It's not too complicated
#Nyaux
edgy pilot
they are character devices. your input device sends them, well, input. they process it according to the flags set (found in manpages). then you just read data from them. writing sends data to the backend driver, like the display
flat nymph
I get that
But how do they take over displays, how are keyboard drivers supposed to interract with them, etc.
what's the role of init server in all of it
edgy pilot
But how do they take over displays, how are keyboard drivers supposed to interra...
just have an object that you create your tty with that contains the input and output drivers
edgy pilot
what's the role of init server in all of it
hm?
flat nymph
(microkernel shenanigans)
flat nymph
because I'd suppose my tty would habe to be a userspace server
what about multiple displays
edgy pilot
n ttys per display?
broken depot
what about multiple displays
Usually those are just cloned actually
brisk zenith
But how do they take over displays, how are keyboard drivers supposed to interra...
the way i'd do it is to have something like kmscon, it interacts with the display driver (probably fb) and input drivers and presents a pty
flat nymph
like I've never written a big userspace application, I just don't have a feeling of how the things are connected
Linux KMS/DRM based virtual Console Emulator
Time to nag @ ma_rv about uDRM
edgy pilot
udrm 
brisk zenith
> Linux KMS/DRM based virtual Console Emulator
Time to nag @ ma_rv about uDRM <...
i mean that part can just be replaced with open("/dev/fb"), mmap, and flanterm
edgy pilot
> Linux KMS/DRM based virtual Console Emulator
Time to nag @ ma_rv about uDRM <...
there are fb based emulators as well
I think one was just called fbcon or fbterm
flat nymph
I mean I have flanterm, and a keyboard driver
And userspace
Just no glue for them to run bash
flat nymph
I mean I have flanterm, and a keyboard driver
(I mean all of that runs in userspace)
flat nymph
And userspace
And even fork
edgy pilot
Just no glue for them to run bash
I was going to say that "you just need to glue them together now"
edgy pilot
And even fork
do clone() and use that for fork
flat nymph
No, I mean I have full fork already
Which creates a new process and clones open files
and I have POSIX pipes...
flat nymph
and execlp
Because it's very inconvenient for a microkernel because of setuid
What if I do the terminal thing in Rust? 
flat nymph
https://tenor.com/view/starwars-no-darth-vader-yell-noooooo-gif-5858146
the kernel stays in C++ 
flat nymph
I mean I also have Go working...
tardy hull
i need someone to tell me to finish PMM, ping me
i need motivation
need someone to tell me to finish memory managment today
the whole thing
flat nymph
@tardy hull
tardy hull
<@468392931217375240>
ty, going to make resize and allocate for my PMM allocator then do another allocator that isnt bitmap and is for random shit physKmalloc
rich trellis
@tardy hull finish the pmm or i will tickle your feet when you go to sleep
tardy hull
<@468392931217375240> finish the pmm or i will tickle your feet when you go to s...
ty
ty guys
also opinion on what allocator for kmalloc?
rich trellis
im struggling with the pmm and vmm tho so idk
worn rain
with the pmm?
cinder plinth
slab is really good
worn rain
- you dont
flat nymph
I just use dlmalloc 
worn rain
- bitmaps are slow af
flat nymph
bitmaps are ok, but why would you use them when freelist exists
tardy hull
bitmaps are ok, but why would you use them when freelist exists
i use it :3
rich trellis
okay but how could i implement a freelist for the pmm?
worn rain
buddy allocator all the way
rich trellis
im kind of a dumbass
flat nymph
For vmm, read the second part of this paper (vmem)
worn rain
he was talking about the pmm
flat nymph
okay but how could i implement a freelist for the pmm?
Just make a list of free pages
cinder plinth
okay but how could i implement a freelist for the pmm?
Do you know what is a freelist?
cinder plinth
its a linked list of pages
Not necessarily, a freelist is not unique to physical memory allocation
worn rain
well, you traverse the layers but thats still O(1)
cinder plinth
buddy is not O(1)
worn rain
tehnically yes
but if you cap the layers to like 10
then its rather wrong to say its logn
coral dove
Im pretty sure most operating systems limit the amount of layers right?
worn rain
ofc
cinder plinth
I may look into buddy eventually
coral dove
It’s pretty easy to implement
rich trellis
i think ill just go with bitmap
coral dove
i think ill just go with bitmap
Why
worn rain
no please dont
rich trellis
i will probably solve it
worn rain
instant allocation
surreal path
from here whats going on
worn rain
instant allocation
xd
surreal path
im tryna nap some more
rich trellis
its less code too
interesting
brisk zenith
why has the nyaux thread become lounge 3 (lounge 2 is uacpi)
worn rain
im tryna nap some more
mute
coral dove
OH SHIT
surreal path
why has the nyaux thread become lounge 3 (lounge 2 is uacpi)
exactly
worn rain
OH I SEE
worn rain
we are in nyaux
worn rain
id prefer for my thread to not be cluttered lol
40K MESSAGES
tardy hull
im tryna nap some more
no sleep for you, only cat bites and cat annoyance
rich trellis
oh shit
freelist just made sense on my head
oh shit
what have i been doing all this time
surreal path
please move to somewhere else thank you
rich trellis
okay
flat nymph
buddy is not O(1)
It is O(1), but slower than freelist since you may have to be splitting pages
edgy pilot
im technically in bash tho
get to ze prompt and you win
there's a chance that I won't have much time to work on my kernel in the coming days
edgy pilot
the osdev race
flat nymph
But I want to do it in Rust
(Which implies that I should finally write the init server)
surreal path
okay so time to implement ttys
rich trellis
tities tty when
ebon needle
r/unixporn...
