OBOS (not vibecoded) | Operating System Development | Page 32

flint idol Dec 31, 2024, 1:02 AM

#

../options/posix/generic/posix_stdlib.cpp:11:10: fatal error: sys/ioctl.h: No such file or directory```

#

from mlibc

#

do I need to define that myself?

weary hound Dec 31, 2024, 1:04 AM

#

qookie@selenium ~/projects/mlibc/src λ find -name ioctl.h
./sysdeps/vinix/include/asm/ioctl.h
./sysdeps/lyre/include/asm/ioctl.h
./options/glibc/include/sys/ioctl.h
``` seems like a bug?

#

can you check git blame for that line meme

flint idol Dec 31, 2024, 1:05 AM

#

e773a516a options/posix/generic/posix_stdlib.cpp  (Geert Custers             2020-08-28 11:36:53 +0200  17) #include <mlibc/posix-sysdeps.hpp>```

#

it's been there for ages

#

who is Geert Custers

#

and astral doesn't seem to have that header either

real pecan Dec 31, 2024, 1:05 AM

#

Think hard

weary hound Dec 31, 2024, 1:05 AM

#

that's not the line

flint idol Dec 31, 2024, 1:05 AM

#

shit I can't read

#

0a20b6aef options/posix/generic/posix_stdlib.cpp (Alexander van der Grinten 2019-07-20 19:12:22 +0200 11) #include <sys/ioctl.h>

#

isn't that portmaster

weary hound Dec 31, 2024, 1:06 AM

#

no

flint idol Dec 31, 2024, 1:06 AM

#

flint idol and astral doesn't seem to have that header either

that's because mathewnd enables glibc, which I assume also imports that header

real pecan Dec 31, 2024, 1:07 AM

#

flint idol isn't that portmaster

Close enough

flint idol Dec 31, 2024, 1:07 AM

#

korona?

lean glen Dec 31, 2024, 1:07 AM

#

no

#

we dont know who that guy is

#

he's a random contributor

#

his name was only heard of in legends

real pecan Dec 31, 2024, 1:07 AM

#

flint idol korona?

Yes

weary hound Dec 31, 2024, 1:07 AM

#

re. geert, i want you to start typing in @ geert without the space (maybe not in here but in a normal channel)

flint idol Dec 31, 2024, 1:07 AM

#

I'm sorry I'm not up to date with all the managarm lore

lean glen Dec 31, 2024, 1:07 AM

#

real pecan Yes

boring

real pecan Dec 31, 2024, 1:08 AM

#

Sorry trl

lean glen Dec 31, 2024, 1:08 AM

#

geert is

#

flint idol Dec 31, 2024, 1:08 AM

#

flint idol korona?

||I found that out through discord search||

weary hound Dec 31, 2024, 1:10 AM

#

anyway you could yoink the lyre or vinix sys/ioctl.h

#

afaics it's the linux one?

#

ah no nvm

#

i misread the paths

#

just enable the glibc option for now meme

flint idol Dec 31, 2024, 1:11 AM

#

ok (I feel like I will regret that)

#

doesn't build

#

white mulch Dec 31, 2024, 1:13 AM

#

yeah that should be fixed, for now you can just add a #define _GNU_SOURCE on top of that file I think

weary hound Dec 31, 2024, 1:13 AM

#

hmm

flint idol Dec 31, 2024, 1:14 AM

#

white mulch yeah that should be fixed, for now you can just add a #define _GNU_SOURCE on top...

well it solves it there, but other errors pop up elsewhere

#

maybe I should just yoink sys/ioctl.h

weary hound Dec 31, 2024, 1:15 AM

#

i think _GNU_SOURCE should be getting defined by the compiler?

flint idol Dec 31, 2024, 1:15 AM

#

weary hound i think _GNU_SOURCE should be getting defined by the compiler?

maybe I don't enable it in my GCC patch

white mulch Dec 31, 2024, 1:15 AM

#

its not because mlibc does cpp_std=c++20 instead of cpp_std=gnu++20

#

I think it would make sense to change that like its already changed to gnu for c

weary hound Dec 31, 2024, 1:16 AM

#

i mean this builds on ci and on managarm

#

and for other oses

flint idol Dec 31, 2024, 1:17 AM

#

flint idol maybe I should just yoink sys/ioctl.h

I did this, to be greeted by

real pecan Dec 31, 2024, 1:17 AM

#

Obos is a bsd confirmed

flint idol Dec 31, 2024, 1:17 AM

#

lol

real pecan Dec 31, 2024, 1:18 AM

#

Obosd

flint idol Dec 31, 2024, 1:18 AM

#

obsd

weary hound Dec 31, 2024, 1:18 AM

#

hm that's missing an ifdef around it

#

checking for __MLIBC_BSD_OPTION

flint idol Dec 31, 2024, 1:18 AM

#

flint idol I did this, to be greeted by

I commented that out

#

and it turns out I can't just yoink that header

#

and I need the glibc option

#

it compiles now

white mulch Dec 31, 2024, 1:23 AM

#

weary hound and for other oses

astral also uses that patch, maybe it depends on some gcc config

weary hound Dec 31, 2024, 1:23 AM

#

possibly

#

we should probably build more combinations of options on ci

#

like ansi+posix without glibc,bsd,etc

vale nymph Dec 31, 2024, 1:26 AM

#

yeah mlibc is kinda iffy when missing a few options

flint idol Dec 31, 2024, 1:27 AM

#

after all that, bash still doesn't build

#

with the same errors

#

about dprintf

#

interesting

#

it doesn't think mlibc has dprintf

#

wth

#

@weary hound I don't believe this should be happening?

weary hound Dec 31, 2024, 1:34 AM

#

did you install the new libc.so?

flint idol Dec 31, 2024, 1:34 AM

#

I will reinstall it

#

just in case

#

that was the problem

vale nymph Dec 31, 2024, 1:39 AM

#

bash soon??

flint idol Dec 31, 2024, 1:39 AM

#

yes

#

bash coming to an obos near you soon™️

#

/home/oberrow/x86_64-obos-bootstrap/bin/../lib/gcc/x86_64-obos/14.2.0/../../../../x86_64-obos/bin/ld: ./lib/termcap/libtermcap.a(termcap.o):/home/oberrow/Code/bash-5.2.37/lib/termcap/termcap.c:298: multiple definition of `PC'; ./lib/readline/libreadline.a(terminal.o):/home/oberrow/Code/bash-5.2.37/lib/readline/terminal.c:109: first defined here
/home/oberrow/x86_64-obos-bootstrap/bin/../lib/gcc/x86_64-obos/14.2.0/../../../../x86_64-obos/bin/ld: ./lib/termcap/libtermcap.a(tparam.o):/home/oberrow/Code/bash-5.2.37/lib/termcap/tparam.c:127: multiple definition of `BC'; ./lib/readline/libreadline.a(terminal.o):/home/oberrow/Code/bash-5.2.37/lib/readline/terminal.c:109: first defined here
/home/oberrow/x86_64-obos-bootstrap/bin/../lib/gcc/x86_64-obos/14.2.0/../../../../x86_64-obos/bin/ld: ./lib/termcap/libtermcap.a(tparam.o):/home/oberrow/Code/bash-5.2.37/lib/termcap/tparam.c:128: multiple definition of `UP'; ./lib/readline/libreadline.a(terminal.o):/home/oberrow/Code/bash-5.2.37/lib/readline/terminal.c:109: first defined here
/home/oberrow/x86_64-obos-bootstrap/bin/../lib/gcc/x86_64-obos/14.2.0/../../../../x86_64-obos/bin/ld: warning: /home/oberrow/x86_64-obos-bootstrap/bin/../lib/gcc/x86_64-obos/14.2.0/crtend.o: missing .note.GNU-stack section implies executable stack

#

nearly there...

#

through some funny buisness I got those to go away

#

[ LIBC  ] Could not satisfy dependency libdl.so
[ LIBC  ] mlibc panicked! exiting program....so

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

weary hound Dec 31, 2024, 1:47 AM

#

there's an option to enable a stub libdl.so iirc

flint idol Dec 31, 2024, 1:47 AM

#

in mlibc?

weary hound Dec 31, 2024, 1:47 AM

#

it doesn't provide anything but exists for programs that want it

#

yeah

flint idol Dec 31, 2024, 1:48 AM

#

it seems to be enabled by default...

#

vale nymph Dec 31, 2024, 1:49 AM

#

if its in /usr/lib in the sysroot then your open or stat syscall is bad, sorry

flint idol Dec 31, 2024, 1:49 AM

#

what stat syscall KEKW

vale nymph Dec 31, 2024, 1:49 AM

#

\s/stat/whatever your stat syscall is

flint idol Dec 31, 2024, 1:50 AM

#

(I have no stat syscall)

#

well fucking lucky me

#

it exists

vale nymph Dec 31, 2024, 1:50 AM

#

get a syscall log and see what happends

flint idol Dec 31, 2024, 1:51 AM

#

ah I see what's happening

#

I forgot to null terminate a string

vale nymph Dec 31, 2024, 1:51 AM

#

xd

flint idol Dec 31, 2024, 1:52 AM

#

thank god it wasn't a bug with open

#

the actual implementation

#

otherwise I would've found the nearest cliff and throw my computer off it

#

"fixed" that

#

(I SEGV now)

#

user mode read on a non-present page

#

0x412f5e50

#

at 0x41241620

#

judging by this, the page is probably swapped out, but I have a bug in that code

#

it could also be a bug in my lazy mapping code

#

[ DEBUG ] Handling page fault at 0x412f5000...
[ DEBUG ] Trying a swap in...
