yea @molten grotto recompiling bash works
#Nyaux
surreal path
ok come on bro
i think i need to recompile nano to
too
same issue on nano
@molten grotto
LOOK NANO WORKS
LETS FUCKING GOO
molten grotto
ionic musk
surreal path
3 weeks ago u worked on it
im considering switching back to mints branch atp
if ur not gonna mantain it
😡
flat nymph
Why was it forked in the first place?
surreal path
sixel support
flat nymph
(I'm still using 1 year old version and it works fine 
)
edgy pilot
fix ur flanterm fork
that's not much information about the issue you're having, is it?
silver yarrow
surreal path
that's not much information about the issue you're having, is it?
im not having an issue
rn
edgy pilot
what do you want fixed then mr
edgy pilot
no
surreal path
thats crazy
edgy pilot
as I said, it was just an experiment
surreal path
surreal path
https://www.reddit.com/r/osdev/comments/1julxdf/nyaux_runs_nano/ 100 upvotes pls thanks
median goblet
nice
kind root
<@379250891838193667>
Damn well done
surreal path
Damn well done
thanks bro#
surreal path
time to debug why nano is using canoical mode??
kind root
Just port vim instead
thorn bramble
time to debug why nano is using canoical mode??
that's normal
coral dove
https://www.reddit.com/r/osdev/comments/1julxdf/nyaux_runs_nano/ 100 upvotes pls...
Bro how the fuck did your shit get 7 upvotes but some idiot with nothing but a kernel toy shell that is 2 c files get 300.
surreal path
that's normal
how so
thorn bramble
instead of "debugging" why it uses a properly supported feature you should implement it
surreal path
instead of "debugging" *why* it uses a properly supported feature you should imp...
i did
tho
thorn bramble
good
thorn bramble
because you implemented it incorrectly probably
surreal path
and moves the cursor down
no but shouldnt nano be using raw mode
first off
and second off
thorn bramble
no idea lmao
surreal path
because you implemented it incorrectly probably
no not really? its literarly just a few lines of code
if (tty->termi.c_lflag & ICANON) {
for (; i < (int)size; i++) {
uint64_t val = 0;
get_ringbuf(tty->rx, (uint64_t *)&val);
((char *)buffer)[i] = (char)val;
if ((char)val == '\n') {
break;
}
}
return i;
}
its just this
coral dove
no but shouldnt nano be using raw mode
No, it should be
surreal path
otherwise i will do this
thorn bramble
i don't think that's correct brother
or Maybe it is
who knows
not me!
I love spreading misinformation :3
surreal path
bru
coral dove
No, it should be
Shit I’m stupid I misread your message
surreal path
Shit I’m stupid I misread your message
?
coral dove
?
You said “shouldn’t nano be in raw mode” and I read it as “nano shouldn’t be in raw mode”
flat nymph
Bro how the fuck did your shit get 7 upvotes but some idiot with nothing but a k...
It's like when Managarm systemd was anounced, a lot of people were asking "why"
coral dove
It's like when Managarm systemd was anounced, a lot of people were asking "why"
Those dumbasses probably don’t know the difference between ring 0 and ring 3
cinder plinth
Bro how the fuck did your shit get 7 upvotes but some idiot with nothing but a k...
Because it's been like an hour lol
surreal path
do i need fcnctl for nano input?
i might be wrong idk
im just reading nano source code and trying to make sense of it
flat nymph
Ironclad reference?!
surreal path
i cant even get raw mode or cannoical mode working
i dont fucking understand why giving it one character doesnt make it happy for raw mode
but wtv
edgy pilot
i cant even get raw mode or cannoical mode working
cooked
surreal path
what
edgy pilot
it's called cooked mode
surreal path
man page says cannoical
edgy pilot
man page is cringe
brisk zenith
posix says canonical as well
crystal scarab
Bro how the fuck did your shit get 7 upvotes but some idiot with nothing but a k...
i saw that, i hate that so much
kind root
Cooked mode and raw mode
Btw why is it called cooked, I still don't know to this day lol
desert haven
cooking transforms stuff
brisk zenith
the raw input and output is cooked into a nicer form
tender gorge
Btw why is it called cooked, I still don't know to this day lol
because the other mode is the raw mode
kind root
the raw input and output is cooked into a nicer form
Makes sense
bleak cedar
They really have been cooking with this rare w name choice
surreal path
i cant get cannoical or raw mode right for userspace to accept it
surreal path
possibly the only way im gonna understand how raw mode vs cannoical mode works is probaly gonna be me writing my own userspace program to test how exactly input works, then have this userspace program also in my userspace for nyaux and compare it to linux and adjust tty code til i get it as close as possible to the results of linux
tmrw i get my weight loss pills (Glucomannan) so i can control my appetie cause recently ive been out of control and its been a major stress point for me
so hopefully tmrw im not as stressed as today
(i ate 3000 calories of chocolate AGAIN)
coral dove
tmrw i get my weight loss pills (Glucomannan) so i can control my appetie cause ...