rich trellis
you can think that will tickle your feet when you sleep if you dont work
surreal path
nah thats crazy exit code 0 is crazy
quick quiver
It is O(1), but slower than freelist since you may have to be splitting pages
It's O(log n) as abbix said as you potentially need to coalesce log n pages to a single huge page
surreal path
It's O(log n) as abbix said as you potentially need to coalesce log n pages to a...
eletrorb said u know everything lmao
worn rain
why would you ever create a single huge page tho
worn rain
just cap the max level
thorn bramble
so you still need to potentially coalesce smaller chunks into bigger ones
flat nymph
It's O(log n) as abbix said as you potentially need to coalesce log n pages to a...
But it's still tecnically O(1)
thorn bramble
no matter what you do
surreal path
watching these to understand unix more and to understand how to impl a tty, after this video series ill read the tty article
flat nymph
But it's still tecnically O(1)
You might have to split one giant page too
worn rain
so you still need to potentially coalesce smaller chunks into bigger ones
thats how it becomes O(1)
when adding a cap
surreal path
watching these to understand unix more and to understand how to impl a tty, afte...
i already know most of what im watching here but
worn rain
lets not talk about this here tho
thorn bramble
no, if you coalesce it never becomes O(1)
surreal path
i already know most of what im watching here but
its good to reinforce
flat nymph
no, if you coalesce it never becomes O(1)
like that number is capped
thorn bramble
ok maybe it comes close for very small N
flat nymph
but yeah, it's not a freelist where you just push to it
flat nymph
my allocator at most just coalesces with free region before it and region after it
quick quiver
But it's still tecnically O(1)
"Technically" a bitmap is also O(1) if your address space is limited by 2^64 ;D
thorn bramble
buddy isn't slow tho
worn rain
we should just agree then that its logn where n is max number of levels
thorn bramble
we should just agree then that its logn where n is max number of levels
we did
surreal path
#allocators please :) u can discord forward to allocators and countinue ur convo there :) cause this isnt related to nyaux and while i dont mind the convo here and there its getting a bit much now
flat nymph
then I agree with everyone
worn rain
lovely
flat nymph
okay bros yapping at this point
no?
surreal path
no?
i mean yapping as in im starting to not understand bro anymore
because hes talking about pipes and
shit
and signals
for some reason
those are a todo to learn in my mind but not rn
need to laser focus on tty shit
flat nymph
like a water pipe literally
thorn bramble
literally tho
rich trellis
wather pihpe
thorny glen
It's O(log n) as abbix said as you potentially need to coalesce log n pages to a...
someone pointed out to me recently that its technically O(1) if you take the amount of phys mem to be a constant
but thats probably less useful for analyzing performance lol
flat nymph
here we go again
surreal path
i think i get how a tty works
silver yarrow
surreal path
theres this thing called a line discipline that can be attached to a tty, its a passive object ofc so it doesnt do anything itself without someone calling it. from what i understand the line discipline has commands and its own buffers. for example with line editing u can have editing commands like backspace and such
the default discipline provided line editing and only one line discipline can be attached to a tty at a time
i think i got that part right
tty has its own fields sure but its kinda useless without a line discipline
did i understand correctly-ish
thats what i got from reading https://www.linusakesson.net/programming/tty/
yea linux documention seems to comfirm it
cinder plinth
It is O(1), but slower than freelist since you may have to be splitting pages
yea I mean it's O(log n) but if that n is constant then it's O(1)
surreal path
yea linux documention seems to comfirm it
from reading the linux documention some more it seems the line disclpine is responable for talking to the low level driver, such as flanterm or serial
cinder plinth
yea I mean it's O(log n) but if that n is constant then it's O(1)
I think?
cinder plinth
from reading the linux documention some more it seems the line disclpine is resp...
id just try running bash and monitor all the ioctls
and implement them
surreal path
id just try running bash and monitor all the ioctls
i dont even ahve a tty yet
cinder plinth
a simple tty is pretty easy
surreal path
id just try running bash and monitor all the ioctls
i dont even have ioctl bro
cinder plinth
i dont even ahve a tty yet
yea so use the ioctls to implement it as you go
median goblet
okay bros yapping at this point
i mean honestly you can probably just think of it like this
cinder plinth
make a useless device file and make it fd 0,1,2
then monitor reads, writes, and ioctls to that device
writes for now you can just call flanterm, but in the future when you support cooked vs raw mode and termios stuff you may want to do something else
surreal path
let me cook here ill try to make an actual tty implementation
does a tty have its own buffer
why does linux have a name for a line discpline
okay reading further into this
linux has it so tty_driver install into a tty, and then the tty driver is passed into the line disclpine for its ops to be called
heres what ill do
cinder plinth
does a tty have its own buffer
yea I think so
you typically have a ringbuffer for input
surreal path
in the tty itself?