[  LOG  ] User thread 6 SIGSEGV```

#

after enabling debug logs that's the last thing I see

#

which means the page does indeed have a page_range associated with it (i.e., it was mapped through sys_mmap)

#

but the kernel failed to handle the page fault for unknown reasons

#

fixed it

#

I think

#

-    if (rng->cow && (ec & PF_EC_RW))
+    if (rng->cow)
     {
+        if (rng->un.cow_type == COW_SYMMETRIC && (~ec & PF_EC_RW))
+            goto not_cow;```

#

[ LIBC  ] In function ttyname, file ../options/posix/generic/unistd.cpp:829
__ensure(Library function fails due to missing sysdep) failed
[ LIBC  ] In function getuid, file ../options/posix/generic/unistd.cpp:1118
__ensure(Library function fails due to missing sysdep) failed
[ LIBC  ] In function getuid, file ../options/posix/generic/unistd.cpp:1119
__ensure(!"Cannot continue without sys_getuid()") failed
[ LIBC  ] mlibc panicked! exiting program...```

vale nymph Dec 31, 2024, 2:08 AM

#

U can return 0

#

For that

flint idol Dec 31, 2024, 2:08 AM

#

ye

#

what's the difference between ppid and pid

#

found the answer

vale nymph Dec 31, 2024, 2:15 AM

#

Haha pp

flint idol Dec 31, 2024, 2:15 AM

#

hahahhaha

#

look under there

under where?

HAHAHA made you say underwear

#

what's the difference between sys_getpid and sys_getpgid

vale nymph Dec 31, 2024, 2:21 AM

#

flint idol what's the difference between `sys_getpid` and `sys_getpgid`

Pgid is program group id which is job control stuff

flint idol Dec 31, 2024, 2:21 AM

#

so bash needs it?

#

or bash only needs it for job control stuff

vale nymph Dec 31, 2024, 2:21 AM

#

You can just stub it as like 1 or something

#

Only for job control yeah

flint idol Dec 31, 2024, 2:21 AM

#

thanks

vale nymph Dec 31, 2024, 2:22 AM

#

vale nymph You can just stub it as like 1 or something

Or getpid

flint idol Dec 31, 2024, 2:40 AM

#

I am able to get it to run a bit further

#

a bunch of unimplemented signal sysdep messages

#

until it ultimately page faults and the kernel panics

#

in Vfs_FdWrite

#

it page faults

#

Something was here I accidentally edited the message

#

no prompt yet

#

BUT I at least get to the part where it writes to the "TTY"

#

@vale nymph do you know why this could be happening?

#

well first I guess I need sys_getcwd

flint idol Dec 31, 2024, 4:08 AM

#

this is weird

#

even though I have sys_getcwd

#

bash complains about it

#

"ENOSYS"

#

bash source code is brainrotten

#

but I would expect it to print on the libc log that it couldn't find the sysdep

#

and not on the TTY

#

hmm

#

bash's configure says GETCWD_BROKEN=1

#

which is odd

#

x86_64-obos-gcc -shared  -L./lib/termcap  -o print print.o 
/home/oberrow/x86_64-obos-bootstrap/bin/../lib/gcc/x86_64-obos/14.2.0/../../../../x86_64-obos/bin/ld: /home/oberrow/x86_64-obos-bootstrap/bin/../lib/gcc/x86_64-obos/14.2.0/crtbegin.o: relocation R_X86_64_32 against hidden symbol `__TMC_END__' can not be used when making a shared object```

#

I decided to make install

#

bash

#

only to get this

#

well it's getting late now

#

so I'm off to sleep

empty kernel Dec 31, 2024, 4:29 AM

#

you still have 1 more day left in the year

flint idol Dec 31, 2024, 4:29 AM

#

yes

#

(1 day EST)

#

so 24 hours 30 minutes

empty kernel Dec 31, 2024, 4:30 AM

#

yeah, I've only got 9 and a half hours left in the year

flint idol Dec 31, 2024, 4:30 AM

#

thick jolt Dec 31, 2024, 4:30 AM

#

while ur having fun omar, im experiencing scheduler ub

#

nooo

#

bro is doing the fun part

flint idol Dec 31, 2024, 4:30 AM

#

thick jolt while ur having fun omar, im experiencing scheduler ub

this ain't fun

thick jolt Dec 31, 2024, 4:30 AM

#

of porting

empty kernel Dec 31, 2024, 4:30 AM

#

thick jolt while ur having fun omar, im experiencing scheduler ub

I thought you had gone to bed

thick jolt Dec 31, 2024, 4:30 AM

#

empty kernel I thought you had gone to bed

nahhh

#

cant sleep

thick jolt Dec 31, 2024, 4:30 AM

#

flint idol this ain't fun

bro its lit porting shit

#

its so fun

flint idol Dec 31, 2024, 4:31 AM

#

flint idol this ain't fun

I'm debugging sys_getcwd not existing

#

when it exists

#

https://tenor.com/view/funny-hahaha-gif-22505124

Tenor

thick jolt Dec 31, 2024, 4:31 AM

#

how about u deal with scheduler UB trl

flint idol Dec 31, 2024, 4:31 AM

#

I have, don't worry

thick jolt Dec 31, 2024, 4:31 AM

#

even worse

#

like bro i just added thread states like running, ready, and zombie

#

crazy

flint idol Dec 31, 2024, 4:31 AM

#

a day or two I had to debug the scheduler

thick jolt Dec 31, 2024, 4:32 AM

#

what was ze issue for u

flint idol Dec 31, 2024, 4:32 AM

#

specifically the context switching

flint idol Dec 31, 2024, 4:32 AM

#

thick jolt what was ze issue for u

idr, but it wasn't fun

empty kernel Dec 31, 2024, 4:32 AM

#

flint idol specifically the context switching

that is always annoying

flint idol Dec 31, 2024, 4:32 AM

#

luckily for me it only happened in the first few context switches of the user program

#

and iirc it had something to do with stack switching

flint idol Dec 31, 2024, 4:33 AM

#

flint idol ``` x86_64-obos-gcc -shared -L./lib/termcap -o print print.o /home/oberrow/x8...

I think that's because I'm still using the "bootstrap gcc"

#

aka gcc built with inhibitLibc=true

#

anyway gn

thick jolt Dec 31, 2024, 4:46 AM

#

nightnight

lean glen Dec 31, 2024, 5:05 AM

#

flint idol I'm debugging `sys_getcwd` not existing

it's not needed

#

I don't have it

#

And bash works fine

flint idol Dec 31, 2024, 5:06 AM

#

Hm

flint idol Dec 31, 2024, 5:07 AM

#

flint idol Something was here I accidentally edited the message

Then why does it complain here

lean glen Dec 31, 2024, 5:08 AM

#

Because you fucked up

flint idol Dec 31, 2024, 5:08 AM

#

Makes perfect sense

flint idol Dec 31, 2024, 12:54 PM

#

I just love my isatty implementation

limber wave Dec 31, 2024, 1:26 PM

#

flint idol I just love my `isatty` implementation

🤫

flint idol Dec 31, 2024, 1:29 PM

#

@vale nymph help

#

pls

#

I decided to remove my "isatty" implementation

#

and stub it out

#

but now bash just exits without doing any prompt

vale nymph Dec 31, 2024, 1:40 PM

#

y stub it out

flint idol Dec 31, 2024, 1:40 PM

#

flint idol I just love my `isatty` implementation

well this is how it's implemented

#

so I figured I should just stub it instead

vale nymph Dec 31, 2024, 1:44 PM

#

implement it properly lol

white mulch Dec 31, 2024, 2:01 PM

#

flint idol well this is how it's implemented

you need pselect at the very least stubbed out to return success and 1 as the result (I just figured that out having that same issue on my kernel)

#

see lib/readline/input.c:833

flint idol Dec 31, 2024, 2:02 PM

#

I'll try that

white mulch Dec 31, 2024, 2:03 PM

#

ofc making it return 1 as the result and always succeeding isn't likely going to make it work very well but it shouldn't exit at least

flint idol Dec 31, 2024, 2:09 PM

#

I did that, still no prompt :(

white mulch Dec 31, 2024, 2:09 PM

#

did you also restore your original isatty stub

flint idol Dec 31, 2024, 2:09 PM

#

no

white mulch Dec 31, 2024, 2:10 PM

#

try that

flint idol Dec 31, 2024, 2:11 PM

#

still no prompt

#

I get a bunch of missing sigaction sysdeps

#

maybe I should fix that

vale nymph Dec 31, 2024, 2:15 PM

#

not really needed for bash

#

@flint idol get a syscall log

#

something somewehre is failing

#

did you compile it for readline?

flint idol Dec 31, 2024, 2:16 PM

#

I just compiled it the same way you compile it

#

./configure --host=x86_64-obos --prefix=$HOME/Code/obos/tar/ --without-bash-malloc```

#

then make all

#

and then I copy bash to my initrd

#

and run that

flint idol Dec 31, 2024, 2:17 PM

#

vale nymph <@841762704766140486> get a syscall log

sure

#

this will take a bit, as I need to implement syscall logging

vale nymph Dec 31, 2024, 2:18 PM

#

hmmm probably it is using readline

#

for readline you need poll

#

if poll fails it exits

#

does mlibc complain about an unimplemented poll?

flint idol Dec 31, 2024, 2:18 PM

#

I was able to get it to stop exiting

flint idol Dec 31, 2024, 2:18 PM

#

vale nymph does mlibc complain about an unimplemented poll?

I'll check

#

no

#

📎 message.txt

white mulch Dec 31, 2024, 2:18 PM

#

I didn't get any polls either

flint idol Dec 31, 2024, 2:19 PM

#

that's all the unimplemented sysdep things I got

vale nymph Dec 31, 2024, 2:21 PM

#

something is erroring out which bash doesnt like lol

flint idol Dec 31, 2024, 2:22 PM

#

btw now it hangs after printing some errors on stdout

#

📎 message.txt

#

there's my syscall log

#

except it has no syscall names

#

most the time, if a syscall returns zero that means success

#

but syscall 22 is mmap

#

which returns a void*

#

and 42 is TellOff

#

syscall 57 is libc log

#

which is void

#

34 is FdAlloc, which allocates a file handle

#

and 35 is FdOpen

#

which fails with 0xb

#

aka OBOS_STATUS_NOT_FOUND

#

it tried to open /etc/passwd

#

I wonder if that's a problem

#

so other than that failed open, all else is fine

#

until it ultimately exits the thread

#

causing the "hang"

#

[  LOG  ] writing shell-init: error retrieving current directory: getcwd: cannot access parent directories: Operation not implemented (ENOSYS)
 to stderr```

#

I also added logs of writes to stderr/stdout

#

[  LOG  ] writing init: initialize_job_control: getpgrp failed: Operation not implemented (ENOSYS)
 to stderr```

white mulch Dec 31, 2024, 2:40 PM

#

flint idol ``` [ LOG ] writing shell-init: error retrieving current directory: getcwd: ca...

