard of it before whatd you think

[carmine swallow]

now it works, thanks

[grand shadow]

epic

[prime juniper]

udev needs NETLINK_KOBJECT_UEVENT

You can ignore all other families if you need to implement this

(Afaics)

[grand shadow]

theres a stub driver in io/audio/hda.c that you can start out of

but theres no msi in pci, only msix

I should get around to implementing it I got completely derailed

yeah while you work on that I will implement msi

[grand shadow]

@carmine swallow do you plan on using multiple msi interrupts

[carmine swallow]

I am pretty sure that there is only one interrupt on the hda

[grand shadow]

then multiple msi interrupt support will be assertos™️

[carmine swallow]

I guess most devices that use multiple interrupts should support msix anyway

[grand shadow]

oops all children of init were running with a full signal mask because I forgot to unset it when spawning a child

well I pushed msi

untested but SHOULD work

I'm going to deal with dinner then play around seeing if I can see what caused the xclock rounding issue (and if that can be fixed without patching)

[carmine swallow]

nice, I got the register structs made and controller stopping if its already running done (that mostly just happens when you use pci passthrough to pass a controller from the host to qemu, then if you don't properly stop it it can became completely unusable until the next host reboot). I think Ill go sleep now and continue with it tomorrow (and also see if the msi's work)

[grand shadow]

alrighty nice work

[craggy tide]

@grand shadow when does summer break end for you?

[gritty shore]

i should stop looking at your thread, it makes me want to work on my thing but I have so much homework

[grand shadow]

uni break started last week so until like 20th of september I'll not have to deal with it

still have to go to work though

[craggy tide]

hm

i will have to play catch up but i am still sitting in limbo here figuring out what to do with my life lol

[grand shadow]

wdym

[craggy tide]

good job though, you've made a lot of progress

like

who when how or if i am going to be paid for Vinix work

[prime juniper]

Unironically jealous of the progress

[craggy tide]

it's fine i'll just steal get inspired by the progress once time comes :^)

[grand shadow]

lyre 3.0 soon™️

I mean if you eventually get paid for vinix I think the first thing you should do is weed out the remaining instabilities and then flesh out the tty stuff and implement ptys to have terminals in xorg

[craggy tide]

yes

that was the plan

[grand shadow]

you can run webkit smh

[craggy tide]

actually it was that alongside looking into starting an aarch64 port perhaps, but not simultaneously

[grand shadow]

also cvttsd2si seems to be the instruction that it uses to round without the roundf

(I have no idea about sse programming)

[craggy tide]

also figure out what's missing that doesn't make gtk work on Vinix (it works on Lyre)

[grand shadow]

it'd really be nice to see some progress on vinix eventually, it was one of the big inspirations for me

[gritty shore]

i think i will work on personal projects on fridays

I chose a schedule that doesnt have classes on fridays

[grand shadow]

hmmm

wonder if 0.5 is treated this way

hm actually seems like this is the expected behaviour of the function

not funciton

instruction

[prime juniper]

Yes and it took a lot of time I’m aware. But I see progress threads where big progress is made quickly and I guess I just wish that Managarm had that too. And yes webkit is based, but I can’t run fucking configure scripts without 20 different autoconf cache variables and what the fuck is a user, I only do that with a shitload of hacks and then it still isn’t done properly

[robust geyser]

i guess that's what happens when the main focus is put on impressive ports

[prime juniper]

or when there’s like 2 people working on it, one of which does magic with drivers but get stuck in review hell and the other one is me

[robust geyser]

that too

[gritty shore]

well tbf you're constantly pushing the boundaries of mlibc and adding code to it so OSes like astral dont have to do the work

[grand shadow]

I've been having to implement some stuff in mlibc

nothing too deep (I barely know c++) but it is some things

[grand shadow]

yeah I think the biggest issue with managarm is just how slow it is to develop sometimes

[robust geyser]

phd code moment

but i like some of the abstractions

[gritty shore]

yeah ofc but managarm has the issue of having not many contributors and also having to maintain mlibc which is huge

[prime juniper]

Yep. Honestly focussing more on stability and bugfixes would be good too but every time we suggest that, we agree, and then fail to keep focus lol

[robust geyser]

same with me going back to osdev

i just write some code

realize how shit it is

and drop the project lol

[gritty shore]

honestly I'm pretty happy with my current iteration

just gotta finish a sidequest first so I can develop it on my mac

[prime juniper]