surreal path
yea I think so
and what about output?
surreal path
i want it to be done well really
cinder plinth
but like think about two users connected to two different ttys
thats not really a thing nowadays, but in old unix it was
so they mustve had a seperate buffer
surreal path
but like think about two users connected to two different ttys
yea then u'd need both a buffer for input and output
can both be ring buffers?
ring buffers are funny
@cinder plinth ?
surreal path
yea then u'd need both a buffer for input and output
how big would that buffer be
surreal path
like 255?
surreal path
big enough
also should i use a ring buffer for both input and output
surreal path
okay, im gonna steal a library that impls a ring buffer because i dont wanna do some shitty implementation
thorn bramble
surreal path
wdym nah
thorn bramble
its 5 lines of code
broken depot
i mean honestly you can probably just think of it like this
No you can't
surreal path
ill screw up those 5 lines of code
thats ze issue
if i wrote it itll js be bad
and
idk
thorn bramble
there is only one way to write a ring buffer and you either do it right or not lol
its literally a pointer and two offsets for read and write head
broken depot
No you can't
That's not how terminal input and output works in shells. The shell doesn't forward from the TTY to the child process, it gives its child process an actual handle to the TTY (unless replaced by a pipe)
flat nymph
its literally a pointer and two offsets for read and write head
memove(buffer, buffer + 1, size--);
broken depot
That's not how terminal input and output works in shells. The shell doesn't forw...
In fact, that handle is typically given by just... not doing anything. It's not marked as CLOEXEC so the exec call after the fork to create the child process won't close the TTY handles unless explicitly replaced by the shell for e.g. pipes.
broken depot
`memove(buffer, buffer + 1, size--);` <:trl:829316641191886888>
O(n²) buffer lmao
Also it's --size because you want to decrement first
flat nymph
but like that's pretty much how you would do it in hardware
broken depot
Sorta
The way you'd do it in hardware is usually actually a ring buffer
You can do what you described but it's less efficient
flat nymph
but you just chain a bunch of registers
broken depot
but you just chain a bunch of registers
Trust me, I've made FIFOs in verilog, I would know.
Like I said: what you say would work but is not how most implementations actually work because of efficiency problems or even clock-domain-crossing problems.
median goblet
No you can't
Why not
median goblet
That's not how terminal input and output works in shells. The shell doesn't forw...
In that manner you can think of it as a two channel pipe duplex
The shell doesn't do line buffering, the terminal does, but in terms of data flow you can think of it like that
broken depot
Why not
Because this implies that the shell is writing data that the first process then reads, which is not the case. The TTY already has that data.
median goblet
Well, it "already has" the data because the shell wrote that data into it.
The same reason the shell "has" the output data because it was piped in from the process
broken depot
Do you understand how TTYs generally work under Linux? Because it isn't this.
broken depot
Like I said earlier, the shell doesn't forward data from the TTY to the process, it merely gives a handle to the TTY to the process.
median goblet
Did I ever say that?
broken depot
Your picture implies it.
median goblet
No. I cut a line through the terminal character device implying it could be represented as two pipes, in a duplex
(The shell -> tty -> process pipe has some extra logic though for line buffering)
broken depot
No. I cut a line through the terminal character device implying it could be repr...
Except for the part where a TTY is one file
median goblet
Sure. A file with two different read ends and two different write ends
The shell gets one set of one read and one write end, the process gets the other, as far as I understand
(I guess this is why you can write to stdin and it will still show up on the shell)
broken depot
The problem I have with this
Is that even if you do have a shell that intercepts FDs 0, 1 and 2
The way you drew that implies something something different than that
median goblet
It implies that you can think of the terminal character device as a duplex of pipes
Nothing more
broken depot
It implies a closed loop where the TTY acts as a glorified pipe to the child process and back, which I am very confident is not what the video tries to convey.
And yes, you can create set-ups like that
broken depot
And yes, you can create set-ups like that
But then where does the actual TTY go? You'd assume the diagram would show it too and not just the communication between the shell and the child processes.
median goblet
I think I did mention it does not work exactly like that but you can think of it as such if it gives you a better idea of the actual workings
median goblet
But then where does the actual TTY go? You'd assume the diagram would show it to...
what features does the TTY have other than line buffering that I mentioned earlier?
broken depot
what features does the TTY have other than line buffering that I mentioned earli...
Why is that even relevant to this diagram?