ah yeah I forgot about that one, it assumes that getcwd is broken by default when cross compiling and uses its own impl

real pecan Dec 31, 2024, 2:40 PM

#

Lol

flint idol Dec 31, 2024, 2:40 PM

#

how do I get it to.. not?

#

I had noticed that in config.log, and assumed whatever

white mulch Dec 31, 2024, 2:40 PM

#

bash_cv_getcwd_malloc=yes to the end of the ./configure line

flint idol Dec 31, 2024, 2:40 PM

#

the ./configure line?

#

or do you mean file

white mulch Dec 31, 2024, 2:41 PM

#

./configure --host=whatever --without-bash-malloc bash_cv_getcwd_malloc=yes

flint idol Dec 31, 2024, 2:41 PM

#

ah

weary hound Dec 31, 2024, 2:41 PM

#

you can also export that on the env instead

#

when invoking configure

flint idol Dec 31, 2024, 2:42 PM

#

btw when I run make install I get linker errors

flint idol Dec 31, 2024, 2:42 PM

#

flint idol ``` x86_64-obos-gcc -shared -L./lib/termcap -o print print.o /home/oberrow/x8...

here

#

which I found weird

white mulch Dec 31, 2024, 2:44 PM

#

pretty sure it should use crtbeginS.o for a shared library thonk

flint idol Dec 31, 2024, 2:44 PM

#

relocation R_X86_64_32 against hidden symbol '__TMC_END__' can not be used when making a shared object

#

well adding that removes the getcwd error

#

but still no prompt thonk

#

[  LOG  ] syscall 37(0x2, 0xcf8008, 0x51, 0x916060, 0x0)
[  LOG  ] writing init: initialize_job_control: getpgrp failed: Operation not implemented (ENOSYS)
 to stderr
[  LOG  ] syscall 37 returned 0x0000000000000000
[  LOG  ] syscall 42(0x0, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 41(0x0, 0x0, 0x1, 0x0, 0x0)
[  LOG  ] syscall 41 returned 0x0000000000000000
[  LOG  ] syscall 42(0x0, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 42(0x0, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 41(0x0, 0x0, 0x1, 0x0, 0x0)
[  LOG  ] syscall 41 returned 0x0000000000000000
[  LOG  ] syscall 42(0x0, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 42(0x1, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 41(0x1, 0x0, 0x1, 0x0, 0x0)
[  LOG  ] syscall 41 returned 0x0000000000000000
[  LOG  ] syscall 42(0x1, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 42(0x1, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 41(0x1, 0x0, 0x1, 0x0, 0x0)
[  LOG  ] syscall 41 returned 0x0000000000000000
[  LOG  ] syscall 42(0x1, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 42(0x2, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 41(0x2, 0x0, 0x1, 0x0, 0x0)
[  LOG  ] syscall 41 returned 0x0000000000000000
[  LOG  ] syscall 42(0x2, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] syscall 42 returned 0x0000000000000000
[  LOG  ] syscall 0(0x0, 0x0, 0x0, 0x0, 0x0)

#

syscall 0 is exit current thread

#

syscall 42 is fd tell off

#

syscall 41 is fd seek

#

syscall 37 is fd write

#

and none of these syscalls fail...

white mulch Dec 31, 2024, 2:50 PM

#

try stubbing getpgid

flint idol Dec 31, 2024, 2:50 PM

#

ok

#

I'm also going to make sys_thread_exit hang

flint idol Dec 31, 2024, 2:54 PM

#

white mulch try stubbing getpgid

now I get SIGILL

#

KEKW

white mulch Dec 31, 2024, 2:54 PM

#

lol

flint idol Dec 31, 2024, 2:55 PM

#

but it gets further than last time

#

seemingly

white mulch Dec 31, 2024, 2:55 PM

#

does it say anything about it

#

or does it just die

flint idol Dec 31, 2024, 2:55 PM

#

last thing it says is [ LIBC ] mlibc: sys_getegid is a stub!

#

then SIGILL

#

I stubbed that and made it return 0

#

I removed the fake stub

#

and now mlibc panics

#

[ LIBC  ] In function getegid, file ../options/posix/generic/unistd.cpp:1110
__ensure(Library function fails due to missing sysdep) failed
[  LOG  ] syscall 57 returned 0x00000000c404a2cf
[  LOG  ] syscall 57(0x916b98, 0x0, 0x0, 0x0, 0x0)
[ LIBC  ] In function getegid, file ../options/posix/generic/unistd.cpp:1111
__ensure(!"Cannot continue without sys_getegid()") failed
[  LOG  ] syscall 57 returned 0x00000000c404a2cf
[  LOG  ] syscall 57(0x412aa0b8, 0x0, 0x0, 0x0, 0x0)
[ LIBC  ] mlibc panicked! exiting program...
[  LOG  ] syscall 57 returned 0x00000000c404a2cf
[  LOG  ] syscall 0(0x0, 0x0, 0x0, 0x0, 0x0)

#

so I put the stub back

#

and ofc it SIGILLed

#

like it did before

#

I found the problem

#

that caused SIGILL

#

I was forgetting to return a value in a stubbed sysdep

#

#

I get a prompt, to an extent

#

it just throws in exit

white mulch Dec 31, 2024, 3:08 PM

#

did you make pselect return 1 in num_events?

flint idol Dec 31, 2024, 3:08 PM

#

yes

#

wait no

white mulch Dec 31, 2024, 3:08 PM

#

then it also tries to read stdio after that

#

so if you don't return anything (or return zero read bytes) it might fail

flint idol Dec 31, 2024, 3:10 PM

#

wait I do

#

it was an ancient bug in my uart driver

#

where it doesn't block in a function that should block

#

at least, iirc I didn't fix it

flint idol Dec 31, 2024, 3:12 PM

#

white mulch so if you don't return anything (or return zero read bytes) it might fail

this is in sys_read right?

#

or is it in pselect

white mulch Dec 31, 2024, 3:12 PM

#

yes

flint idol Dec 31, 2024, 3:12 PM

#

int sys_pselect(int num_fds, fd_set *read_set, fd_set *write_set, fd_set *except_set, const struct timespec *timeout, const sigset_t *sigmask, int *num_events)
{
    *num_events = 1;
    return 1;
}

white mulch Dec 31, 2024, 3:13 PM

#

you should make it return 0, 1 indicates a failure

flint idol Dec 31, 2024, 3:13 PM

#

oops

white mulch Dec 31, 2024, 3:13 PM

#

it uses pselect to wait for input and then after that reads from stdio

flint idol Dec 31, 2024, 3:16 PM

#

why is bash seemingly trying to read zero bytes

#

nvm

#

it's trying to read one, but I have a bug somewhere in fd read

#

fuck clang

#

while(port->in_buffer.szBuf < blkCount)
    ;```

#

it optimized this loop into an infinite loop

#

#

FINALLY

#

bash

#

except I don't have fork so nothing happens KEKW

#

but hey the exit command works KEKW

white mulch Dec 31, 2024, 3:23 PM

#

better than on my kernel, I haven't yet added proper stdio read

flint idol Dec 31, 2024, 3:24 PM

#

just get a serial driver

#

my stdin opens /dev/COM1

#

void OBOS_OpenStandardFDs()
{
    handle hnd_stdin = Sys_FdAlloc();
    handle hnd_stdout = Sys_FdAlloc();
    handle hnd_stderr = Sys_FdAlloc();
    OBOS_LockHandleTable(OBOS_CurrentHandleTable());
    obos_status status = OBOS_STATUS_SUCCESS;
    handle_desc* stdin = OBOS_HandleLookup(OBOS_CurrentHandleTable(), hnd_stdin, HANDLE_TYPE_FD, false, &status);
    handle_desc* stdout = OBOS_HandleLookup(OBOS_CurrentHandleTable(), hnd_stdout, HANDLE_TYPE_FD, false, &status);
    handle_desc* stderr = OBOS_HandleLookup(OBOS_CurrentHandleTable(), hnd_stderr, HANDLE_TYPE_FD, false, &status);
    OBOS_UnlockHandleTable(OBOS_CurrentHandleTable());
    Vfs_FdOpen(stdin->un.fd, "/dev/COM1", FD_OFLAGS_READ);
    Vfs_FdOpen(stdout->un.fd, "/dev/COM1", FD_OFLAGS_WRITE);
    Vfs_FdOpen(stderr->un.fd, "/dev/COM1", FD_OFLAGS_WRITE);
}```

white mulch Dec 31, 2024, 3:25 PM

#

yeah that would be the easiest option

flint idol Dec 31, 2024, 3:26 PM

#

time to implement fork

#

@thick jolt guess who has bash now!!11

lean glen Dec 31, 2024, 3:26 PM

#

white mulch see lib/readline/input.c:833

you can also build bash without readline

white mulch Dec 31, 2024, 3:26 PM

#

ah lol

flint idol Dec 31, 2024, 3:27 PM

#

wait what

#

wtf

#

removing syscall logs changed

#

the behaviour

#

of bash

lean glen Dec 31, 2024, 3:27 PM

#

pro tip: ./configure --help

#

grep

flint idol Dec 31, 2024, 3:29 PM

#

what the hell is happening lol

#

the first command ran is the only one that runs

#

so if I did echo hi as the first comamnd

#

echo hi is the command that works

#

through out the entire session

#

lean glen Dec 31, 2024, 3:30 PM

#

why are you using putty

flint idol Dec 31, 2024, 3:30 PM

#

cuz

#

I have qemu set to put serial on localhost:1534

#

so I'm using putty to connect to that

lean glen Dec 31, 2024, 3:31 PM

#

just

#

-serial stdio

#

nooo

flint idol Dec 31, 2024, 3:31 PM

#

stdio is taken

#

by the monitor

lean glen Dec 31, 2024, 3:31 PM

#

you can access the monitor via the gui anyway

flint idol Dec 31, 2024, 3:32 PM

#

fair

flint idol Dec 31, 2024, 3:32 PM

#

flint idol `echo hi` is the command that works

nvm, bash sometimes wants to run a comamnd

#

*command

#

sometimes it decides not to

#

I assume comamnds like ls are in coreutils, right?