Anyway, off to sleep and crying about systemd porting. There’s a printf issue in printing help and I suspect there’s a linker bug related to rpath. If that works, I need to debug udev and cry

Then debug systemd as init actual

Which would involve a lot of shit

All in the name of stability right? (Better init)

[robust geyser]

hopefully lol

[gritty shore]

did you try porting openrc

[robust geyser]

gn and good luck

[gritty shore]

or runit

[prime juniper]

Runit yes

Should work I think idr

But systemd

[gritty shore]

most people don't like systemd tho

[prime juniper]

Yet a lot of people use it

[grand shadow]

I tried porting openrc and it needed quite a few missing(?) linux stuff

[prime juniper]

But it’s already compiled, now it just needs to work

[robust geyser]

its the necessary evil

(systemd)

[grand shadow]

just use sysvinit

its that simple

[prime juniper]

I don’t think I have mlibc patches for the compilation part

[gritty shore]

the good thing about systemd is its UX

[prime juniper]

Which is based

[gritty shore]

runit fucking sucks in that sense kinda

[gritty shore]

but the problem with it is it tries to do everything all at once

so its fucking huge

[grand shadow]

bruh so
_update is 0, update is 0
_update * 1000 is 0, + 0.5 is 0.5
the (int) does a truncation which turns it into 0

which then is a div by 0 in time_intervals

is it really wanting (int) to round it??

ooor

maybe _update is being passed wrong

w->clock.update is 0

could it possibly be that the missing scanf %f is at fault here?

it seems like w->clock.update is initialized by X and it seems to have some default

passing -update doesn't seem to change it

I AM SO FUCKING SMART HOLY SHIT

@prime juniper IT LIVES

basically the problem seems to be: w->clock.update needs to be initialized by whatever the fuck in the X11 stack it needs to be, but it isnt (again, I am betting on the missing scanf %f), so adding a

w->clock.update = 0.5;

on the Initialize() function fixed it

funnily enough, theres a check for an invalid update being passed but it was commented out for some reason?

    /* make invalid update's use a default */
    /*if (w->clock.update <= 0) w->clock.update = 60; */

[gritty shore]

awesome

[grand shadow]

bleh no idea then

but yeah that is why xclock doesnt tick or work normally: its missing update for some reason

well digging into xorg is something I dont want to do

this is where the default is defined

[grand shadow]

I lied

the fuck is this sscanf magic

actually

yeah thats equal to %f

I was right

thats why mlibc needs %f in scanf, so xclock can tick

[grand shadow]

so yeah, w->clock.update isn't initialized because %f inst implemented for scanf and it does some sscanf magic to convert it in libXt

maybe this (missing %f) is also whats causing the tyr-glquake glitches?

[analog berry]

I wish u could explain literally every bug with this

[fleet osprey]

Oooh nice debugging job! Love seeing deep dives into obscure issues like this

[prime juniper]

Nice work

[carmine swallow]

msis appear to work fine except that there was a typo in pci_msisetbase inside the if (is64bit), address > 32 should be address >> 32

[grand shadow]

Omg howd I miss that

[grand shadow]

Ill let you push the fix with hda nothing else is really gonna use msi for now

[carmine swallow]

now time to figure out the cursed stuff ig (though not that I even implemented anything other than created those device files with empty logging ioctl impls, the hda side is also missing things like queuing audio because I want to first know how exactly it should work, I am pretty sure that there is a signal along with a write ioctl)

[grand shadow]

Damn

Has it played any audio?

[carmine swallow]

no because I didn't yet implement the queueing stuff but I did implement all the stuff needed to eg. set the format and start the playback

[grand shadow]

Ohhh neat

If you need me to implement anything lmk

[carmine swallow]

choosing a supported sample rate is kinda cursed though, its like billion ifs because there are two base rates (44100 and 48000) and then you can multiply that by 1/2/3/4x and divide by 1/2/3/4/5/6/7/8x (so you have to find the closest match that it supports)

[carmine swallow]

realloc accepting null would be nice, though I already implemented that locally (well its literally just a if (!ptr) return alloc(size)) but other than that I haven't yet needed anything new

[grand shadow]

You can push it witj the hda driver too then, thatd be how Id implement it anyways

[analog berry]

wait u implemented hda in like a day?

[grand shadow]

Qwinci probably has got it down to an art atp I think he wrote one for vinix and for his oses

[carmine swallow]

well I already knew how a lot of the basic stuff is done because I did it before two times (and its still missing some things ofc, its like the minimum for playback eg. no hdmi support or anything)

[analog berry]

nice, thats p cool

[prime juniper]

Can I push Managarm into the queue for a HDA driver?