Also the TTY has conversions for newlines
Arguably part of line buffering
median goblet
because you say "where's the TTY" when I split it up into two pipes
broken depot
You're not getting the point I'm making
median goblet
again. my point wasn't to accurately describe the operation of a terminal character device. my point was to give an idea of how communication through a TTY works: you can think of it like two pipes in a duplex
cinder plinth
robot you seem like you're being overly pedantic
broken depot
If you do the analogy of splitting into two pipes, then the way you draw it on this diagram implies to fundamentally change what that square actually does.
Because when it is just one square you're not implying that the shell uses it to send data to the child process, but with the specific way you drew, it does.
Which, once again, is fine, but not a TTY at that point
Because now the actual TTY would be somewhere else and only the shell would have access to it
median goblet
the shell uses a pipe to send and another pipe to receive
broken depot
Of course it does
But I don't think you're getting what I'm trying to explain to you.
I'm not arguing against the dual pipe analogy, I'm arguing against how you drew it here.
median goblet
to reiterate. YOU CAN THINK of it as a pipe duplex. doesn't mean that it ACTUALLY is. in particular there are certain things the TTY does (line buffering and new line conversions) that dumb pipes simply don't.
broken depot
to reiterate. YOU CAN THINK of it as a pipe duplex. doesn't mean that it ACTUALL...
To reiterate. What you drew implies something different than what I think you actually mean to me.
median goblet
Other than the pipe duplex system? What does it imply?
broken depot
The original example shows one handle to the TTY being shared to the child processes and the shell right?
For your analogy of dual pipes, I would expect the division to be such that there's two pipe objects shared by the shell and child processes, each is either written by both or read by both, and the other end is "the physical device™️".
And like I said, what you drew isn't inherently invalid but it doesn't describe the situation I think you want it to.
broken depot
robot you seem like you're being overly pedantic
Certain kinds of simplifications can be harmful. A simplification that actually represents a different but equally valid system is, in my opinion, one of them.
median goblet
The original example shows one handle to the TTY being shared to the child proce...
No it doesn't
Here's another attempt
Is this better
broken depot
I say this because of the sheer amount of times some random youtuber did exactly such a harmful simplification causing me problems solved by looking up the actual system.
broken depot
Here's another attempt
This describes the same thing; I think the video's intent is not to show this connection.
It could be if it had said something like "virtual TTY"
But the text reads to me like it's the actual /dev/tty* file
surreal path
okay so ive impled ring buffers
whata should the 2 read and write buffers on a tty device be called
broken depot
whata should the 2 read and write buffers on a tty device be called
The name is a bit arbitrary. There will be one used for reads and another for writes from the process. If you have a kernel driver that e.g. uses a serial port for /dev/tty0 then you might call them tx and rx for writing to and reading from the TTY respectively.
If you have a virtual TTY (a special file where one's read is the other's write and vice versa), you might even just call them A and B depending on how the data structures are represented in your kernel.
cinder plinth
Certain kinds of simplifications can be harmful. A simplification that actually ...
jesus
i think you have a stick up your ass
broken depot
i think you have a stick up your ass
The disrespect in that message aside, I'm just trying to be helpful here. I've watched the video and I know how TTYs approximately work and all I'm trying to do is clear up a misconception waiting to happen.
surreal path
If you have a virtual TTY (a special file where one's read is the other's write ...
ill call it tx and rx js cause
broken depot
I never said any of you weren't well-intentioned and I don't judge anyone for not knowing certain things or having a misconception about it. What I do judge people for is needlessly throwing insults. We're both human and, regardless of who's right in this situation, I think you wouldn't want to be described as "having a stick up the ass" either.
surreal path
umm
ringbuffer seems to kill kernel lol
my code
struct ring_buf *init_ringbuf(size_t thesize) {
struct ring_buf *ringbuf = kmalloc(sizeof(struct ring_buf));
ringbuf->size = thesize;
ringbuf->write_idx = 0;
ringbuf->read_idx = 0;
ringbuf->buf = kmalloc(align_up(thesize, 8));
return ringbuf;
}
// referenced from wikipedia but i sorta undestand it
int put_ringbuf(struct ring_buf *buf, uint64_t data) {
if ((buf->write_idx + 1) % buf->size == buf->read_idx) {
return 0;
}
assert(buf != NULL || buf->buf != NULL);
buf->buf[buf->write_idx] = data;
buf->write_idx = (buf->write_idx + 1) % buf->size;
return 1;
}
int get_ringbuf(struct ring_buf *buf, uint64_t *value) {
if (buf->read_idx == buf->write_idx) {
return 0;
}
assert(buf != NULL || buf->buf != NULL);
*value = buf->buf[buf->read_idx];
buf->read_idx = (buf->read_idx + 1) % buf->size;
return 1;
}
void resize_ringbuf(struct ring_buf *buf, size_t resize) {
assert(buf != NULL || buf->buf != NULL);
void *new = kmalloc(resize);
void *old = buf->buf;
memcpy(new, old, buf->size);
buf->size = resize;
}
Did you mean to use ``` ?
surreal path
yea
struct ring_buf {
uint64_t *buf;
size_t size;
size_t write_idx;
size_t read_idx;
};
honestly if I acted like you I wouldnt mind it
surreal path
is there anything wrong with this code @broken depot ??
cinder plinth
you should free the old buffer
surreal path
other then that
cinder plinth
and set buf->buf to new
broken depot
is there anything wrong with this code <@322418405095505920> ??