#

as well as cat

lean glen Dec 31, 2024, 3:34 PM

#

yes

flint idol Dec 31, 2024, 3:34 PM

#

k

#

so I'll be porting that next

#

after implementing fork

flint idol Dec 31, 2024, 3:56 PM

#

I have a bug which idk how to solve

#

so I segfault after a GPF

#

when I pass a 2nd argument to the program

#

i.e., argc=2

#

I look into it, and it faults on an instruction where the pointer passed should be aligned to a 16-byte boundary

#

movaps   %xmm0, -0x20(%rbp)

#

but rbp=0x00000000009129d8

#

and I can't align the stack when I hand off the program to the dynamic loader, because then it can't see the aux values

#

and panics

weary hound Dec 31, 2024, 4:00 PM

#

? the aux values don't change alignment

#

the only thing to account for is argc + argv[] + null + envp[] + null

#

and you can account for that before pushing the auxv by pushing an extra 0 if needed

flint idol Dec 31, 2024, 4:01 PM

#

oh yeah

#

thanks

flint idol Dec 31, 2024, 4:08 PM

#

flint idol

using bash --verbose, I discover that it's because sometimes there are missing bytes

#

in stuff read from the serial driver

#

as can be seen here

#

anyway, now that I have bash, I want to continue making the thing

#

obos-strap

flint idol Dec 31, 2024, 6:41 PM

#

I implement the signal-related sysdeps

#

so I decided to try it out using kill -s SIGINT 6

#

(6 is the main thread of the bash process)

#

but then disaster strikes KEKW

#

and of course in a debugger, this doesn't reprod

thick jolt Dec 31, 2024, 7:26 PM

#

flint idol <@787341689333350450> guess who has bash now!!11

dayum

#

meanwhile im over here struggling with scheduler UB STILL

#

@flint idol good progress though happy u reached bash

#

!!

timid elk Dec 31, 2024, 7:47 PM

#

flint idol `obos-strap`

what significant feature do you want it to have

flint idol Dec 31, 2024, 7:47 PM

#

nothing

timid elk Dec 31, 2024, 7:47 PM

#

asking so i can implement it in nixstrap

flint idol Dec 31, 2024, 7:48 PM

#

I was in bed yesterday and thought that would be fun to try out

timid elk Dec 31, 2024, 7:48 PM

#

i want nixstrap to be able to generate runners

timid elk Dec 31, 2024, 7:48 PM

#

timid elk i want nixstrap to be able to generate runners

and i want a helper application for nixstrap

#

so i can nixstrap configure or something to set qemu args

#

then it will generate a shell script for that

#

and maybe i could put that in home manager

#

and i can have a script like gbox run

#

generated automagically

thick jolt Dec 31, 2024, 7:50 PM

#

so work on nixstrap lol

#

im still waiting

timid elk Dec 31, 2024, 7:50 PM

#

no

#

i will work on it in the summer

thick jolt Dec 31, 2024, 7:50 PM

#

pulls out glock 19

#

im gonna reach userspace before the summer

#

nooo

timid elk Dec 31, 2024, 7:50 PM

#

tomorrow i might work a couple hours on mlibc in nixpkgs

flint idol Dec 31, 2024, 7:50 PM

#

thick jolt im gonna reach userspace before the summer

use obos-strap trl

#

when it comes out

thick jolt Dec 31, 2024, 7:51 PM

#

lmao

timid elk Dec 31, 2024, 7:51 PM

#

thick jolt im gonna reach userspace before the summer

do your own thing

#

also i want ISOs in nixstrap

flint idol Dec 31, 2024, 7:51 PM

#

that's probably gonna somehow end up deleting your entire root

timid elk Dec 31, 2024, 7:51 PM

#

timid elk also i want ISOs in nixstrap

it can only generate raw disk images rn

flint idol Dec 31, 2024, 7:51 PM

#

like it is somehow gonna do rm -rf --no-preserve-root /

thick jolt Dec 31, 2024, 7:51 PM

#

nyauxstrap rm -rf --no-preserve-root /

flint idol Dec 31, 2024, 7:51 PM

#

if @thick jolt tried to make a package manager

thick jolt Dec 31, 2024, 7:51 PM

#

trl

flint idol Dec 31, 2024, 7:51 PM

#

or whatever these are called

lean glen Dec 31, 2024, 7:52 PM

#

flint idol use obos-strap <:trl:829316641191886888>

just use jinx

timid elk Dec 31, 2024, 7:52 PM

#

timid elk it can only generate raw disk images rn

disko supports various VM image formats so that's a very simple addition to nixstrap but ISOs are a problem

timid elk Dec 31, 2024, 7:52 PM

#

lean glen just use jinx

yea jinx is cool

#

nixstrap is cooler ofc but it's not done

lean glen Dec 31, 2024, 7:52 PM

#

how is nixstrap cooler

#

im genuinely asking

timid elk Dec 31, 2024, 7:52 PM

#

#1279525023127830609 and scroll up

#

it's cool because

you can use the entirety of nixpkgs (which is super useful for bootloaders and stuff, and if you add your OS as a supported platform in nixpkgs you can probably get userspace apps from nixpkgs as well, and if you're ABI compatible with linux you can get unpatched linux software from there
declarative and modular (it's like using nixos modules, and several modules are provided by nixstrap by default like installing limine or generating limine.conf or stuff for image generation
you get a lockfile and nix has reproducibility guarantees

flint idol Dec 31, 2024, 7:56 PM

#

oh yeah for obos-strap I had the brilliant idea of using... named semaphores for locking

timid elk Dec 31, 2024, 7:57 PM

#

timid elk it's cool because - you can use the entirety of nixpkgs (which is super useful f...

and you can spin up a nix cache server like how NixBSD has one to cut down on compile times for your contributors

sharp pike Dec 31, 2024, 7:57 PM

#

lean glen how is nixstrap cooler

i don't think jinx and nixstrap are the same thing and the same purpose tbh lol

lean glen Dec 31, 2024, 7:58 PM

#

neither do i

sharp pike Dec 31, 2024, 7:58 PM

#

just like xbstrap is different and not comparable to jinx

lean glen Dec 31, 2024, 7:58 PM

#

jinx is literally called so because of xbstrap tho

timid elk Dec 31, 2024, 7:58 PM

#

i mean using one would replace the other right?

lean glen Dec 31, 2024, 7:58 PM

#

jinx and xbstrap are comparable

sharp pike Dec 31, 2024, 7:58 PM

#

yes but it literally means "jinx is not xbstrap" :^)

timid elk Dec 31, 2024, 7:58 PM

#

timid elk i mean using one would replace the other right?

however different they may be they do the same core thing

#

which is to bootstrap an OS or distro

sharp pike Dec 31, 2024, 7:59 PM

#

yeah sure

timid elk Dec 31, 2024, 7:59 PM

#

jinx vs nixstrap is basically ABS (arch) vs nix

sharp pike Dec 31, 2024, 7:59 PM

#

in that sense they are comparable, fair

timid elk Dec 31, 2024, 8:00 PM

#

anyhow the only time i tried building managarm

#

i got frustrated over xbstrap

#

i don't remember why tho

#

jinx worked every time i used it for other OSes here but i never used it for a project of mine (none of them progressed to that point

sharp pike Dec 31, 2024, 8:01 PM

#

jinx W

timid elk Dec 31, 2024, 8:01 PM

#

yep thanks for your work on jinx

#

i may be a nixstrap evangelist but i appreciate jinx

#

jinx is what inspired nixstrap's predecessor which i probably didn't mention here at all

#

that predecessor was lever and it was a build system / os bootstrapper written in lua & rust with recipes in lua and it used udisksctl for mounting and image generation

#

then i tried making it at least somewhat reproducable

#

and i got really mad

#

then scrapped it and did nixstrap

#

i say "lua" but it was more a dialect of lua where unresolved functions spawned the program in path with the name of that function and it had ""pipelines"" so you could do cat("test.txt"):tee("foo.txt")

#

anyhow happy new year

sharp pike Dec 31, 2024, 8:25 PM

#

timid elk yep thanks for your work on jinx

❤️

sharp pike Dec 31, 2024, 8:25 PM

#

timid elk anyhow happy new year

likewise

flint idol Dec 31, 2024, 9:02 PM

#

I think it's time to merge https://github.com/OBOS-dev/obos/pull/27

GitHub

Userspace work by oberrow · Pull Request #27 · OBOS-dev/obos

Countless changes to the kernel, adding userspace, rewriting the VMM, PCI interface, power management stuff, and many other miscellaneous changes.

#

small changes

empty kernel Dec 31, 2024, 9:04 PM

#

16k insertions, wow you've done a lot in the week that branch has existed

flint idol Dec 31, 2024, 9:04 PM

#

this branch has existed for 5 months

#

*4 months

#

it's just that the PR was made 1 week ago

empty kernel Dec 31, 2024, 9:04 PM

#

Oh

#

Just noticed that

#

Still lots of progress

flint idol Dec 31, 2024, 9:05 PM

#

indeed

timid elk Dec 31, 2024, 9:39 PM

#

hello from 2025

thick jolt Dec 31, 2024, 9:39 PM

#

it is not 2025

timid elk Dec 31, 2024, 9:39 PM

#

it is

#

GMT+3

vale nymph Dec 31, 2024, 9:41 PM

#

timid elk hello from 2025

Is 2025 finally the year of the linux desktop

flint idol Dec 31, 2024, 9:41 PM

#

*obos desktop

#

trl

weak kestrel Dec 31, 2024, 9:41 PM

#

vale nymph Is 2025 finally the year of the linux desktop

90% desktops running Linux this year????

