#Astral
grand shadow
bronze nexus
rivtard world rain
regal stone
i hate wacatac
wild marsh
i wonder what the detection heuristics are for wacatac because i've seen it be detected as a false positive a bunch of times
grand shadow
if (download.is_kernel && download.kernel_is_good) detect("wacatac");
analog berry
if (download.is_kernel && download.kernel_is_good) detect("wacatac");
if (strfind(downloaded_file, "memory is not kernec")) detect("wacatc")
bronze nexus
Wacatac is apparently a class of malwares and not a specific thing
That'd also explain why there are so many false positives of it.
grand shadow
got it to do a full build with the patches gcc, now it is time to update mlibc and do another clean build
my laptop probably wants to cry
grand shadow
almosst done, compiling wine
grand shadow
oh no
xterm stopped working
and so did me launching serial
did login stop working as well
ok I dont think its that
oh no it is that
creating /var/run makes it work
why would it not display an error though?
yeah creating that makes xterm also work
so weird though, that it makes both hang
bronze nexus
Smells like something is missing error handling
grand shadow
time to go through the ports and see what works still works
grand shadow
Smells like something is missing error handling
indeed, might be in mlibc
as ^C still kills the process
sudo is broken, doesnt seem to have setuid bit
grand shadow
seems like dns is also borked
bronze nexus
Damn, what changed to break all this shit?
grand shadow
I updated mlibc and patched gcc 💀
its probably the mlibc update, I was like a good year behind I think?
bronze nexus
Mlibc regressions 💀
grand shadow
it does seem that just dns is down
gotta figure out if its my udp or something else
it used to work
seems like it is just the dns issues
tested the other ports
su also does not work
I think this doenst work more cuz of the jinx update
hm
yup
lets see how gloire handles this
oh they dont
grand shadow
had to port xbps and will do post install hooks for this
grand shadow
in a side quest to implement advisory locking now
so that xbps can run so that I can do post install hooks so that I can fix sudo et cetera
fleet osprey
next side quest: fixing PS/2 
cyan junco
I'm literally not gonna add it to my kernel
how are you gonna do sudo ?
bronze nexus
how are you gonna do sudo ?
Well not
Getting superuser will be done by talking with init to get it to spawn a privileged process.
wild marsh
Getting superuser will be done by talking with init to get it to spawn a privile...
systemd's run0 style
bronze nexus
systemd's run0 style
yep
cyan junco
Getting superuser will be done by talking with init to get it to spawn a privile...
oh cool way
grand shadow
grand shadow
My todo list rn is finish advisory locks, get xbps working, add post install hooks to a bunch of packages which need it, figure out why dns stopped working and then continue whatever I was doing
Small derailing
Worth
fleet osprey
small derailing from the end goal of fixing PS2 
grand shadow
one day a deity will descend from the heavens and deliver a pr for a perfectly working ps2
fleet osprey
Just blindly assume it's there and wait for interrupts for it like I did
https://github.com/techflashYT/Techflash-OS/blob/old-bootboot/kernel/hardware/kbd.c
also please excuse the godawful code style
I don't write code like that anymore
this is old code
also since I was a fucking dumbass and decided that writing code under a custom license was somehow OK - you can use that code under whatever license for whatever you want
(if you do decide to use it or parts of it, for whatever horrific reason)
grand shadow
alright, I think I got advisory locks implemented
time to see if everything explodes
ewww stop doing permission checks I am your creator you must respect me
oh woops, forgot to add flock to the mlibc sysdep
nope, borked
me when I forget to intiialize variable
it ran but gpf'd while closing a file, likely due to the advlock
grand shadow
okay
I need to get this to run on the first boot now
xdddddd
there we go
I might want to just pipe that into a logfile, though
and if I run it again it does not run
perfect
there we go
and the xbps-reconfigure logs will be in /var/log
now I just gotta write the post install hooks
I'll do that tmr
and then I gotta see what broke dns
prime juniper
Nice work
grand shadow
@prime juniper mr port man I have a mlibc related question, does mlibc support vdsos
I looked into rtld and it seems like it only prints the value of the auxval
sharp carbon
upstream does not but depending on your style of vdso (is it required to make syscalls (hard) or only an optimization (easy)) it's relatively easy to add
grand shadow
upstream does not but depending on your style of vdso (is it required to make sy...
Itd mostly be for some syscalls like gettimeofday or getting the cpu id
sharp carbon
for the "only an optimization" style this is all that's necessary to implement https://github.com/proxima-os/mlibc/commit/6ec4368abea33c0b95a4066d2b939bc4db911896
for the "required to make syscalls" style rtld itself hard-depends on the vdso and so the self-relocation needs to be changed which is much more extensive (https://github.com/proxima-os/mlibc/commit/08ea7f5a145f416d9862cf2ca23533549e9f3e85)
sharp carbon
for the "only an optimization" style this is all that's necessary to implement <...
note that i haven't verified that this works correctly for linux-style vdsos (where the vdso provides symbols with the same name as libc, LD_PRELOAD style)
unlike linux's, hydrogen's vdso is listed in DT_NEEDED and uses unique symbol names not provided by any other objects, and for that usage style it works correctly
but idk how that code interacts with vdsos that aren't listed in DT_NEEDED (does it get inserted into the symbol resolution order at all? if so, where?)
sharp carbon
note that i haven't verified that this works correctly for linux-style vdsos (wh...
looking at the linker code, if you add vdso_soname to the preloads vector the linux semantics should be implemented correctly
grand shadow
It would be interesting to have upstream mlibc vdso support with jinix being a thing now and other hobby kernels looking into it
opal lotus
@sharp carbon 
sharp carbon
yeah yeah i know
grand shadow
If I get the hooks done today depends on that the bf wants to do
grand shadow
that is done
gotta figure out what broke dns now
ok so mlibc expects an /etc/services now
and if that is not present it just does not do dns
ugh
I mean its in posix so I gotta
carmine swallow
even windows has etc/services (and etc/protocol + etc/hosts) in C:/Windows/System32/drivers/etc/
grand shadow
there we go that works now
analog berry
Damn
Will astral be the only hobby os with a working chromium port at this pace
I wonder what you're even missing for that
Well DRM I guess
bronze nexus
It's really in outer space compared to the other OSes
analog berry
Real
grand shadow
Will astral be the only hobby os with a working chromium port at this pace
still a loong way
theres kernel work to do but I will take a small detour on a lil project
analog berry
theres kernel work to do but I will take a small detour on a lil project
What's the project?