[carmine swallow]

I am pretty sure Leo wanted to do that tho

[prime juniper]

Leo is really busy tho

[carmine swallow]

I mean maybe, then there would need to be a new server tho and idk how do they work (though ig I could come up with something based on the existing ones)

[formal patio]

yeah I won't touch sound for a while

iGPU stuff is torchering me enough already lol

[analog berry]

what are u working on atm and how is the progress

[formal patio]

skylake displayport modesetting

it's not working yet

(obv)

[analog berry]

i see

[robust geyser]

i swear to god

everything that qwinci touches

has to play audio

(not a bad thing)

[grand shadow]

Quake with audio soon???

I think tyrquake has audio

And can use sdl for it

I think I disabled it cuz it was segfaulting or something lmao

[carmine swallow]

now I ran into something that would be nice to get fixed, apparently mmap doesn't check if the underlying file supports mmaping + succeeds even if it doesn't and then fails in an assert in vmm pagefault handler when the mmap'd memory is accessed

at least if I understood whats happening correctly

[grand shadow]

Ill fix that when Im home

[obsidian pivot]

@grand shadow have you made it so that building astral is easier

now

[grand shadow]

yeah you can just make DISTROTYPE=minimal and it wont build llvm and gcc twice (only gcc once)

and you can add the packages you want to the makefile

[obsidian pivot]

I'm considering unshittifying the astral vmm

by adding swap

• working set things

it won't be the best, but it will help ease out oom conditions

since I don't want to work on ttys rn, I'll start

[obsidian pivot]

I'll try to abstract it away as much as I can

as to not interfere with drivers

and stuff

[grand shadow]

I mean hey if you want to go ahead

just tell me exactly how you are gonna do it/did it so I can know when I eventually need to change something

[obsidian pivot]

(just note, this might include some slight obos curse in astral)
((depends on how skill issued I am))

[grand shadow]

the mm stack is pretty stable so anything I will blame on you

[obsidian pivot]

it's probably just gonna be some extra linked lists to vmm ctx (both global and per-process) + some way to interface with swap

from the vmm

[grand shadow]

as long as kernel memory doesnt get swapped out it should be fine

I have a specific requirement that all kernel memory needs to be allocated at map time

to simplify things

[obsidian pivot]

no I'm forcing you to support that

[obsidian pivot]

i.e., ways to allocate/free swap file space, and doing io on said allocated swap regions

dam your vmm looks really simple

this is gonna be fun

[grand shadow]

yeah I tried to keep things simple in astral

its just a linked list of regions

[obsidian pivot]

I haven't look inside yet tho

[grand shadow]

a lot of the magic is just thanks to struct page

[obsidian pivot]

no optimized ways to fetch struct page?

[grand shadow]

wdym

[obsidian pivot]

like no fast way

to fetch a struct page from a virt. address

[grand shadow]

not rly no

[obsidian pivot]

.

[grand shadow]

but you just get the physical and then the struct page from that

[obsidian pivot]

and does struct page represent a virt. page or phys. page

[grand shadow]

physical

[obsidian pivot]

so how do you get info for a virtual page

I'm going to quite a bit of metadata for this

[grand shadow]

using the vmmrange_t its in

there is no struct page for virtual pages

[carmine swallow]

also this was specifically with a file in the devfs where I think it gets proxied through the devfs generic ops? that might make it harder, though ig there could be some kind of flag or something like that

[grand shadow]

I think I might just have the map address being some magic value like -1 being a "check if mapping is actually possible"

[obsidian pivot]

I'll start in a bit

gtg now

[grand shadow]

so VOP_MMAP(vnode, -1, ...) would return ENODEV if mapping isnt possible

cya

I need to actually write docs if people are gonna start contributing now

[carmine swallow]

the good thing is that the code alone is pretty easy to understand

[grand shadow]

lol thanks

well I pushed the fix

soon I will get around to making vmm_map be able to return an errno

I think for that I will just have a macro VMM_MAP_OK() and a VMM_MAP_ERRNO() and have any return values < PAGE_SIZE be an errno

[analog berry]

lol

why astral instead of obos

[grand shadow]

actually I wouldn't even need to have a VMM_MAP_OK

just VMM_MAP_ERRNO == 0 when its ok like literally everywhere else

[carmine swallow]

thanks

[obsidian pivot]