Well you're trying to send characters right?
Why is the pointer for uint64_t * in that case?
That might be what's causing your error.
surreal path
Why is the pointer for `uint64_t *` in that case?
its a generic ring buffer impl
broken depot
That's fine
In that case you should multiply the size you give kmalloc with sizeof(uint64_t)
cinder plinth
why are you writing 8 bytes at once
surreal path
oh i just realized why
surreal path
In that case you should multiply the size you give kmalloc with `sizeof(uint64_t...
yes
my bad
lol
buffer overflow like usual lmao
3bugs related to buffer overflows
how funny
broken depot
buffer overflow like usual lmao
Long live the programming language actively opposed to memory safety 
surreal path
works now
broken depot
Very cool
surreal path
its a generic impl
broken depot
its a generic impl
If you want to make it generic, couldn't you just add a field like size_t item_size to it?
Then use the uint8_t * for the pointer
tender gorge
maybe some hardware wants a ring buffer that uses 8 bytes idfk
realtek on their way to try and put even less SRAM into their network chips:
cinder plinth
maybe some hardware wants a ring buffer that uses 8 bytes idfk
your tty wont
surreal path
If you want to make it generic, couldn't you just add a field like `size_t item_...
i could but im lazy plus eepy
broken depot
lol
tender gorge
i could but im lazy plus eepy
can i offer: hugs
coral dove
Long live the programming language actively opposed to memory safety <:KEKW:9137...
C is a joy
tender gorge
<:hug:647834498758737921>
💙
coral dove
We love you nyaux you’re the goat
coral dove
```C
struct ring_buf *init_ringbuf(size_t thesize) {
struct ring_buf *ringbuf ...
I loaded this on mobile and I felt like I was reading the Declaration of Independence
tender gorge
it's on this server
isnt
surreal path
anyways my tty will look like this
struct tty_device {
}
struct tty {
struct line_disclpine *cur_disclpine;
struct ring_buf *tx;
struct ring_buf *rx;
struct tty_device *device;
}
// and then my tty would impl dev fs ops
void tty_install_device(struct tty *tty, struct tty_device *device);
void tty_install_linedisclpine(struct tty *tty, struct line_disclpine *loc);
sorry i cant spell
😭
is this a good way of doing things
its very similar to how linux does it
tender gorge
anyways my tty will look like this
```C
struct tty_device {
}
struct tty {
...
i think thats about correct?
idk its been a while since ive looked at stuff like this
surreal path
still refactoring
surreal path
how would i go about implementing sys_isatty
im a little confused on how to do that
this is all i have
edgy pilot
im a little confused on how to do that
how do you open ttys?
edgy pilot
is your code on github
surreal path
yes
edgy pilot
surreal path
how will i get that tho
edgy pilot
it's your code
surreal path
whats the best way to approach this i mean
surreal path
yes
edgy pilot
this is all i have
file descriptor
surreal path
but not all vnodes are dev devices
edgy pilot
not description
surreal path
file descriptor
too lazy to rename
edgy pilot
this is all i have
that doesn't match with the code on github
edgy pilot
too lazy to rename
literally select it and f2
surreal path
that doesn't match with the code on github
hold on
reload ur page
i just pushed upstream
surreal path
literally select it and f2
done
edgy pilot
but not all vnodes are dev devices
your vnode contains a pointer to data
and that data contains a pointer to major and minor ids
surreal path
i cant just blindly assume data is a devfsnode
edgy pilot
check?
surreal path
how so
surreal path
wait ik what to do
edgy pilot
do you write code without testing it 😭 😭
flat nymph
I just spam asserts
surreal path
I FUCKING DO SET THE NODE
edgy pilot
I just spam asserts
microkernel and full of asserts
surreal path
bro
edgy pilot
are you sure your os is not managarm?
surreal path
I FUCKING DO SET THE NODE
i need help wtf
flat nymph
are you sure your os is not managarm?
It's different
I have separate assert and panic
edgy pilot
lol
surreal path
bru
flat nymph
are you sure your os is not managarm?
And I support more architectures than they
Btw, Nyaux for RISC-V when 
flat nymph
My kernel quality has increased a lot after porting to it
rich trellis
just make a cpu emulation that runs the os and port it to all platforms
broken depot
just make a cpu emulation that runs the os and port it to all platforms<:trl:829...