fleet osprey
nvm this too cancer
average experience after working on the PS/2 driver (definetly what you were doing )
grand shadow
So true
I will do some mlibc sysdep work now, make some scheduling and smp syscalls etc
grand shadow
for getcpu I will just probably use rdtscp
actully taht could be intel specific hm
lemme check the amd manual
wait no I'm dumb I think its on both
vdso looking mad fine rn
might just query cpuid for it in __mlibc_entry and if it is present go through the fast path
grand shadow
nah just gonna keep it simple and slow
grand shadow
or nah I'll do that another time, just implementing getcpu
grand shadow
noooo updating mlbic seems to have broken links
@prime juniper what does sys_inet_configured search for exactly?
just if its running with ipv4 or ipv6?
prime juniper
I’d need to check
One sec
Seems like it
Tho it does some rtnetlink stuff in our sysdep
@formal patio might know more
grand shadow
I guess I can just hardcode it ipv4 = true
I dont support nor do I plan on supporting ipv6 for quite a while
otherwise I would have to implement a way to iterate through interfaces and check if those interfaces have an ip assigned to them from netd
and I refuse to implement netlink
yeah it really doesnt matter if theres no interface configured it'll just get a routing error in its face cuz of no route in the ip tables
formal patio
<@261497806307459072> what does sys_inet_configured search for exactly?
It's for querying what posix calls checking if Internet is configured
So if you have no link up or should return false
formal patio
otherwise I would have to implement a way to iterate through interfaces and chec...
You should do that tho 
rtnetlink is good tho
Better than any other of the cursed interfaces for that that I have seen so far
grand shadow
oh well
if it breaks something in the future I do that otherwise wawawawa
added it to the mlibc pr as well so I do not have to keep it in the local patch
grand shadow
I should finish setting up xbps and see if I can download and install a package
grand shadow
syscall return: pid 41 tid 42: 18446744073709551615 EAFNOSUPPORT (956 cached pages, 501055 free pages)
proably uz of this
oh right, I dont think I have special stuff in the type field
like cloexec
BROTHER I AM RUNNING AS ROOT
lmao
astral too secure
robust geyser
x86_64-repodata btw :^)
grand shadow
oh woops ty
this is so weird, no syscall in that process fails with EPERM
oh seems like xbps does return EPERM itself for some functions
oh right I'm blind
certificate veritfication failed
wait I dont think I install the certificates in the minimal distro
5head moment
robust geyser
can't wait to be able to install astral from a live usb
you should port dosfstools and e2fsprogs :^)
also something you can use to partition the drive
then you could do a full astral install from within astral
which would be very cool
grand shadow
I got e2fsprogs and I plan on porting parted
though I still need a fat driver for the efi boot partition and some real nic driver
grand shadow
astralbros we are so back
analog berry
Damn
neon crane
astralbros we are so back
implement the way neofetch does uptime next
cyan junco
implement the way neofetch does uptime next
/proc/uptime
grand shadow
wtf windows supports /proc real
cyan junco
wtf windows supports /proc real
OMG
neon crane
wtf windows supports /proc real
TIL
cyan junco
wtf windows supports /proc real
maybee it don't actually
and it alaways fallback to the commands stuff
grand shadow
maybe I could start adding some stuff to proc
analog berry
$ cat /proc/gdt
$ ... OK
REAL
grand shadow
I do wonder, how does procfs handle say, a process doingle multiple reads of the same file
@analog berry linux man do you know
like reading half of it and then the other half
sharp carbon
iirc it prints the content into a kernel buffer on first read and then returns from that buffer
analog berry
<@353593948523986945> linux man do you know
nope
havent looked into the vfs and that stuff at all yet
pine ledge
astralbros we are so back
Wow bro, amazing work seriously
sage locust
real
sage locust
that's a decent pick
grand shadow
fixed a bug in vionet that randomly broke networking
the bug?
a typo.
I typed tx instead of rx
fml
spent like 3 hours looking for the bug
but now! networking seems perfectly stable
I wanna see what error jinx makes
awwwww
it doesnt explode it just says fuck off
fleet osprey
absolutely unacceptable, we need Astral support in Jinx
grand shadow
uploading all package files I have to the repo
so that theyre there if needed
ig that means astral has working package management now
yippe
sage locust
that's really cool
@grand shadow update Jinx, i upstreamed a new release :3
it makes it faster and whatnot
grand shadow
yippieeeee it did work (after some trial and error to find out which libs to install... I really need to make an xorg-base meta package or something)
grand shadow
<@215823874615541761> update Jinx, i upstreamed a new release :3
will do tomorrow, it is 0:41 and I have to wake up 6
sage locust
gn then lol
grand shadow
gnnn
fleet osprey
will do tomorrow, it is 0:41 and I have to wake up 6
imagine not going to bed at 1:30AM and needing to wake up at 5:40AM
(I do this often)
grand shadow
mlibc did the thing
fleet osprey
lmao
grand shadow
well damn, locale being a stub in mlibc just breaks a very specific part of parted
fucking atari-fdisk
noboduy will use that
so I am just removing the init functions!
okay I got into a prompt
funny
%P terminator
fucked up?