timid elk Dec 31, 2024, 9:41 PM

#

vale nymph Is 2025 finally the year of the linux desktop

i mean i have yet to use windows

#

so for me

#

all desktops (and laptops) have been running linux

weak kestrel Dec 31, 2024, 9:41 PM

#

timid elk i mean i have yet to use windows

I do not recommend it

timid elk Dec 31, 2024, 9:42 PM

#

i wonder how long i can go without windows

#

probably 2 days

#

because school

flint idol Dec 31, 2024, 9:43 PM

#

I've gone 7 months

#

(best 7 months ever)

timid elk Dec 31, 2024, 9:43 PM

#

we are in the first month oberrow

#

there's no way you went 7 months

weak kestrel Dec 31, 2024, 9:44 PM

#

he's a time traveler

timid elk Dec 31, 2024, 9:45 PM

#

i have knowledge from the future

#

ask me what you want to know

#

i should probably be in another channel tho

weak kestrel Dec 31, 2024, 9:45 PM

#

timid elk ask me what you want to know

will I get into the uni I want

flint idol Dec 31, 2024, 9:46 PM

#

timid elk ask me what you want to know

What will happen to Syria

timid elk Dec 31, 2024, 9:46 PM

#

flint idol What will happen to Syria

iraq 2.0 let's hope someone doesn't accidentally kill 1.5 million

flint idol Dec 31, 2024, 9:46 PM

#

Damn it

#

Was hoping to visit it soon

timid elk Dec 31, 2024, 9:47 PM

#

weak kestrel will I get into the uni I want

i need details on who you are

#

give me SSN or ID number or whatever

#

exact location

#

birth date and time

#

full legal name

flint idol Dec 31, 2024, 9:47 PM

#

1234 5678 is ID number

timid elk Dec 31, 2024, 9:47 PM

#

the numbers on your card

flint idol Dec 31, 2024, 9:47 PM

#

123 sesame street

timid elk Dec 31, 2024, 9:47 PM

#

flint idol 1234 5678 is ID number

invalid

timid elk Dec 31, 2024, 9:48 PM

#

flint idol 123 sesame street

9/11

#

will happen

flint idol Dec 31, 2024, 9:48 PM

#

timid elk birth date and time

9/11/2001 at aroud 7am

flint idol Dec 31, 2024, 9:48 PM

#

timid elk full legal name

Ben Dover

timid elk Dec 31, 2024, 9:48 PM

#

flint idol 9/11/2001 at aroud 7am

lol that's one day after my sister

#

tho

weak kestrel Dec 31, 2024, 9:48 PM

#

Georgia, Tbilisi
Birth date: April 20 1889
Full name: albus percival wulfric brian dumbledore
ID: 6942069420

timid elk Dec 31, 2024, 9:48 PM

#

americans use MM/DD

timid elk Dec 31, 2024, 9:49 PM

#

weak kestrel Georgia, Tbilisi Birth date: April 20 1889 Full name: albus percival wulfric bri...

you will move abroad

#

to baguette land

#

or terf island

weak kestrel Dec 31, 2024, 9:49 PM

#

timid elk to baguette land

I would rather die

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

weak kestrel Dec 31, 2024, 9:50 PM

#

timid elk or terf island

which one is that

timid elk Dec 31, 2024, 9:50 PM

#

the one on the right

#

the one on the left isn't terf island

#

that's ireland

weak kestrel Dec 31, 2024, 9:51 PM

#

ah

flint idol Dec 31, 2024, 10:45 PM

#

https://github.com/OBOS-dev/obos/pull/27

GitHub

Userspace work by oberrow · Pull Request #27 · OBOS-dev/obos

Countless changes to the kernel, adding userspace, rewriting the VMM, PCI interface, power management stuff, and many other miscellaneous changes.

#

merged

#

@thick jolt you are now in the contributors list

#

now time to see if it even compiles

thick jolt Dec 31, 2024, 10:46 PM

#

yay

flint idol Dec 31, 2024, 10:50 PM

#

after some compiler errors were fixerd

#

it compiles

#

but it PFs

#

but with the pretty panic screen

#

I made a few months ago

flint idol Jan 1, 2025, 12:02 AM

#

damn it

#

it works on debug mode

#

but not release

#

meaning I've UB

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

#

I guess I will be git diffing

#

to find the problem

empty kernel Jan 1, 2025, 12:06 AM

#

flint idol but with the pretty panic screen

the commit ID starts with bad...

flint idol Jan 1, 2025, 12:06 AM

#

it's git's fault then meme

#

I need to git commit --amend to remake the commit's hash

#

then the problem is solved

#

fixed that seemingly

#

oh wait it's UB

#

that's not how it works

real pecan Jan 1, 2025, 12:15 AM

#

Is 2025 the year of bash on obos

flint idol Jan 1, 2025, 12:15 AM

#

yes

#

well 2024 is

#

#1141057599584878645 message

real pecan Jan 1, 2025, 12:15 AM

#

What are 2025 goals

flint idol Jan 1, 2025, 12:15 AM

#

idk

#

but I need to get the crashes on master fixed

#

I'm going to use git bisect

real pecan Jan 1, 2025, 12:21 AM

#

flint idol idk

Self hosting?

flint idol Jan 1, 2025, 12:21 AM

#

seems far out of my reach

#

currently

real pecan Jan 1, 2025, 12:21 AM

#

Well if your bash already somewhat works

#

All you need is more syscalls

flint idol Jan 1, 2025, 12:22 AM

#

it's not the lack of syscalls really, it's the lack of sysdeps

real pecan Jan 1, 2025, 12:22 AM

#

Even easier

flint idol Jan 1, 2025, 12:23 AM

#

the only syscalls I can even think of needing rn are:

TTY syscalls
tgkill
sockets

#

I also need dup(2)

#

symlink syscalls

#

I need file creation/removal

#

I need poll

white mulch Jan 1, 2025, 12:24 AM

#

at least you have signals

flint idol Jan 1, 2025, 12:24 AM

#

although I already have pipes, I need to add syscalls for them

flint idol Jan 1, 2025, 12:25 AM

#

white mulch at least you have signals

true

#

except none, but the "exception" signals are automatically sent by the kernel as of now

#

and by exception signals I mean SIGSEGV, SIGILL, SIGFPE, SIGTRAP

flint idol Jan 1, 2025, 12:27 AM

#

flint idol I need `poll`

timer related syscalls as well

#

int sys_waitid(idtype_t idtype, id_t id, siginfo_t *info, int options);```

#

probably this too

#

also process group related syscalls probably

white mulch Jan 1, 2025, 12:28 AM

#

tldr: everything lol

flint idol Jan 1, 2025, 12:28 AM

#

out of the 60 syscalls I have, almost none I have are used

#

by mlibc

#

so that's something

white mulch Jan 1, 2025, 12:29 AM

#

there are actually a huge ton of functions in posix, I hadn't realized that before I wrote my own libc

flint idol Jan 1, 2025, 12:29 AM

#

I also need the dreaded fork function

#

although I have half of the work for that done

real pecan Jan 1, 2025, 12:29 AM

#

white mulch there are actually a huge ton of functions in posix, I hadn't realized that befo...

Crescent is like the good serenityos

white mulch Jan 1, 2025, 12:30 AM

#

how lol

empty kernel Jan 1, 2025, 12:30 AM

#

flint idol I also need the dreaded `fork` function

do you have execve?

flint idol Jan 1, 2025, 12:30 AM

#

yes

#

syscall 56

empty kernel Jan 1, 2025, 12:31 AM

#

so you can execute something, but you can't still have a process after starting a new one

flint idol Jan 1, 2025, 12:31 AM

#

what execve does is make the current pid have an identity crisis

real pecan Jan 1, 2025, 12:31 AM

#

white mulch how lol

More competent

flint idol Jan 1, 2025, 12:31 AM

#

flint idol what `execve` does is make the current pid have an identity crisis

which thus causes it to change itself into a new process

#

fork starts a child process that is effectively a clone of the current one

empty kernel Jan 1, 2025, 12:36 AM

#

I find it weird that you have execve, but not fork. You can start a new program, but the current one has to die in the process.

flint idol Jan 1, 2025, 12:36 AM

#

after fixing these bugs, I will:

implement the execve sysdep
implement fork
implement fork sysdep

flint idol Jan 1, 2025, 12:36 AM

#

empty kernel I find it weird that you have execve, but not fork. You can start a new program,...

well many moons ago I decided to do it, but then I stopped after having finished the VMM part

#

so I have mm/fork.h defining Mm_ForkContext

#

/*
 * oboskrnl/mm/fork.h
 *
 * Copyright (c) 2024 Omar Berrow
 */

#pragma once

#include <int.h>
#include <error.h>

#include <mm/context.h>

obos_status Mm_ForkContext(context* into, context* toFork);

#

but no Sys_Fork

empty kernel Jan 1, 2025, 12:37 AM

#

fork is a pain, so I can understand why you would abandon it

flint idol Jan 1, 2025, 12:37 AM

#

yup

white mulch Jan 1, 2025, 12:37 AM

#

real pecan More competent

tbh I don't even know much about serenity, I haven't really looked much into it (well I do know what it is and whatever but more than that) KEKW

real pecan Jan 1, 2025, 12:38 AM

#

white mulch tbh I don't even know much about serenity, I haven't really looked much into it ...

Well crescent is like serenity because full nih

#

Is what I meant

flint idol Jan 1, 2025, 12:38 AM

#

flint idol yup

although the hard part which, to most people appears to be the VMM part, is the only part I did

white mulch Jan 1, 2025, 12:39 AM

#

real pecan Well crescent is like serenity because full nih

ah

real pecan Jan 1, 2025, 12:39 AM

#

And c++

white mulch Jan 1, 2025, 12:40 AM

#

flint idol although the hard part which, to most people appears to be the VMM part, is the ...