QEMU OS
broken depot
surreal path
thats lit my tty atm 💀
broken depot
Now it's my turn to fix my scheduler and implement TTYs I guess
What a funny coincidence we had the same next things on our TODO lists
surreal path
indeed
flat nymph
@kind root does Ultra kernel only work on x86?
kind root
which one
the new one supports aarch64 as well
although im thinking of dropping that
flat nymph
Why
kind root
i dont want to support crapberry pi etc
drop until they have a proper standardized platform
kind root
yeah but ya know
flat nymph
So like you only support x86?
kind root
anyway arm is not a priority, the new ultra will be focused on hardware support, primarily gpus probably
my goal is to run some cool game in 3d accel on a real laptop
with cp'ed linux userland apps
flat nymph
With your own GPU protocol you said?
kind root
protocol?
flat nymph
Or uDRM lives?
kind root
i will have full linux abi compat, so no its drm
kind root
Or uDRM lives?
well thats marvin's project, I will see about making my impl generic, whether it's even possible
flat nymph
But it's part of uAPI
surreal path
unkempt relic
there is no royal road to the bash prompt, and only those who do not dread the fatiguing climb of its steep paths have a chance of gaining its luminous summit
nyaux is on its way up
unkempt relic
bash prompt
congrats
surreal path
bash prompt
@kind root
brisk zenith
yo nice
kind root
surreal path
surreal path
bash prompt
@edgy pilot
broken depot
bash prompt
tender gorge
bash prompt
NICEEEE
supple robin
But it's part of uAPI
uapi is kill
desert haven
bash prompt
nice one 🎉
surreal path
been fixing more vfs bugs
edgy pilot
how do u free vnodes
this is how I do it currently
silver yarrow
lmao
I just re-implemented my scuffed vfs from elysium5
I need to properly do ref counting this time
: )
not looking forward to that
edgy pilot
it's on my todo list
silver yarrow
I a not moving on before I do this
My main concern tbh is string ownership in my kernel
like I kind of pass around strings randomly
and uhh, it's very possible they just get freed when I start dropping vnodes and their associated data
edgy pilot
My main concern tbh is string ownership in my kernel
C++ 😎
especially with smart pointers
I can just do vnode.reset() and everything is freed
(if there are no other references to member shared pointers)
I still need refcount for other things
silver yarrow
(if there are no other references to member shared pointers)
and if there are references?
like how do u deal with that
edgy pilot
and if there are references?
for example two vnodes are linked to the same resource, one is freed, the refcount is decreased
surreal path
solved vfs bugs
edgy pilot
and the resource is freed once both vnodes are freed (refcount == 0)
surreal path
with some hacks and work arounds
silver yarrow
i am just planning on saying "u can only use this data as long as u keep the ref to the associated vnode" but that completely depends on me making sure that is followed
surreal path
bash with readline and ncurses @thorn bramble
edgy pilot
why does it say ls-5.2
snow wolf
My main concern tbh is string ownership in my kernel
a model I find quite nice is that the underlying storage of a namecache name is refcounted and immutable. for e.g. rename, you want some namecache entry to adopt the name of some other entry, so you decrement the refcount of the old one (and free it if it reaches zero), increment the refcount of the name you are adopting, and store the pointer to the new name in the namecache entry.
thorny glen
i got refcounted interned strings
snow wolf
passing namecache entries to underlying filesystem functions can also be nice. e.g. lookup(inode *dir, namecache_entry *dirent);
snow wolf
a model I find quite nice is that the underlying storage of a namecache name is ...
one thing that can complicate this scheme is short-string optimization (storing < N byte strings inline). however with a good abstraction it isn't a big deal
thorny glen
generally speaking if i have some long term string storage ill look it up in a hash table of interned strings and only allocate new space if i cant find it in there already
name cache node names, thread names, process names, etc
orchid dawn
Why is it ls-5.2 lol
surreal path
no idea
surreal path
Why is it `ls-5.2` lol
u awake btw
fuck i need stat and i aint pulling anymore shit from fucking mlibc
i need to
include mlibc headers
somehow
no idea how to set that up
orchid dawn
u awake btw
yes
orchid dawn
let me know once you've got a build for the Wii
surreal path
chuck me a build
I can try it on some shit later
surreal path
i got the tty to accept input from serial
and it works with bash
bash has no idea who i am rn
tender gorge
bash has no idea who i am rn
Congrats! 
kind root
<@353593948523986945> big?
median goblet
generally speaking if i have some long term string storage ill look it up in a h...
that's a pretty cool way to do such things
what brought you to do something like that?
flat nymph
let me know once you've got a build for the Wii <:trl:829316641191886888>
I'm waiting for @odd summit 's bootloader 
orchid dawn
lol
flat nymph
https://tenor.com/view/ron-swanson-nick-offerman-inshallah-gif-13027660
How is it going btw?
surreal path
i want to put a rope on myself
ebon needle
i want to put a rope on myself
why
surreal path
cause i hate being alive
ebon needle
and why you hate being alive
ebon needle
cause i hate being alive
what is happened ?
bleak cedar
please don't 🥺
surreal path
if i fail this attempt i swear to god ima js figure out a way to buy a bullet in ireland then 3d print a weapon to shoot myself
tender gorge
if i fail this attempt i swear to god ima js figure out a way to buy a bullet in...