I already have that in obos for one

two

I don't wanna implement ttys

this is a procrastination strategy

[analog berry]

makes sense

[grand shadow]

ttys are fun until you have to session and process group

that took me a while to figure out how I would do it and I still don't think I do it 100%

but hey it works well enough for bash job control so worth

[obsidian pivot]

flip you I'm doing that anyway (if your vmm isn't too shit)

[carmine swallow]

idk how useful that is tho, swap in general is pretty useless on modern machines

[obsidian pivot]

stop you'll scare fadanoid

iprogramincpp

and I

[carmine swallow]

lol

[obsidian pivot]

I would've said hyenasky, but he ain't here no more

[analog berry]

i used to think this

but this is a skill issue

(to think it is)

[obsidian pivot]

make[1]: warning: jobserver unavailable: using -j1. Add '+' to parent make rule.

hmm

[carmine swallow]

to think that its useless?

[analog berry]

yeah

evne if u have lots of ram

[carmine swallow]

I have just disabled swap completely on my pc as I am not ever running out of 48gb ram (and I don't want to use ssd as a swap anyway)

[analog berry]

At least two reasons:

1. if u do it well and have a fast swap backing u can really make it feel like u have lots more memory (see MacOS/XNU)
2. Less resident pages -> less fragmentation -> more contiguous memory (do i need to explain why this is good?)

ability to do more advanced things like compress on swap etc

linux swap is just dogshit so everyone disables it

[obsidian pivot]

you spent too much time writing this, the valid answer is "swap is fucking cool" /hj

[analog berry]

Not to mention just more free memory == more memory to use for caching stuff

[obsidian pivot]

also

page cache can be paged out

to make it more of a cache

[carmine swallow]

that sounds cursed

[obsidian pivot]

not rly

[analog berry]

paged out back to the file it came from? lol

[obsidian pivot]

idr how this worked exactly

[analog berry]

page cache is not paged out, thats the entire point

its just written back to file if dirty

its desigend so it can be reclaimed easily

[obsidian pivot]

#1141057599584878645 message

[obsidian pivot]

that's what I meant

[grand shadow]

its shit in the way that by design faults in the kernel are instantly an error and the locking is shit for handling faults like that in the kernel (say youre loading memory from swap and want to vmm_map some meory to use, you cant because of locking unless you make a recursive mutex or something)

[obsidian pivot]

btw when you make the kernel pageable, not everything becomes pageable

[carmine swallow]

fair, ig there are some good things about it but eh I still don't see much use for it personally

[analog berry]

"page cache" is view of memory on disk

[obsidian pivot]

vmm stays out of it

[analog berry]

u dont swap it out

u just remove it

[grand shadow]

^

[obsidian pivot]

yeyeyyeyeyeyeyye eyyeyeyyeye

[obsidian pivot]

scheduler too

[grand shadow]

in astral pages stay in the page cache until all pages in the free list are gone and you have to start taking from the standby list

[analog berry]

in a hobby kernel only if ur really interested in vmm design and stuff

other than that i wouldnt bother

[analog berry]

yeah on linux kswapd thread wakes up then reclaims the page cache

then wakes up kcompactd to compact the pages

[obsidian pivot]

vmm.c

amazing file name

[grand shadow]

I also dont have the concept of a paged and nonpaged pool so you'd probably have to add that to my slabs

[obsidian pivot]

O_O

1000 line file

[grand shadow]

ext2.c is 1.9k thats nothing

[analog berry]

inb4 oberrow introduces the curse into astral

[grand shadow]

I would cry

I worked so hard to make it at least stable

[obsidian pivot]

fnuy typo

[grand shadow]

lmao I forgot to fix that

ah well doesnt matter flanterm supports it

its an easter egg now

[obsidian pivot]

how should I write commit messages

[analog berry]

kerenC

[grand shadow]

git commit -m "thing: change"

[analog berry]

look at the history

[obsidian pivot]

too much work

[analog berry]

i do -v so i can see the code im writing the message about

[grand shadow]

wait really

[carmine swallow]

also if you want a compile commands json I have a cmakelists I can borrow (just use it with host gcc and run cmake but don't actually build it)

[analog berry]

also git add -p if u didnt already know

[grand shadow]

git add -p is useful

I just didn know about -v

also it doesnt highlight the diff, cringe

[analog berry]

yeah i use it all the time

i never remember what i changed

[obsidian pivot]

I'll just use bear or whatever the tool was called

[analog berry]

depends on your editor

[grand shadow]

it uses nano by default

I should change it to vim

[analog berry]

it uses the variable from the config file