(C99; not in SUSv2) For a conversion, the double argument is converted to hexadecimal notation (using the letters abcdef) in the style [-]0xh.hhhhp±; for A conversion the prefix 0X, the letters ABCDEF, and the exponent separator P is used. There is one hexadecimal digit before the decimal point, and the number of digits after it is equal to the precision. The default precision suffices for an exact representation of the value if an exact representation in base 2 exists and otherwise is sufficiently large to distinguish values of type double. The digit before the decimal point is unspecified for nonnormalized numbers, and nonzero but otherwise unspecified for normalized numbers.
this is the only thing I found where P is mentioned
wait no
tomorrow I'll debug this more
sage locust
insans
grand shadow
I am just a parted port, dosfstools and a fat driver away from having this be able to install a distro of itself, exciting stuff
Maybe even only a parted port if I do mbr 
robust geyser
you still need the fat partition for limine to load limine-bios.sys from
and the config
and the kernel and everything
grand shadow
Wiay I thought lime went back to supporting ext2
robust geyser
unless you go with grub, but i don't remember astral having mb2 support
grand shadow
Did that change again
robust geyser
a long time ago
grand shadow
Damn
grand shadow
hm
oh wait this is decimal 50
I was looking at hex 50 oops
its complaining about the %....2
that does seem like a valid printf?
example from man 3 printf: %2$*1$d
ughh this is gonna be fun
oh wait seems like mlibc only handles it in scanf
well """handles"""
as in, stubs
actuall wait
my frigg is probably like outdated as hell
maybe its been implemented in printf
wait no in frigg it seems to be handled but is breaking somehow?
oh yeap
seems like even after upgrading frigg it still dies
oh thank you beos devs
gonna open a mlibc issue for this so someone smarter than me can figure this out
what does bro want me to do
something is very wrong here
maybe its buggy cuz I dont have it working with readline let me try with it
grand shadow
still buggy, time to see whats it even reading back from that scanf
funny
surely mlibc's scanf supports %lf?
wait does mlibc scanf not even support %f
💀
oh wait I do remember, it was blocked on them trying to decide which library to use to handle the float conversion..... and it has been so for years now
genuinely considering dumping parted and just doing util-linux fdisk at this poin t
or writing my own
or I could also add a very barebones %lf support to mlibc locally
yeah doesnt seem to bad
float/double/long double tmp = 0;
float/double/long double div = 10;
get characters until dot or null
if not dot then tmp = tmp * 10 + digit
if dot then tmp = tmp + digit / div, div *= 10
then set the pointed address passed in scanf to tmp
grand shadow
small detour to implement that
still not working on real disks though, need to check my port
:D
as one last test I will add a second disk, add a gpt to it, add a partition, and install bash and chroot into it
then I go eep
grand shadow
damn, it all worked until xbps-install
it just doesnt want to work with a --rootdir for some reason
maybe I'm misunderstanding that option
figured it out ig?
seems like you have to copy the keys and the xbps config before doing the command
and that works
so yeah it is that
ugh to lazy to do everything now
at least I know about that xbps thing for an eventual installer
bronze nexus
genuinely considering dumping parted and just doing util-linux fdisk at this poi...
I use sgdisk
grand shadow
@carmine swallow I do wonder if the missing float handling in scanf is the reason why mlibc factorio does not wotk
It also causes parted to not work, xclock to not tick (and anything from xorg with float configs really) and xscreensaver-gfx to segfault
I wonder if thats also why fvwm had those funky windows, I didnt really boot into the wm since adding that patch
grand shadow
did some more mlibc work, now finishing the float conversions to (hopefully) be able to upstream
grand shadow
okay that is doneish
after my other pr is upstream I will open a pr for this
time to see if I can get meson and ninja ported
grand shadow
ok I got ninja but still trying to figure out meson
prime juniper
Yeah I’m not looking forward to packaging python modules
robust geyser
Yeah I’m not looking forward to packaging python modules
you can just look at jinix :^)
also i think i made a function for packaging perl modules if you ever need that lol
grand shadow
we do a little
wait FUUCK I have to get the linux headers in too
thank you scritps folder for the get linux headers script
trolled
guess I gotta port xz
prime juniper
you can just look at jinix :^)
Yeah I’m aware for both
prime juniper
guess I gotta port xz
Basic stuff
prime juniper
Real
grand shadow
Wonder how difficult it is to get mesa and virtio gpu accel working
If that needs drm then boooring
Damn it do
grand shadow
udrm when
I might take a look at it some time in the future then
At least for virtio gpu accel
grand shadow
Becoming the thing I swore to destroy :(
bronze nexus
I'm still wondering if it'd be possible for me to contribute this kinda thing to the OSDev community, but my Rust codebase would make that hard to do
Not that I'm at DRM yet at all
grand shadow
I need to see what causes unpacking a tar archive to take this long, in theory it should all go straight to the page cache
but unpacking the linux source is being sloow
espeiclaly when it gets to drm
sharp carbon
probably file creation speed if I had to guess
the classic lots of small files is slow thing
formal patio
just implement profiling infra 
profiling stuff on managarm has revealed many unexpected timesinks, some of which turned out to be avoidable
grand shadow
okay I think I found the issue
oversight on my part when implementing ext2
when growing a file I actually allocated the blocks instead of just creating a hole in the end that would get filled later
so the later reads would read the page fo the disk from there instead of memseting a page to zero
grand shadow
BRUHHH
I waited all this time
for this
its stable but there is something slowing it down
grand shadow
I need to figure out what causes it to slow down so much after a while
proper meteor
I need to figure out what causes it to slow down so much after a while
have you tried profiling your kernel?
I have a tool to do it through QEMU: https://github.com/maestro-os/kern-profile
although I haven't used it for a long time so it may be broken
grand shadow
I can take a look later
I do think its how Im hashing the vnode, offset for the cache table
Ill confirm that in a bit, gotta cook lunch
grand shadow
ok it is 100% the hashing I just need to do a field test by checking how many entries are put in every bucket
grand shadow
nvnm
I think it was just filling up the 2gbs of memory and having to do more disk io
linux does use a few gigs
- the other stuff
ah I'll figure that out later I want to see if I can compile mlibc
so I'll add the headers to the sysroot and then port rsync
grand shadow
so meson runs but dies
dying on scandir makes me thinkg it is passing too small of a buffer to readdir
but meson does still seem to be kind of working and complaining
does not seem to be that
syscall log is needed, then
also it really wants sbrk() for some forsaken reason
grand shadow
I have been spending time with bf so havent looked into it
Tomorrow the fun resumes
I also wanna look into getting a fatfs implementation running
neon crane
I have been spending time with bf so havent looked into it
consider telling him about it
if he's also a programmer he could join
proper meteor
@grand shadow I am looking at your semaphore code and I have a question
https://github.com/Mathewnd/Astral/blob/1aeb85967754916ea9b236c7e1e58c167c65c564/kernel-src/lib/semaphore.c#L80
Can't a process get woken up for another reason than the semaphore being released? (example: woken up by a timer)
grand shadow
<@215823874615541761> I am looking at your semaphore code and I have a question
...