Please dont
ebon needle
if i fail this attempt i swear to god ima js figure out a way to buy a bullet in...
what attempt exactly?
ebon needle
@surreal path what are you trying to do
tender gorge
@surreal path
bleak cedar
@surreal path I promise, I wanted to install a older version of Python to compile ICU 69 today
Did someone call any emergency services?
ionic musk
That's pretty worrying
bleak cedar
Did someone call any emergency services?
I'll call them now
tender gorge
Did someone call any emergency services?
Please call them
ionic musk
How do you call them if you're not from Ireland?
tender gorge
How do you call them if you're not from Ireland?
European 112
ionic musk
You can have them redirect you to another emergency service in another EU country?
tender gorge
Yes
bleak cedar
ok, police is coming to me
tender gorge
ok, police is coming to me
To you?
bleak cedar
yes, they want to check the message themself
tender gorge
Ah ok
hexed sluice
authorities have been contacted, by multiple people, this thread will be locked and left here as evidence
nyaux has been banned as to comply with Discord ToS. I don't personally like it but that's what the law of the land is
messages won't be deleted because they are evidence
hexed sluice
as an update, from what we know, Nyaux is now safe and in police custody
hexed sluice
if you people want to talk to Nyaux, he is on the OSDev matrix space #osdev:matrix.org
hexed sluice
@surreal path
surreal path
okay so UPDATE to everyone here, nyaux has reached bash since a while ago as we know. ive implemented fork() a shitty implementation but it works and now im implementing sys exit properly
after sys exit will be sys exceve
we ballin
honestly probs waitpid
before sys exceve
kind root
welcome back
flat nymph
osdev server is saved
magic kernel
@surreal path welcome back
flat nymph
Circular lists?
steady flume
surreal path
Circular lists?
no i have a previous variable
the fucking operator lol
😭
solved it
rigid fable
DAJE ROMA DAJE 💪
hexed sluice
DAJE ROMA DAJE 💪
daje Roma li mortacci
surreal path
chat what are we saying
hexed sluice
dw about it
surreal path
oki
rigid fable
porco dio e porco allah
Bro-
rigid fable
chat what are we saying
Some not halal stuff 💀
hexed sluice
i apologise on streaks's behalf
he's Spanish and doesn't know what the fuck he's saying
rigid fable
😔
alpine mulch
I am literally quoting a video of an AI generated shark with sneakers
alpine mulch
}
tender gorge
how do i tell mr thread x "hey ur child process died"
send a signal
aka implement posix signals
aka have fun :3
tender gorge
the signal is SIGCHILD
surreal path
poop
tender gorge
oh youre talking about the waitpid syscall
surreal path
yes
tender gorge
no that does not need signals
rigid fable
looks complicated
surreal path
then how is the thread told
its child process died or its child process is alive idk what bash wants
lol
tender gorge
waitpid waits for the process with the PID to exit
how exactly you implement that in your kernel is not something posix cares too much about iirc
tender gorge
waitpid waits for the process with the PID to exit
wait() waits for the next child process to finishj executing
tender gorge
what would be the best way to implement that
iirc the way linux does it is by keeping the process structures in memory/as a process until it has been waited on
which is also i think what posix wants?
surreal path
where is it stored
tender gorge
as in, if you never wait() a process, it will stay alive as a zombie process (which linux reports correctly as a zombie)
surreal path
yes it is in state zombie
but where is the process stored
it has no more connected to no threads
tender gorge
with the other processes
surreal path
all its threads are dead
tender gorge
it has no more connected to no threads
correct, it has no threads
it still exists as a process though until it has been waited on
surreal path
correct, it has no threads
i dont have an area to store processes
tender gorge
i dont have an area to store processes
ok so you probably want one
surreal path
i dont have an area to store processes
i only store threads
tender gorge
i think
probably best to ask korona/qwinci/oberrow on that, i have never gotten past very simply TTY stuff 
so uhh, i dont really know here
surreal path
ok so you probably want one
would something like storing it in the cpu local struct be suitable
because that would be the easiest approach
alpine mulch
processes and threads are necessarily stored separately
unless you are linux and threads are processes
you cannot have a thread process separation and have emergent processes as info only stored on threads
alpine mulch
would something like storing it in the cpu local struct be suitable
threads usually have a notion of which process they are part of
surreal path
then i shall store a list of processes
js a linked list
of processes
what ill doo
broken depot
@rigid fable }}}}}}}}
broken depot
lets fucking go
nice
surreal path
idk if bash is suppsoed to not return to prompt or whatever
rigid fable
<@456226577798135808> }}}}}}}}
Gud
}
frigid cliff
yippee nyaux is back
tardy hull
welcom backkkkk
surreal path
im eepy
ebon needle
<:letsgo:1282155513156468747>
hi
coral dove
i am trying to figure out how to build but idk -_-
Build what
tardy hull
The kernel, I tried doing make but it didn't work and said that cc-runtime git something
surreal path
execve works without deallocate_all_user_regions() but does this if i do
void deallocate_all_user_regions(pagemap *target) {
VMMRegion *cur = (VMMRegion *)target->userhead;
while (cur != NULL) {
if (cur->nocopy) {
cur = (VMMRegion *)cur->next;
continue;
}
kprintf_vmmregion(cur);
arch_unmap_vmm_region(target, cur->base, cur->length);
slabfree(cur);
cur = (VMMRegion *)cur->next;
}
}
thats lit it
surreal path
heres where we are atm
execve is implemented and working
however ncurses is being a piss of poop
and not finding our terminfo even tho everything i provide to it is correct?
broken depot
and not finding our terminfo even tho everything i provide to it is correct?