I should also implement that but I still don't even have demand mapping KEKW

#

and I hate fork with passion

flint idol Jan 1, 2025, 12:41 AM

#

white mulch and I hate fork with passion

same

#

yet it's the only way with mlibc (unless mlibc has a clone sysdep)

#

afaict

white mulch Jan 1, 2025, 12:42 AM

#

or you just patch the usages of fork within everything you want to port

flint idol Jan 1, 2025, 12:42 AM

#

I could probably do fork in userspace without much trouble

#

as long as I expose Mm_ForkContext, of course

#

since any file descriptors appear to be closed

lean glen Jan 1, 2025, 12:51 AM

#

fork isn't that bad

lean glen Jan 1, 2025, 12:51 AM

#

flint idol since any file descriptors appear to be closed

No? Fork copies the fds

flint idol Jan 1, 2025, 12:51 AM

#

oops

lean glen Jan 1, 2025, 12:52 AM

#

Fork is an exact copy of the parent process basically

#

Which is why execing directly after is kinda stupid

#

Because you throw it all away (apart from fds)

flint idol Jan 1, 2025, 12:53 AM

#

iirc they are

#

as in, fds are thrown away

#

but I will check the man page for htis

#

*this

lean glen Jan 1, 2025, 12:53 AM

#

No

#

Unless they're marked cloexec

#

That's how stdio works

flint idol Jan 1, 2025, 12:53 AM

#

oh yeah

#

oops

lean glen Jan 1, 2025, 12:54 AM

#

Inherited from init

flint idol Jan 1, 2025, 12:54 AM

#

I misread my execve code

real pecan Jan 1, 2025, 12:54 AM

#

Is o_cloexec a joke to you LULW

flint idol Jan 1, 2025, 12:54 AM

#

if (tbl->arr[i].un.fd->flags & FD_FLAGS_NOEXEC)
    Sys_HandleClose(i | (tbl->arr[i].type << HANDLE_TYPE_SHIFT));```

#

no

flint idol Jan 1, 2025, 12:54 AM

#

flint idol ```c if (tbl->arr[i].un.fd->flags & FD_FLAGS_NOEXEC) Sys_HandleClose(i | (tb...

(this is in execve code)

#

((in a loop freeing handles))

#

real pecan Jan 1, 2025, 12:55 AM

#

flint idol ((in a loop freeing handles))

What if there's another thread in this process opening fds at the same time

#

Do u protect against that

flint idol Jan 1, 2025, 12:56 AM

#

good thing you reminded me to that

#

because I should be locking the handle table in that loop

#

actually, nvm

#

all other threads of that process would be killed by then

#

for (thread_node* curr = Core_GetCurrentThread()->proc->threads.head; curr; )
{
    thread* const thr = curr->data;
    curr = curr->next;
    if (thr == Core_GetCurrentThread())
        continue;

    OBOS_Kill(Core_GetCurrentThread(), thr, SIGKILL);
    while (~thr->flags & THREAD_FLAGS_DIED)
        OBOSS_SpinlockHint();
}```

#

according to this loop

real pecan Jan 1, 2025, 12:57 AM

#

U sure thats compliant behavior? KEKW

flint idol Jan 1, 2025, 12:57 AM

#

I think I remember reading threads of the process are not preserved

real pecan Jan 1, 2025, 12:58 AM

#

Makes sense for exec, but are u supposed to kill them with sigkill

flint idol Jan 1, 2025, 12:58 AM

#

idk

#

it's not like the threads can even do anything about SIGKILL

#

real pecan Jan 1, 2025, 12:58 AM

#

Anyway that's a separate issue

flint idol Jan 1, 2025, 12:59 AM

#

it only becomes a problem when it breaks something

real pecan Jan 1, 2025, 12:59 AM

#

It breaks my heart

#

(Jk lmao)

flint idol Jan 1, 2025, 1:00 AM

#

lmao

#

anyway I need to fix this bug

#

before new year where I am

#

I need to get obos' fireworks test working

real pecan Jan 1, 2025, 1:01 AM

#

Like userspace?

flint idol Jan 1, 2025, 1:01 AM

#

no

#

(t yet)

real pecan Jan 1, 2025, 1:01 AM

#

Didnt u have that test Workin

flint idol Jan 1, 2025, 1:01 AM

#

yes

#

I merged userspace-work

#

causing regressions

#

and crashes while loading some other driver

#

because of ||memory corruption||

#

(obos curse always returns)

real pecan Jan 1, 2025, 1:03 AM

#

💀

flint idol Jan 1, 2025, 1:03 AM

#

I am currently checking the diff to find out where I went wrong

vale nymph Jan 1, 2025, 1:09 AM

#

real pecan Well crescent is like serenity because full nih

Astral is like serenity (in implementation meme )

real pecan Jan 1, 2025, 1:09 AM

#

vale nymph Astral is like serenity (in implementation <:meme:575445094589661215>)

LULW

#

Well no

flint idol Jan 1, 2025, 1:09 AM

#

I'm going to start copy pasting files from before I merged the branch until something changes

real pecan Jan 1, 2025, 1:09 AM

#

astral has working smp
no mutexes taken under spinlocks

#

ext not corrupting itself after making 2 files

white mulch Jan 1, 2025, 1:10 AM

#

wait serenity doesn't even have working smp?

real pecan Jan 1, 2025, 1:10 AM

#

Nah

#

It crashes a few seconds after boot

#

Depending on your luck

white mulch Jan 1, 2025, 1:10 AM

#

bru

flint idol Jan 1, 2025, 1:11 AM

#

flint idol I'm going to start copy pasting files from before I merged the branch until some...

I suspect it's something silly like me forgetting to pass a correct size to Free

vale nymph Jan 1, 2025, 1:11 AM

#

real pecan >ext not corrupting itself after making 2 files

What the fuck

real pecan Jan 1, 2025, 1:11 AM

#

white mulch bru

They never bothered stabilizing it

flint idol Jan 1, 2025, 1:11 AM

#

flint idol I suspect it's something silly like me forgetting to pass a correct size to Free

the crash specifically is in Mm_Allocator

white mulch Jan 1, 2025, 1:12 AM

#

at least I have been trying to keep my kernel relatively stable fixing all bugs that I find even if they only happen once in a large number of boots

real pecan Jan 1, 2025, 1:12 AM

#

Exactly

flint idol Jan 1, 2025, 1:13 AM

#

I am afraid this is a.. allocator bug

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

real pecan Jan 1, 2025, 1:13 AM

#

flint idol I am afraid this is a.. allocator bug

Literally how

flint idol Jan 1, 2025, 1:13 AM

#

that's what I am trying to figure out

#

luckily for my sanity, it seems to prod every boot in around the same place

#

nBytes = (size_t)1 << (64-__builtin_clzll(nBytes));```
this code is correct to round up size to the nearest power of two, right?

vale nymph Jan 1, 2025, 1:15 AM

#

Just copy astral's slab, bam no more allocator bugs meme

white mulch Jan 1, 2025, 1:15 AM

#

I also found out funny bugs in my vmem today (or well yesterday, its 3:14 here already) that happen when you pass in a min addr + max addr (like I did for MAP_FIXED when splitting regions), it only looked at a freelist one order higher than where the free was put at because the size wasn't power of two KEKW

real pecan Jan 1, 2025, 1:15 AM

#

F

flint idol Jan 1, 2025, 1:16 AM

#

but why would it be an allocator bug if the allocator worked in the exact same configuration, but before merging master

white mulch Jan 1, 2025, 1:17 AM

#

flint idol ```c nBytes = (size_t)1 << (64-__builtin_clzll(nBytes));``` this code is correct...

yes, though its going to overallocate if the size is exactly power of two

flint idol Jan 1, 2025, 1:17 AM

#

that's fine

#

I'd rather it over allocate than under allocate

white mulch Jan 1, 2025, 1:17 AM

#

that's also easy to fix, just pass it nBytes - 1

flint idol Jan 1, 2025, 1:17 AM

#

I'll fix that after fixing the bug

white mulch Jan 1, 2025, 1:18 AM

#

though you do have to make sure that nBytes is >= 2 when you do that or otherwise you get UB

flint idol Jan 1, 2025, 1:18 AM

#

the size is rounded up to 16 if it's < 16

#

so that's fine

#

weirdly that gets the kernel to boot further

#

where it then crashes in the non paged pool allocator

#

but the weirdest part about this bug is the fact that it crashes here:

remove_node(c->free, c->free.tail);```

#

(remove_node is a macro)

#

I guess I need kasan now

#

obos doesn't build on GCC anymore

#

it complains about symbols that should be defined

#

which in fact are defined

#

in a static library

#

fixed

#

TIL you can use lto to get extended results from -fanalyzer

#

wait nvm

#

so yeah this all worked perfectly fine before the merge

#

after I made sure

#

uh

#

when I add printfs it no longer crashes

vale nymph Jan 1, 2025, 2:04 AM

#

Ub!!!!! :D

flint idol Jan 1, 2025, 2:04 AM

#

indeed

#

I did "test the allocator in userspace"

#

(obos userspace, that is)

flint idol Jan 1, 2025, 2:04 AM

#

flint idol (obos userspace, that is)

which doesn't say much

#

I can reproduce on gcc

#

ubsan doesn't complain

#

but everytime it faults, it faults here:

remove_node(c->free, c->free.tail);```

#

what the hell

#

c->free.tail->next is non-null

#

which can only mean one thing, memory corruption

#

specifically it's some random address in the kernel

#

amazing, under a debugger the bug goes away

vale nymph Jan 1, 2025, 2:36 AM

#

@flint idol what even is your reasoning for having ipl be a per thread thing

#

Irql

#

Whatever you call it

flint idol Jan 1, 2025, 2:37 AM

#

idr, it was a decision I made months ago

#

and never bothered to document in any way, shape, or form

#

ub is so fun 🥰

#

especially when UBSan catches nothing

#

nor does an entire static analyzer

empty kernel Jan 1, 2025, 2:45 AM

#

I thought you had escaped the worst of your curse in this kernel, but obviously not

flint idol Jan 1, 2025, 2:48 AM

#

ok I got it to reproduce in a debugger

#

I was able to find a bug

#

I was freeing an object with a different allocator than the one I was allocating with

#

fixed

#

for future reference, always free with the same allocator you allocate with

#

but I swear this is the third time I've fixed the exact same bug

#

this same bug happened in the same file, but in a different context

#

and I had fixed it in commit b5462267e9d71dd76e63323509d524b95a908515

main girder Jan 1, 2025, 3:14 AM

#

Why do u have multiple alligators

flint idol Jan 1, 2025, 3:14 AM

#

sorry, alligators?