They can have a different wakeup reason in that case
I dont think theres any code off the top of my head that sleeps on a semaphore but has a timer
Unless its like an userspace timer, which just gives it an interrupted reason and sends EINTR down
neon crane
is he younger than you?
grand shadow
is he younger than you?
A few months, hes just switching from maths to cs
He was doing physics then switched to math then wants to switch to cs now
proper meteor
They can have a different wakeup reason in that case
oh right, sched_yield returns a wakeup reason
grand shadow
I feel like working on a fat driver so I will put the mlibc compile stuff on hold
gonna do a driver for fat{12,16,32}
fleet osprey
NTFS driver when
grand shadow
NTFS driver when <:trl:829316641191886888>
analog berry
vfat is just an extension that applies to all fat widths
grand shadow
Yesterday I got the typedefs and defines and stuff done, now Im going to actually do the fat
Start with mounting, dir traversal, fat traversal, etc
I wrote a fat32 implementation to lyre many eras ago and so I know what to expectish
grand shadow
This will hopefully be the final stretch before an installation program
Its not that hard, but I think ext2 js easier
Especially for integrating in the vfs
sharp carbon
imo it depends on how advanced you want the driver to be
a basic fat driver (no lfn, bad performance) is easier than a basic ext2 driver
but a good fat driver (lfn, good performance, etc) is harder than a good ext2 driver
sharp carbon
long file names
aka more than 8 chars before the last dot and/or more than 3 chars after it and/or mixed case names and/or Unicode names
analog berry
what's lfn btw?
Multiple chained directory entries to make one long name
proper meteor
thx
opal lotus
gnu tar has the same concept i think
formal patio
it can be a surprisingly large hit
we've noticed that on managarm where our block allocation on ext2 was suboptimal but that lead to severe issues in some circumstances
optimizing placement + flushing of the bitmaps basically made writing out a coredump go from 60+ sec to sub 100 ms
grand shadow
did more fat work, working on directory parsing
proper meteor
did more fat work, working on directory parsing
<insert joke about grease/oil>
bronze nexus
obese filesystem
grand shadow
<insert joke about grease/oil>
opal lotus
<insert joke about grease/oil>
<grease/oil.h>
grand shadow
I am still pondering whether I should port openrc or keep my simple init
having a more complete init system would be cool, but also having to implement more linux shit that openrc would likely expect would be ugh
I mean it runs on bsd so cant be THAT bad
fleet osprey
it runs on BaSeD 
opal lotus
I am still pondering whether I should port openrc or keep my simple init
just keep your init
i am porting openrc because i actually implement linux interfaces
prime juniper
i am porting openrc because i actually implement linux interfaces
Systemd when
high whale
systemd demons
grand shadow
So excited for when I get fat to fully work and can write the install script finally
Then its a matter of writing a bunch more drivers for real hardware and forcing people to install astral
I dont have any real hw NICs and the only real disk driver is nvme
analog berry
damn the year of astral desktop is close
grand shadow
damn the year of astral desktop is close
Would have been last year if I didnt have the habit of vanishing for months at a time because I burn out
I think for the installer Im just gonna install a base system (maybe offer both a networkless and networked install images) and have that base system have everything to xbps-install more packages
arch astral
analog berry
yeah thats nice
analog berry
Would have been last year if I didnt have the habit of vanishing for months at a...
be like me and spend weeks on useless stuff
analog berry
first os here to ship an iso you can use as an installer
i think some osses here can do that
the one in basic iirc
robust geyser
ok but that one took every shortcut you could take lol
analog berry
ig
robust geyser
astral makes no compromises
robust geyser
yeah :^)
can't wait for linux namespaces support so you can build the full distro on astral itself
grand shadow
Self hosting work stopped on mlibc, meson ran into some issues which I didnt debug because I wanted to get fatfs
But gcc and binutils and many other things do build just fine
Including the kernel
So clearily, I blame this on managarm picking a bad build system :^)
proper meteor
first os here to ship an iso you can use as an installer
(actually I did it too)
but my installer is not great
proper meteor
So excited for when I get fat to fully work and can write the install script fin...
I guess you need FAT to read the USB stick's filesystem? How would you determine which device is the one with the installer on it?
My installer uses an initramfs with everything in it so I didn't have to figure this out 
hearty fulcrum
(actually I did it too)
proper meteor
https://tenor.com/view/cat-edit-i-saw-what-you-edited-i-saw-gif-27672816
liar
banned
robust geyser
I guess you need FAT to read the USB stick's filesystem? How would you determine...
just do what most linux-based usb installers do and look for a partiton guid or partiton label
proper meteor
makes sense
grand shadow
Its mostly for limine, I dont have xhci
And if I were to use a filesystem for the usb itd be ext2
neon crane
Its mostly for limine, I dont have xhci
then how would you load the data to install
just load it into RAM?