It might expect certain environment variables
broken depot
Then i dunno
surreal path
broken depot
hey u awake
Not for very long
If you'd asked maybe 30 minutes earlier I could help debug but I'm literally just finishing up so I can go to bed.
broken depot
Good luck :3
surreal path
nyaux after trying to open bash after implementing stat mode correctly
my stack pointers fucked
???
rigid fable
my stack pointers fucked
Did you buy it dinner first? 💀
rigid fable
I'm sorry I had to. I won't do it again
rigid fable
😅😂
edgy pilot
isnt you was nyauxmaster 5.0
ebon needle
https://tenor.com/view/grammar-police-drift-police-car-gif-17312786
where !??!?!
edgy pilot
where !??!?!
I am the grammar police
ebon needle
I am the grammar police
and what you want to do with me ????
edgy pilot
and what you want to do with me ????
I'm going to put you in prison for life unless you fix your grammar
ebon needle
I'm going to put you in prison for life unless you fix your grammar
are you duolingo worker
edgy pilot
are you duolingo worker
Duolingo is my subordinate
ebon needle
Duolingo is **my** subordinate
is google translator your enemy
edgy pilot
no
edgy pilot
do you accept bribes???
the only bribe I accept is a grammatically correct sentence
kind root
still wrong after the edit btw
surreal path
in order to debug further
i need to figure out why is ncurses getting its shit stripped
so i can load it in
and view whats up
idk whos stripping this shit
but i need those debug symbols
i know it crashes in that file
cinder plinth
I'm not sure why you're debugging ncurses, this is like 99% a kernel or libc issue
surreal path
I'm not sure why you're debugging ncurses, this is like 99% a kernel or libc iss...
because i dont understand my kernel issue so i need to look into ncurses to figuyre out whats happening
cinder plinth
do you trace syscalls
surreal path
yes
cinder plinth
do that
surreal path
i didnt return an error code
it was valid
it stated the fiel
i gave the file stat
structure
and it closed
cinder plinth
and then it hangs?
surreal path
then it page faults on orq 0x0,(%rsp)
cinder plinth
so your stack is fucked?
surreal path
yep
cinder plinth
how big is it
brisk zenith
that looks like a stack probe
surreal path
brisk zenith
does rsp look correct? if so you probably just need to increase your userspace stack size
brisk zenith
I mean it's low but it doesn't look obviously wrong
maybe it's just at the bottom of its stack
brisk zenith
looks like it yeah
surreal path
korona said its plassuible
brisk zenith
userspace generally expects a stack region of like a few megabytes
kind root
8k is still nothing
brisk zenith
is rsp still near the bottom of the stack?
try giving it like 8 megabytes or something I think that's linux's default
kind root
does nyaux have on demand memory allocation
surreal path
no
supple robin
no
kind root
bruh
edgy pilot
give it a megabyte
or 2
supple robin
i give 2mib
cinder plinth
ah no way! I was right 
surreal path
heres whats next for nyaux
Nyaux Next Milestone: Run Doom
- support x2apic and ps2 driver
- implement more syscalls like access()
- posix signals
- devfb
- dev mouse
- dev keyboard
- profit???
hollow goblet
Nyaux PS/2 driver
cinder plinth
Nyaux Next Milestone: Run Doom
- support x2apic and ps2 driver
- implement more ...
you dont need signals or access to run doom
it's piss easy
just get keyboard input working
glossy walrus
low effort troll
tardy hull
Nyaux Next Milestone: Run Doom
- support x2apic and ps2 driver
- implement more ...
yippeeeeeee
surreal path
oh yeah @kind root i havent updated uacpi in a while lmao
i should do that
@edgy pilot any updates on the flanterm fork i should pull?
kind root
yeah there have been a few fixes since
surreal path
yeah there have been a few fixes since
yea i should pull and make sure its update to date
also btw
opinion on tic running?
in bash
infocmp also runs and same with zdump
but no signals
actually mostly everything runs expect for this one program but its to be expected cause no access()
kind root
tic?
edgy pilot
have you considered sleeping?
surreal path
i cant
surreal path
i cant
surreal path
idk
edgy pilot
it's easy: you get to a bed and put yourself in a horizontal position, optionally put a blanket over yourself
surreal path
cant
edgy pilot
president trump says: falsch
there are some things you still need to try and do, however pointless they must seem
While sleep is only for the weak, it's still important
not even sleeping, just resting
you might feel like you don't deserve it or you should feel worse or something, but think about it
is it really logical?
what would spock say?
🖖
In other news, it's 6 am here