main girder Jan 1, 2025, 3:15 AM

#

Allocators

flint idol Jan 1, 2025, 3:15 AM

#

oh allocators

#

one is for the memory manager
one is for allocating non-paged memory
one is for allocating "normal" memory

#

that design is shit, I hate it

#

and I do wish I hadn't done it

#

but now I'm stuck with it

#

that is unless I want to replace 633 occurences of allocation with a new interface

flint idol Jan 1, 2025, 3:20 AM

#

flint idol one is for the memory manager one is for allocating non-paged memory one is for ...

however, they all have the same underlying implementation, except for the fact that they map memory differently

thick jolt Jan 1, 2025, 3:26 AM

#

flint idol one is for the memory manager one is for allocating non-paged memory one is for ...

please just have a slab allocator then a vmm address space allocator

#

simple

#

simple dimple

flint idol Jan 1, 2025, 3:26 AM

#

go away

thick jolt Jan 1, 2025, 3:27 AM

#

trl

#

i mean

#

it would work better tho

flint idol Jan 1, 2025, 3:27 AM

#

yes it would

flint idol Jan 1, 2025, 3:27 AM

#

flint idol that is unless I want to replace 633 occurences of allocation with a new interfa...

but this

thick jolt Jan 1, 2025, 3:27 AM

#

flint idol but this

||get to work /j||

#

seriously tho maybe i could help with that tmrw

#

ill see

flint idol Jan 1, 2025, 3:28 AM

#

I don't plan on doing that anytime soon

thick jolt Jan 1, 2025, 3:28 AM

#

understandable

flint idol Jan 1, 2025, 3:29 AM

#

the commit would be at least +633 -633

#

changes

#

just to change it to use the new interface

thick jolt Jan 1, 2025, 3:29 AM

#

fair

flint idol Jan 1, 2025, 3:31 AM

#

if I were to make a new interface, it would look like:

enum {
    ALLOCATOR_REGION_GENERAL,
    ALLOCATOR_REGION_NON_PAGED,
    ALLOCATOR_REGION_VMM,
};
void *OBOS_Allocate(size_t sz, uint32_t alloc_region);
void *OBOS_ZeroAllocate(size_t cnt, size_t sz, uint32_t alloc_region);
void OBOS_Reallocate(void* what, size_t new_sz, size_t old_sz);
void OBOS_Free(void* what, size_t sz);```

#

I was able to get the obos fireworks test working again

short mortar Jan 1, 2025, 11:43 AM

#

empty kernel I find it weird that you have execve, but not fork. You can start a new program,...

posix_spawn?

flint idol Jan 1, 2025, 1:22 PM

#

I have been working on substituting certain strings in commands for others (like variables ig) in obos-strap
(example) "${nproc} $PWD $DISPLAY"
${varname} substitutes a preset variable (e.g., prefix directory) with it's actual value
Example: ${nproc} would get substituted with the amount of processors on the system
Example: ${prefix} would get with a full path to the prefix
$ENV_NAME is substituted with whatever the value of ENV_NAME is
Example: $PWD gets substituted with the current working directory

#

hopefully I can get this tool working by the end of today, although it won't be the fanciest, I don't really care

#

as I can always expand it

flint idol Jan 1, 2025, 3:31 PM

#

    struct stat st = {};
    if (stat(pkg->name, &st) == -1)
        mkdir(pkg->name, 0666);
    int dir_fd = open(pkg->name, O_DIRECTORY);
    printf("Entering directory %s\n", pkg->name);
    if (fchdir(dir_fd) == -1)
    {
        perror("chdir");
        return false;
    }```
because bah, this `fchdir` fails with EACCES

#

because "Search permission was denied on the directory open on fd." according to the manpage

#

but the question is how I allow it

white mulch Jan 1, 2025, 3:34 PM

#

flint idol ```c struct stat st = {}; if (stat(pkg->name, &st) == -1) mkdir(...

you need to allow execution within the directory

flint idol Jan 1, 2025, 3:34 PM

#

oh

#

thanks qwinci

white mulch Jan 1, 2025, 3:35 PM

#

np

flint idol Jan 1, 2025, 3:40 PM

#

I just built the first test package using obos-strap

#

{
    "name": "hello",
    "description": "Print a friendly, customizable greeting.",
    "url": "https://ftp.gnu.org/gnu/hello/hello-2.12.1.tar.gz",
    "depends": [],
    "patches": [],
    "bootstrap-commands": [
        [ "${repo_directory}/hello-2.12.1/bootstrap" ],
        [ "${repo_directory}/hello-2.12.1/configure", "--prefix", "${prefix}" ]
    ],
    "build-commands": [
        [ "make", "-j${nproc}" ]
    ],
    "install-commands": [
        [ "make", "install" ]
    ]
}

#

next up is supporting git-url and git-commit

#

which is for cloning

haughty abyss Jan 1, 2025, 3:42 PM

#

imagine using json

#

also you dont want -j $(nproc)

white mulch Jan 1, 2025, 3:43 PM

#

its a variable

haughty abyss Jan 1, 2025, 3:43 PM

#

yes same thing

#

you dont want that

white mulch Jan 1, 2025, 3:43 PM

#

no

#

?

haughty abyss Jan 1, 2025, 3:43 PM

#

the point is that you dont want to use -j anything

flint idol Jan 1, 2025, 3:44 PM

#

haughty abyss the point is that you dont want to use -j anything

I know, it's just there in case something needs it

haughty abyss Jan 1, 2025, 3:44 PM

#

you want to parallel build multiple packages

flint idol Jan 1, 2025, 3:44 PM

#

flint idol I know, it's just there in case something needs it

and I was testing substitution

haughty abyss Jan 1, 2025, 3:44 PM

#

and not build one package in parallel

flint idol Jan 1, 2025, 3:44 PM

#

haughty abyss you want to parallel build multiple packages

TODO:

white mulch Jan 1, 2025, 3:44 PM

#

eh

haughty abyss Jan 1, 2025, 3:44 PM

#

haughty abyss you want to parallel build multiple packages

because that is significantly faster

#

because it lets you parallelize configure

flint idol Jan 1, 2025, 3:44 PM

#

I will be doing that

#

I just want to get the base build command done

haughty abyss Jan 1, 2025, 3:45 PM

#

or alternatively you want to run with -j but with a custom jobserver that limits concurrency at the global level

white mulch Jan 1, 2025, 3:46 PM

#

it also depends on the packages whether its faster, with autogarbage ones sure but if they use cmake/meson the configure time is usually insignificant

flint idol Jan 1, 2025, 3:47 PM

#

when I do parallel package builds, I will need to make sure that if two+ packages depend on the same thing (i.e., bash and coreutils depending on mlibc), that only one of them builds the mlibc package

haughty abyss Jan 1, 2025, 3:47 PM

#

flint idol when I do parallel package builds, I will need to make sure that if two+ package...

yes

haughty abyss Jan 1, 2025, 3:47 PM

#

white mulch it also depends on the packages whether its faster, with autogarbage ones sure b...

yeah but autogarbage is like

#

most of the stuff you build

haughty abyss Jan 1, 2025, 3:47 PM

#

haughty abyss yes

but thats easy enough to do

flint idol Jan 1, 2025, 3:47 PM

#

ye

haughty abyss Jan 1, 2025, 3:48 PM

#

you build a dep graph

#

you make sure that it is acyclic

#

and then when you finish building a package P, for each package Q where P is in Q.depends, you decrement Q.num_missing_deps

#

and if Q.num_missing_deps == 0 then you add it to the queue

#

repeat until the queue is empty

white mulch Jan 1, 2025, 3:50 PM

#

maybe I should do something like that too with some kind of way to indicate that a package uses all parallelism on its own (for use in packages like llvm)

#

or well ig you could monitor cpu usage and do it automatically, idk how hard would that be to do (as you'd have to take into account that there might be spikes in usage too)

haughty abyss Jan 1, 2025, 3:51 PM

#

white mulch or well ig you could monitor cpu usage and do it automatically, idk how hard wou...

you can use the jobserver protocol

#

every build system worth its salt handles it

#

and that lets you limit concurrency globally

haughty abyss Jan 1, 2025, 3:51 PM

#

haughty abyss every build system worth its salt handles it

make, ninja, cargo do at least

#

and other ones dont exist anyway

flint idol Jan 1, 2025, 3:51 PM

#

shell script trl

haughty abyss Jan 1, 2025, 3:52 PM

#

not parallel

#

so doesnt matter

white mulch Jan 1, 2025, 3:57 PM

#

haughty abyss make, ninja, cargo do at least

does ninja? I only see it in a pr

haughty abyss Jan 1, 2025, 3:57 PM

#

white mulch does ninja? I only see it in a pr

if it doesnt that would suck

flint idol Jan 1, 2025, 5:55 PM

#

/*
 * build - DONE
 * clean - DONE
 * buildall - TODO
 * rebuild - DONE
 * setup-env - DONE
 * force-unlock - DONE
 * install - DONE
 * installall - TODO
 * chroot - DONE
 * git repos - TODO
 */```
I am nearly done the base of `obos-strap`

#

before it becomes usable for OBOS

#

for git repos is it really worth it to use a library like libgit2

#

I would probably save a lot of time if I just used the git command

#

within my C code

lean glen Jan 1, 2025, 5:57 PM

#

do you do stuff in a container?