grand shadow
then how would you load the data to install
The internet or just load into ram yeah
The base system is not that big
I do have to write a usb stack at some point (or force some poor soul to do it for me)
analog berry
I wish someone made a uUSB
grand shadow
I wish someone made a uUSB
Be the change you want to see in the wo
I also still need to unfuck ps2 :^)
hearty fulcrum
I wish someone made a uUSB
uSB
analog berry
You picked the name, now all that's left is code
flat copper
unironically you can calamares for installer :^)
prime juniper
It’s what I’ve looked at for Managarm at one day
opal lotus
@grand shadow do you have a CI running for xbps packages
or did you simply upload the packages to the server
robust geyser
the latter
grand shadow
Yeah I simply uploaded, its not something super nescessary for now
grand shadow
rotund filesystem
fat12 and fat16 work, fat32 seems a bit borked so figuring out whats going on
oh right, forgot to set some metadata for the root cluster
bronze nexus
Wait you didn't have FAT yet
grand shadow
no
grand shadow
alright, fat32 works now, had a super stupid typo
so that is a working fat read-only driver
still gotta do write and optimizations
but thats a start
grand shadow
I am inching closer to getting write working, been busy
after that I will upstream some mlibc stuff
that has been sitting around for a while and I've been meaning to upstream
grand shadow
you know, why is fsinfo a thing if the data there is not guaranteed to be right
ugh
I love fat
grand shadow
alright, I got it to be able to cp /usr/bin/neofetch /mnt/EXISTING_FILE
dosfsck complains about freeing unused clusters, so I gotta check whats up
oh wait it could be because I am not writing to all fats
grand shadow
alright, dosfsck is happy now
so I think file read/write is done
just need to do directories
which I will do tomorrow
neon crane
alright, dosfsck is happy now
chkdsk is the canonical disk checker, so consider testing w that as well
neon crane
wtf thats tiny
my readonly ext2 driver looks like this
meanwhile nanoshell's read+write ext2 implementation looks like this
bronze nexus
loc != complexity
neon crane
sure
analog berry
and fat12/16
ofc, yeah
grand shadow
wtf thats tiny
skill /j
its not the best thing in the world, I'm gonna optimize it more when I'm done with making it read/write without exploding
grand shadow
chkdsk is the canonical disk checker, so consider testing w that as well
probably yeah
grand shadow
Astral running on a satellite when
How to bring down a sattelite
grand shadow
the average line length is 300 characters <:trl:829316641191886888>
More like 400
grand shadow
time for directory operations
still missing those
link I just have to support linking a file with no disk dents
grand shadow
there were some troubles with long filenames that are now resolved, and I have written a function to create a dent in a dir
just gotta glue it and I will have create and link
neon crane
link I just have to support linking a file with no disk dents
fat has neither hard links nor soft links
so ENOTSUP time
grand shadow
fat has neither hard links nor soft links
Yes but you can still support something like keeping a file unlinked and then linking it again while keeping nlinks <= 1
neon crane
Yes but you can still support something like keeping a file unlinked and then li...
link() takes a path name. once you unlink something you can't link it again because the (1) path name is no longer there
grand shadow
Something like that can work in my design
Since I pass vnode pointers and the final pathname
So you could use linkat with AT_EMPTY_PATH to relink it at the same filesystem without a loss of data
Obviously wouldnt work across filesystem boundaries
And the file data gets released on a refcount == 0
grand shadow
fat64 support for large disks
this does impress me as to how neofetch manages to be a big shell file
shit uses over 200kb
fleet osprey
this is why fastfetch is better
grand shadow
I think VOP_CREATE for fatfs is done
tomorrow I will do VOP_UNLINK and VOP_LINK work
and VOP_INACTIVE as well to go with VOP_UNLINK
which will leave just VOP_RENAME leftr
grand shadow
Did some unlink work, it works perfectly with short filenamew. Havent tested a long filename dent yet
grand shadow
Hoping to get fat work done soon so I can work on something akin to a install util (or at least for now install instructions like the gentoo manual) and then go back to doing self hosting work
Though if I ever want this installable on real hardware Im gonna need at least a not shit ps2 driver, ahci, some nics and maybe xhci
Contributions welcome
grand shadow
only two major functions left
will do that at some other time
then do the few remaining TODOs I feel are important, then do some more hands on testing aka install limine and boot out of it, and then optimization time
but glad to see fat working
gonna be real exciting when I cook up a install manual
grand shadow
VOP_RENAME is done
and I did VOP_SYNC in the meantime
so all of the vfs glue has been implemented
time to try something fun out
but for that I need to port dosfstools
grand shadow
And how do u generate inode numbers
it is a bit cursed
but since fat only allows one hard link per file, I use the offset of the directory entry on disk
this is to keep it consistent with readdir() without having to allocate and keep the readdir() vnodes in memory
of course, if the file gets rename()d it breaks
but blame the filesystem for being shit
grand shadow
but since fat only allows one hard link per file, I use the offset of the direct...
this only works because I have a 64 bit ino_t and will never support a 32 bit machine
analog berry
Idk supporting 32 bit kernel builds at this point seems dumb af
32 bit userspace I guess, but the kernel is pointless
grand shadow
How big is the driver now
analog berry
Tiny
analog berry
Do u cache fat chains?
I cache vectors of contiguous spans in my driver
So best case is one entry, worst case is entry per block
grand shadow
I dont, at most it uses the page cache on the disk to at least not have to read the disk every next cluster
analog berry
Oh well I guess that works
grand shadow
I am not super worried about making it very fast, since at most it'll be used as the boot partition and maybe to move stuff in a usb drive when I get an xhci driver in a trillion years
analog berry
Still kinda nice to abuse the fact that most files will be contiguous so u save tons of cycles and also have fast binary search lookup
bronze nexus
of course, if the file gets rename()d it breaks
I solve this by faking the concept of inode numbers entirely for FAT.
Every time a VNode is opened, I just atomic_fetch_add a fake ino counter
And rely on the namecache for making sure VNodes are opened only once
analog berry
So if you reboot the PC all inode numbers are gone?
bronze nexus
Not just reboot, even dropping the last FD to a VNode will change the ino
analog berry
☠️
bronze nexus
But here's the kicker: It's FAT and I only need it for bootfs so whoms't care
analog berry
I guess
bronze nexus
it does not violate POSIX semantics if you ignore the part where a file will have its ino changed even though it's still the same file
It's also pretty unlikey an application will rely on checking the ino but not opening the file and getting confused when it checks the ino again
Which would be vulnerable to that race condition with proper POSIX vfs anyway
grand shadow
what the fuck is parted cooking
bronze nexus
I forgot who suggested it, IIRC hyenasky did. Fixed all the problems I was having with FAT and ino
scenic epoch
what the fuck is parted cooking
dont let it cook
grand shadow
wuh
why is there a hardlink to
ld
ld.bfd
and why did this only happen after adding parted
on ubuntu this is not a hardlink
oh according to blfs it is a hardlink
ah whatever
I will just implement the vfs_link call in initrd.c
oohohhhh
I just never put binutils in the initrd before?
no I did
but now theres a package that pulls it
what
I'll add verifying if libuuid really needs binutils in userspace to my todo list
grand shadow
I am cooking I am cooking
grand shadow
can it run doom? <:trl:829316641191886888>
yea
grand shadow
can run quite a lot of things
scenic epoch
you have achieved peak
grand shadow
scenic epoch
can run quite a lot of things
like what?
grand shadow
like what?
idk bash gcc x11 wine java
scenic epoch
idk bash gcc x11 wine java
nice
wine is peak
port qemu on it you'll get something somewhat self sustaining /hj
grand shadow
port qemu on it you'll get something somewhat self sustaining /hj
about that
fleet osprey
holy shit based
scenic epoch
about that <:troll:885239769557266432>
congratulations you can daily drive it now
just run linux on qemu
grand shadow
so true
grand shadow
I wish I could update the screenshot on the thread, that one is not even this current iteration of the os
scenic epoch
I wish I could update the screenshot on the thread, that one is not even this cu...
i hate this discord needs to fix it so fucking bad 😭
grand shadow
audio was being worked on by qwinci but that is now deep in pr limbo
scenic epoch
they probably expected forums to be used like actual forums as in like ask question, receive answer
definetely not long term
grand shadow
I AM COOKING
grand shadow
cooking
grand shadow
these fucking unices i cant understand anything 😔
I am about to do something epic
grand shadow
oh wait I gotta update the packages in the repo, been a while
grand shadow
RAHHHHHH
forgot to add the kernel to the xbps install command oops
unpacking stuff......
:3c
I forgot to pass the flag to xbps install to not configure the packages so some stuff like sudo mkight be missing setuid
but a system there is!
and the kernel
time to try to install limine to that disk
oh yeah I can also chroot
aghh I'm just so excited
this is so cool
bruhh
how the fuck
I copied it from the /usr/share !!!
maybe it corrupted or something?
oh it is all zeros
I think I forgot to sync
grand shadow
no I'm trying to install astral to disk
grand shadow
I partitioned, made the filesystems, installed the base systems thru the package manager and now I am having some issues
nah
ok yeah I defo forgot to sync the fat fs
I'll just reformat it
HAHAHAHAHAHAHAHAHAHAHAHA
first boot of an astral install
that has been installed from the package manager
with the partitions made in astral
and the filesystems formatted in astral
gg
and the bootloader installed from astral
@sage locust look limine works in astral
and xbps seems to be working from the new install, installing gcc
oh oop I think that sync I forgot finally caught up to me
tomorrow I will do this again
and document all steps needed
grand shadow
Im still thinking about the init system stuff, I might port sysvinit
I am slightly familiar with it and its much less linux specfic I think
Compared to something like openrc
grand shadow
you should implement shutdown :3
it is I just forgot to
though its named poweroff instead
it does the normal SIGTERM wait 5 secs SIGKILL sync poweroff
(in init)
poweroff itself writes to /tmp/.initcmd
which is a fifo read by /init
grand shadow
limine 8.x holy
yeah it is a bit old
fleet osprey
last updated when I updated the config file
do I need to become the certified Limine Updater for Astral
grand shadow
I dont have the patience to update userspace stuff so prs always welcome :^)
fleet osprey
tossed it onto my todo list
I'll get back to you in 3-5 business years
(said list is over 100 lines long)
sage locust
damn awesome!
prime juniper
I dont have the patience to update userspace stuff so prs always welcome :^)
Sorry not sorry but unless I manage to land a full time paid job to update userspace components of hobby OSses I’m already behind with Managarm (and that’s only cuz menix doesn’t use my ports yet and mint keeps Gloire up to date 
)
Else I wouldn’t mind doing it (but time)
formal patio
Im still thinking about the init system stuff, I might port sysvinit
systemd
grand shadow
Sorry not sorry but unless I manage to land a full time paid job to update users...
I will pay you a beer and some paperclips per month :meme
grand shadow
systemd <:meme:575445094589661215>
too much linux stuff needed smh
okay so, from what I remember from yesterday, the install process is:
format disk with an astral partition and a boot partition:
parted /dev/DISK
reboot
set up networking:
netd INTERFACE
make filesystems:
mkfs.ext2 /dev/DISKp1
mkfs.fat /dev/DISKp2
mount them:
mkdir /mnt
mount -d /dev/DISKp1 /mnt ext2
mkdir /mnt/boot
mount -d /dev/DISKp2 /mnt/boot fatfs
prepare xbps in the new root:
mkdir -p /mnt/usr/share
mkdir -p /mt/var/db/xbps
cp -r /var/db/xbps/keys /mnt/var/db/xbps
cp -r /usr/share/xbps.d /mnt/usr/share
install the base system:
xbps-install --rootdir /mnt -S -y astral mlibc bash coreutils init distro-files vim nano mount shadow sudo xbps net-base neofetch limine dosfstools e2fsprogs
chroot into the installed system:
mkdir /mnt/dev
mount /mnt/dev devfs
chroot /mnt
source /etc/profile
cd
write a limine.conf and copy required limine files to /boot (from now on using bios mbr):
cp /usr/share/limine/limine-bios.sys /boot
install limine
limine bios-install /dev/DISK
and at last:
reboot
I need to add some sort of command to allow you to not need to reboot after partitioning
like an ioctl on the raw disk or something
I remember the linux parted port having something like that
I should also probably make a base system metapackage
maybe make a sort of xbstrap script like arch has pacstrap
sage locust
xbstrap you say?????? :p
grand shadow
holy shjit mnaganarm reference
hearty fulcrum
lmao
grand shadow
time to try these steps
I really gotta make syncing faster
maybe have more worker threads that just do the i/o
wild marsh
maybe make a sort of xbstrap script like arch has pacstrap
not sure what the script would need to do, xbps supports a sysroot arg already
grand shadow
not sure what the script would need to do, xbps supports a sysroot arg already
more so copy the keys and config to the sysroot
since the --rootdir argument uses the stuff from inside the sysroot
grand shadow
I mean gentoo does that
and void has an option to either do the config copy or tarball
either way, the install instructions work
just installed a perfectly working system and ran xbps-install gcc inside it
20 minute video and like a good chunk of it is syncing disks 💀 I really need to make that faster
robust geyser
I remember the linux parted port having something like that
blkpg :^)
it's an interface that parted uses to tell linux about partition addition, deletion and size changes
but if you don't want to support linux specific things maybe you could add support for astral in libparted, idk how hard that would be
grand shadow
but if you don't want to support linux specific things maybe you could add suppo...