flint idol Jan 1, 2025, 5:57 PM

#

no

#

and I don't plan on doing so anytime soon

lean glen Jan 1, 2025, 5:57 PM

#

cringe

flint idol Jan 1, 2025, 5:59 PM

#

flint idol I would probably save a lot of time if I just used the `git` command

static bool extract_archive(const char* url, const char* name, char* archive_path)
{
    // TODO: Use a library?
    string_array argv = {};
    string_array_append(&argv, "tar");
    string_array_append(&argv, "-xf");
    string_array_append(&argv, archive_path);
    int ret = run_command("tar", argv);
    if (ret != EXIT_SUCCESS)
        printf("Could not run program 'tar'. Exit status: %d\n", ret);
    return ret == EXIT_SUCCESS;
}```
just like how I decided I couldn't be damned to use a library to extract the archive

lean glen Jan 1, 2025, 5:59 PM

#

oh

#

god

haughty abyss Jan 1, 2025, 5:59 PM

#

i agree its kinda cringe

#

also yeah oh god

flint idol Jan 1, 2025, 5:59 PM

#

yeah ik

haughty abyss Jan 1, 2025, 5:59 PM

#

also what if you used a good language

#

i.e. not C

lean glen Jan 1, 2025, 5:59 PM

#

go or rust would be fine here

haughty abyss Jan 1, 2025, 5:59 PM

#

yeah

lean glen Jan 1, 2025, 5:59 PM

#

even python

haughty abyss Jan 1, 2025, 5:59 PM

#

i was about to suggest go and rust

lean glen Jan 1, 2025, 5:59 PM

#

doesnt need speed

haughty abyss Jan 1, 2025, 5:59 PM

#

python no

flint idol Jan 1, 2025, 5:59 PM

#

I don't know neither go, rust, nor python

haughty abyss Jan 1, 2025, 5:59 PM

#

python is significantly slower

lean glen Jan 1, 2025, 6:00 PM

#

I've done a similar tool before in python meme

haughty abyss Jan 1, 2025, 6:00 PM

#

flint idol I don't know neither go, rust, nor python

something something learning

flint idol Jan 1, 2025, 6:00 PM

#

bah

haughty abyss Jan 1, 2025, 6:00 PM

#

you should learn go

flint idol Jan 1, 2025, 6:00 PM

#

https://tenor.com/view/-gif-5323316

Tenor

Dramatic Chipmunk

▶ Play video

lean glen Jan 1, 2025, 6:00 PM

#

wonder if there's some kind of cross platform container library thing

#

I mean you could just call podman too

haughty abyss Jan 1, 2025, 6:01 PM

#

lean glen wonder if there's some kind of cross platform container library thing

docker api lmao

lean glen Jan 1, 2025, 6:01 PM

#

yeah but like

haughty abyss Jan 1, 2025, 6:01 PM

#

and its only mildly attrociously slow

#

also runc

lean glen Jan 1, 2025, 6:01 PM

#

some kind of library that spins up VMs on e.g macOS

haughty abyss Jan 1, 2025, 6:01 PM

#

oh i dont think that exists

#

youd have to mess around with docker

lean glen Jan 1, 2025, 6:01 PM

#

yea

haughty abyss Jan 1, 2025, 6:01 PM

#

which sucks already

lean glen Jan 1, 2025, 6:01 PM

#

or podman

haughty abyss Jan 1, 2025, 6:01 PM

#

or podman

lean glen Jan 1, 2025, 6:01 PM

#

podman handles it already

haughty abyss Jan 1, 2025, 6:02 PM

#

yeah docker too

lean glen Jan 1, 2025, 6:02 PM

#

no you need docker desktop on macOS

haughty abyss Jan 1, 2025, 6:02 PM

#

they both have nice go apis too

haughty abyss Jan 1, 2025, 6:02 PM

#

lean glen no you need docker desktop on macOS

same thing

lean glen Jan 1, 2025, 6:02 PM

#

podman you can just do podman machine init/create

flint idol Jan 1, 2025, 6:02 PM

#

haughty abyss oh i dont think that exists

system("qemu-system-aarch64 -drive file=macos.img,format=raw")

#

trl

haughty abyss Jan 1, 2025, 6:02 PM

#

flint idol `system("qemu-system-aarch64 -drive file=macos.img,format=raw")`

thats shitty and slow

#

hm

lean glen Jan 1, 2025, 6:02 PM

#

lean glen podman you can just do `podman machine init`/create

but even then that sucks

haughty abyss Jan 1, 2025, 6:02 PM

#

maybe that is a viable option

haughty abyss Jan 1, 2025, 6:02 PM

#

lean glen but even then that sucks

theres a go api

lean glen Jan 1, 2025, 6:02 PM

#

because you cant have multiple machines started/created

lean glen Jan 1, 2025, 6:02 PM

#

haughty abyss theres a go api

it doesnt support it

#

the python one didnt atleast

haughty abyss Jan 1, 2025, 6:03 PM

#

lean glen it doesnt support it

which one doesnt

flint idol Jan 1, 2025, 6:03 PM

#

anyway, I already wrote this code, and I don't wanna rewrite it, so when I want to deshittify this, I will do so

weak kestrel Jan 1, 2025, 6:03 PM

#

push to github and run a workflow

haughty abyss Jan 1, 2025, 6:03 PM

#

there is a literal package named github.com/containers/libpod/v2

lean glen Jan 1, 2025, 6:03 PM

#

haughty abyss which one doesnt

idk the python one didnt

#

and it's like the official one

haughty abyss Jan 1, 2025, 6:03 PM

#

https://podman.io/blogs/2020/08/10/podman-go-bindings.html#connect-service

Podman Go bindings | Podman

podman logo

#

the go shit is official too

#

yeah idk

#

i should try the podman bindings

flint idol Jan 1, 2025, 6:03 PM

#

flint idol anyway, I already wrote this code, and I don't wanna rewrite it, so when I want ...

*it's not like anyone even is missing tar on their systems anyway trl *

haughty abyss Jan 1, 2025, 6:03 PM

#

the docker ones suck ass

lean glen Jan 1, 2025, 6:03 PM

#

haughty abyss i should try the podman bindings

https://docs.podman.io/en/v5.2.2/markdown/podman-machine.1.html

#

they dont support this

#

so you need to call it yourself

#

which suuuucks

flint idol Jan 1, 2025, 6:04 PM

#

I do use libcurl, as opposed to running the curl command

#

at least

lean glen Jan 1, 2025, 6:04 PM

#

https://github.com/rdmsr/ck-bootstrap/blob/main/meta/plugins/bstrap.py

GitHub

ck-bootstrap/meta/plugins/bstrap.py at main · rdmsr/ck-bootstrap

Bootstrapping OS distributions using cutekit. Contribute to rdmsr/ck-bootstrap development by creating an account on GitHub.

#

here's my attempt

#

not that bad, but it's a bunch of calling the shell

#

there's a recipe example in recipes/hello.json

flint idol Jan 1, 2025, 6:06 PM

#

does everyone use gnu hello for testing their bootstrap thing

#

{
    "name": "hello",
    "description": "Print a friendly, customizable greeting.",
    "url": "https://ftp.gnu.org/gnu/hello/hello-2.12.1.tar.gz",
    "depends": [],
    "patches": [],
    "bootstrap-commands": [
        [ "${repo_directory}/hello-2.12.1/bootstrap" ],
        [ "${repo_directory}/hello-2.12.1/configure", "--prefix", "${prefix}" ]
    ],
    "build-commands": [
        [ "make", "all" ]
    ],
    "install-commands": [
        [ "make", "install" ]
    ]
}```

#

because that's my test thing

#

hmm maybe I should merge bootstrap and build steps

#

they are done right after each other in the build code anyway, so merging them simply removes some complexity from the recipe

lean glen Jan 1, 2025, 6:09 PM

#

https://github.com/rdmsr/ck-bootstrap/blob/main/recipes/hello.json

GitHub

ck-bootstrap/recipes/hello.json at main · rdmsr/ck-bootstrap

Bootstrapping OS distributions using cutekit. Contribute to rdmsr/ck-bootstrap development by creating an account on GitHub.

#

I think this was good

flint idol Jan 1, 2025, 6:09 PM

#

yeah it looks good

#

anyway, I will be implementing these stuff next:

git repo fetching (may or may not just run a git command or two)
buildall command
installall command

#

then I will make some recipes for obos probably

#

and I'll be making these recipes

bootstrap-gcc
mlibc
gcc
bash

#

bootstrap-gcc is with inhibitLibc=true

#

mlibc is built with that

#

then gcc is built with mlibc

#

then bash is built with mlibc+gcc

#

ig

#

so:
gcc: depends mlibc
mlibc: depends bootstrap-gcc
bash: depends gcc, mlibc
bootstrap-gcc: no dependencies

#

and thinking about it, I might want to separate host recipes and target recipes

#

which I will do after I add git repository fetching

lean glen Jan 1, 2025, 6:17 PM

#

flint idol anyway, I will be implementing these stuff next: - git repo fetching (may or may...

dont do git really

#

just do tarballs

haughty abyss Jan 1, 2025, 6:19 PM

#

do git

#

its the cool thing to do

flint idol Jan 1, 2025, 6:20 PM

#

lean glen dont do git really

but then how do I apply target-specific changes to the thing

#

to the downloaded tarball

haughty abyss Jan 1, 2025, 6:20 PM

#

you can apply patches as normal

lean glen Jan 1, 2025, 6:20 PM

#

what does that mean

flint idol Jan 1, 2025, 6:20 PM

#

probably a stupid question

lean glen Jan 1, 2025, 6:20 PM

#

??

flint idol Jan 1, 2025, 6:20 PM

#

lean glen what does that mean

how do I apply patches

#

to the extracted tarball

lean glen Jan 1, 2025, 6:20 PM

#

using patch ?

flint idol Jan 1, 2025, 6:20 PM

#

oh that exists

lean glen Jan 1, 2025, 6:21 PM

#

bruh

flint idol Jan 1, 2025, 6:21 PM

#

silly me

haughty abyss Jan 1, 2025, 6:21 PM

#

lmfao

#

you should use git though

lean glen Jan 1, 2025, 6:21 PM

#

patches existed way before git did

flint idol Jan 1, 2025, 6:21 PM

#

well obviously

lean glen Jan 1, 2025, 6:21 PM

#

did you think people just applied them by hand?

#

lol

flint idol Jan 1, 2025, 6:25 PM

#

yeah I will be adding git in case I somehow come across a tool/thing I want to port without a tarball

#

which sounds unlikely, as anything that is on github and has a release will have a tarball

#

but just in case

lean glen Jan 1, 2025, 6:27 PM

#

even then everything on github is tarballed iirc

#

like you can fetch a tarball of a repo even without it being published (the tarball, that is)

flint idol Jan 1, 2025, 6:27 PM

#

flint idol which sounds unlikely, as anything that is on github and has a release will have...

.

lean glen Jan 1, 2025, 6:27 PM

#

no need for releases

#OBOS (not vibecoded)