I did add an astral port to libparted
robust geyser
ah that makes sense lol
you're probably familiar with what's needed for partition probing then
sage locust
that's really cool
analog berry
20 minute video and like a good chunk of it is syncing disks 💀 I really need to...
port fio to stress your disk/fs drivers
very nice tool
also nice for finding fs bugs ig
grand shadow
Will keep in mind ty
analog berry
but the progress you've achieved is crazy still
literally maybe some speedups + drm and its daily drivable
neon crane
very cool
grand shadow
literally maybe some speedups + drm and its daily drivable
Yeah some stuff is just too slow for daily use
And drivers
I need to have a lot more drivers
flat copper
20 minute video and like a good chunk of it is syncing disks 💀 I really need to...
if this is wsl2 you should at least try windows qemu with whpx, nvme with nested virt has very weird performance at least for me
grand shadow
if this is wsl2 you should at least try windows qemu with whpx, nvme with nested...
Hmm it is wsl2, Ill take a peek lTer
bronze nexus
holy shjit mnaganarm reference
Now I'm gonna call it Mnaganarm forever
prime juniper
.!t managarm
rigid timberBOT
prime juniper
(Or derivatives )
grand shadow
mammogram
scenic epoch
mammogram
is it an inside joke or did you read my brainfart from yesterday
lounge 0
analog berry
my hello world is somehow 1/3 of that and it doesnt do anything
or actually thats probably because tests make up a lot of the code and u dont have those
so thats why
gritty shore
my hello world is somehow 1/3 of that and it doesnt do anything
dont you have like
an overengineered logging system
analog berry
its only a few k
gritty shore
its only a few k
why tho
analog berry
idk
gritty shore
whats the use of it
analog berry
#1385970208631427173 message
opal lotus
quantity != quality
analog berry
yeah im surprised its that small
managarm is probably way bigger
if you count frigg etc
or even without it
analog berry
I have been told I write dense code so
And tbh u barely have drivers right
grand shadow
And tbh u barely have drivers right
yeah I havent spent a lot of time writing drivers
so I am missing a lot of lines of code lol
qemu with whpx works on my desktop but not on my laptop
this is funny
grand shadow
you know, between sysvinit and openrc I might as well just pick the nuclear option
and have both like gentoo does
gritty shore
runit
prime juniper
you know, between sysvinit and openrc I might as well just pick the nuclear opti...
Systemd. Easy
scenic epoch
Systemd. Easy
read what he said about systemd 😔
prime juniper
I’m aware
grand shadow
okay I think I know why meson is complaining when building mlibc
apparently I am missing a very specific F_DUPFD_CLOEXEC op
tbhank you meson
BWAHAHAHAHAHA
that was it
that was so stupid
running ninja...
seems to be working
gg
and it installs...
wild marsh
now ninja install without DESTDIR
grand shadow
grand shadow
now `ninja install` without DESTDIR <:trl:829316641191886888>
so true
actually true I could tyr that
grand shadow
now `ninja install` without DESTDIR <:trl:829316641191886888>
ez
wild marsh
nice
grand shadow
astral first os to actually build mlibc inside mlibc
big step towards self hosting
I really want to try building some of the other base system packages now
lets see, I know these build:
binutils, gcc, mlibc, the kernel
I gotta try gettext, intl, libxcrypt for the base libraries
coreutils, bash, ncurses, readline for shell stuff
some astral tools like init, netd and mount are just simple c files on makefile, I wont even bother
maybe some tools like grep, sed, make, tar
and the libraries those would need
something tells me to try doing astral from scratch..
grand shadow
I have been thinking about how I would approach that stuff
if I would do /proc or some other astral tool
and how /proc would work
wild marsh
one thing i was thinking about is making the proc files binary and providing utils to write out the linux-style output from them
(not in managarm ofc)
wild marsh
one thing i was thinking about is making the proc files binary and providing uti...
this would also extend to /proc/<pid> being special files instead of dirs
grand shadow
you should port fastfetch
What is the benefit other than the creator of neofetch is mia
analog berry
neofetch is abandoned
sage locust
neofetch is dead and fastfetch is actually genuinely much better
hearty fulcrum
What is the benefit other than the creator of neofetch is mia
it's fast :P
proper meteor
it's fast :P
And also fetch
opal lotus
What is the benefit other than the creator of neofetch is mia
it's written in c/c++
and has more information
grand shadow
what more do I need before I can have the satisfaction of calling astral fully self hosting and farming reddit karma?
analog berry
Just show a timelapse of downloading the sources, then building and running in qemu, all from inside astral
grand shadow
honestly if I make astral faster I think I can do that 💀
I need to track down what is making it sluggish sometimes
I have a gcc build running in the background and its not really an acceptable speed
analog berry
I have a gcc build running in the background and its not really an acceptable sp...
If you enable external profiling you will probably notice what's taking most of the time real fast
grand shadow
I'll take a look
also I just noticed
I'm gonna have to port perl to have autotools proper
perl doesnt work in managarm very well IIRC but idk if its managarm or mlibc
grand shadow
If you enable external profiling you will probably notice what's taking most of ...
google is not being much of help, is it a qemu thing or the gcc -pg thing
analog berry
Well there are different ways to implement it
You can just dump qemu registers every N us, or record stacks in your kernel
grand shadow
I think having it in kernel would probably be better, when I did the qemu register dump thing it was insanely slow after a specific amount of time per dump
analog berry
I think having it in kernel would probably be better, when I did the qemu regist...
Yeah I think so
wild marsh
we have profiling via pmus, the pmu is configured to raise an nmi after every n clock cycles, and in the nmi handler we log the ip where we were interrupted
grand shadow
wait
I think the reason why it might have been super slow
is that it was building gcc with -g by default???
why does gcc do that
I'm gonna leave it building with "-O2 -pipe" and without that -g to see if it helps any
while I look into the profiler
analog berry
Isn't g just debug info
marsh sparrow
(not in managarm ofc)
for what else?
grand shadow
Yes but my fs driver isnt very fast and its a lot of info
marsh sparrow
perl doesnt work in managarm very well IIRC but idk if its managarm or mlibc
I don't really know what the progress on perl is but i do know that perl is pretty anal about locale support
wild marsh
Isn't g just debug info
its a not insignificant amount of data to write out
wild marsh
for what else?
some personal project, likely paaliaq when/if i get to that point
marsh sparrow
paaliaq?
marsh sparrow
ah
analog berry
fork managarm
wild marsh
fork managarm <:meme:575445094589661215>
that'd require having a c++ compiler :^)
grand shadow
port gcc
grand shadow
I don't really know what the progress on perl is but i do know that perl is pret...
yeah if perl does not work well enough for automake I might be in for some trouble if I want to fully self host
bronze nexus
lily-cc 
grand shadow
when then means I gotta research into locales and implement them in mlibc
wild marsh
that'd require having a c++ compiler :^)
another issue is that the cpu has 4x less bits :^)
grand shadow
get 4 cpus then?
wild marsh
incredible
grand shadow
that was suspiciously quick?
make -j2 with 8g ram and vioblk
and 4 cores
ls is super slow for some reason
something is funky
marsh sparrow
@prime juniper probably knows if perl works on mlibc linux
grand shadow
I think something is funky with the page cache
I really need to rewrite this at some point
and do it completely differently
analog berry
fs code?
grand shadow
page cache code
I think I do it the way svr4 did which is like just not good
or some other old unix, I rmeember getting inspiration from oen but I forgot which
prime juniper
I don't really know what the progress on perl is but i do know that perl is pret...
It’s meh at best. I have a working 2048.pl but installing automake and friends and invoking it gave module errors
grand shadow
It’s meh at best. I have a working 2048.pl but installing automake and friends a...
did you ever figure out if it was mlibc or managarm
prime juniper
Nope
neon crane
@grand shadow as a shortcut couldnt you just compile the kernel and then copy all the userspace stuff from the host into the guest
or is that cheating
grand shadow
<@215823874615541761> as a shortcut couldnt you just compile the kernel and then...
I mean you could, thats kind of the same as just installing the rest of the userspace with a package manager while compiling the kernel
i want to be able to build the userspace stuff in astral
sage locust
tbf what you have already counts as self hosting to me
you can literally make git commits inside Astral and push them for your kernel, and then rebuild said kernel, install it, and reboot
it honestly doesn't get more selfhosting than that as a kernel
btw idr if you tried building Limine inside Astral too, but i don't see why it shouldn't build
grand shadow
you can literally make git commits inside Astral and push them for your kernel, ...
I mean yeah, theres one specific commit that I pushed from inside astral to fix a typo.... that added another typo xd
grand shadow
btw idr if you tried building Limine inside Astral too, but i don't see why it s...
I did and it did build after some fiddling around unimplemented mlibc stuff
which was iconv_open
but now thats in libintl and not in mlibc
so even that shoudlnt be an issue
analog berry
locale stuff is probably the most annoying shit ever
i wish everyone spoke the same langauge and used the same formats for everything
marsh sparrow
At least what's required from libc is not that bad
You basically need to be able to tell which unicode chars satisfy isalpha(), isdigit(), islower() etc
and you need to be able to convert to upper case etc
Both of these tasks are solved by large lookup tables that you export from ICU :^)
that's how glibc does it
Actually, I'm not sure if glibc exports from icu or if it maintains its own copies
grand shadow
me thinking my tcp driver shat itself while downloading the binutils tarball
but no it was just ftpmirror.gnu.org giving me a botswana server 💀
it aint even close bruh
wild marsh
huh i didn't know you're in botswana :^)
analog berry
me thinking my tcp driver shat itself while downloading the binutils tarball
How big is your TCP driver btw
grand shadow
How big is your TCP driver btw
analog berry
Damn
Literally every astral component is 1 loc and somehow works better than everyone else
fleet osprey
so true
astral network stack is about to become the world's leading embedded network stack
replacing lwIP
analog berry
Lwip was huge iirc?
fleet osprey
exactly
if Astral's is Just Lighter™, the joke was that it'd 100% replace lwIP
in all seriousness, it probably never would, but maybe it might see some use? would be neat
analog berry
True
grand shadow
I mean it is certainly not *good* but it works well enough
considering it was my first time writing a tcp driver
fleet osprey
I should steal it and use it for https://ptb.discord.com/channels/440442961147199490/1421414518034006066 once I implement literally any networking hardware support
grand shadow
implement astral kexec and use astral as the bootloader 5head
fleet osprey
(it'll probably absolutely shit itself on PowerPC, since you don't target anything big-endian)
grand shadow
big endian cpus arent real wake up sheeple theres a little fairy in your computer that switches the bytes while you arent looking
grand shadow
I mean it is certainly not \*good\* but it works well enough
it is also not super complete
gritty shore
Lwip was huge iirc?
uIP when
grand shadow
I should do a small change on the kernel build stuff and move the regenerate stuff from jinx into a ./prepare.sh and add a kernel-src/BUILD.md detailing how to build just the kernel