Okay man sounds good I hope you feel better
We love you man
Don’t ever rush the OS
surreal path
thanks
coral dove
you’re very talented and we care about your health brother
surreal path
thank you :)
thorny glen
possibly the only way im gonna understand how raw mode vs cannoical mode works i...
might be helpful
surreal path
ive read this
surreal path
L
😭
blame the article for not bothering on explaining implementation side of things
and just pointing me to fucking linux source code
yea thats fun
:^)
coral dove
and just pointing me to fucking linux source code
Which may or may not be readable depending on the mood torvalds was in when he allowed the commit to be pushed lol
surreal path
Which may or may not be readable depending on the mood torvalds was in when he a...
real
gilded mantle
possibly the only way im gonna understand how raw mode vs cannoical mode works i...
the good thing is that you don't need everything
just newline conversion, EOF, EOL, erase, echo, echoe and signals is enough for nearly everything
surreal path
the good thing is that you don't need everything
just newline conversion, EOF, E...
new line conversion?
gilded mantle
yeah 😭
surreal path
what is that?
gilded mantle
why exactly do i need to do that?
cuz in raw mode enter sends '\r' counterintuitively
surreal path
cuz in raw mode enter sends '\r' counterintuitively
and what sends \n in raw mode?
surreal path
since ctrl just strips off bits 5 and 6 of what the character would've been
never knew that but ok lol
brisk zenith
yeah i only figured it out when implementing my vt
surreal path
in raw mode does every character get printed by the program or by the vt itself?
brisk zenith
echo is actually independent of canon but yeah
surreal path
echo is actually independent of canon but yeah
that makes sense cause the quick test program i wrote still prints characters, even in non canon mode
surreal path
i see
surreal path
wtf?
flat nymph
turn off optimisations and lto if you have them
surreal path
i dissabled optimiztiations and its still doing this
flat nymph
(at some point my kernel wasn't working without optimizations)
edgy pilot
lmao
surreal path
im gonna cry in a corner
surreal path
did you push this?
on github?
flat nymph
yeah
surreal path
@flat nymph sorry for leaving
edgy pilot
me when base 10:
thorn bramble
surreal path
i fell asleep
tender gorge
Hi Rayan 🫂 💙
edgy pilot
i fell asleep
your punishment for that insolence will be more nyaux work!
how could one even sleep, unimaginable. 
surreal path
interesting
this issue seems more complex then the getcwd problem
time to put printfs all over nano source code
surreal path
what
i know im joking
(thought it was obvious with the rage face lmao)
bagel ur morning
median goblet
also have i never spoken before in this thread wtf
broken depot
And then the whole diagram of a pipe thing happened 
flat nymph
Wait, how do you search within a thread?
surreal path
infitely allocating here
median goblet
Wait, how do you search within a thread?
I didn't
I had to find messages in the progress report thread channel which shows messages from all threads
flat nymph
infitely allocating here
while (true) {} is UB?
velvet fog
while (true) {} is UB?
Yeah but no compiler would actually optimize it out when it's so explicit, that would break a lot of programs
median goblet
while (true) {} is UB?
yes
i think the program that breaks expectations looks like this
#include<stdio.h>
int main() {
while(1);
}
void what() {
printf("what???");
}
nano is infintely crying here
infitely allocating here it seems
which makes sense?
senes*
senes
senses*
BRO I CANNOT SPELL 😭
back to the ncurses dungeons
molten grotto
yes
not since c11 afaik and there was a defect report in c++ to make it non-ub as well (though clang was the only compiler that fully optimized the loop away and caused weird stuff, since like clang 18 or smth it doesn't do that anymore to comply)
median goblet
i think the program that breaks expectations looks like this
```c
#include<stdio...
i swear to fucking god if i am being gaslighted
surreal path
i swear to fucking god if i am being gaslighted
gaslighted by the c compiler
richard stallman laughing with his beard
edgy pilot
i swear to fucking god if i am being gaslighted
?
it was a "bug" in clang that has been fixed recently
edgy pilot
it used to be
edgy pilot
in C++
surreal path
back to trying to understand ncurses code for like 50 hours lmao
#ihategnustylecode
edgy pilot
I believe it's still UB in C
surreal path
#gnudevelopersshouldbepolitelysenttocastledungeons
edgy pilot
I remember reading a proposal for C++ to define that loop
surreal path
its either gnu code is too big brain for me to understand
or its too shit for me to read
surreal path
or its too shit for me to read
can someone tell me why all gnu software code is like this
genuinely
i should make a gnu code style bingo card
- shit indentation
surreal path
- macros everywhere
molten grotto
I believe it's still UB in C
I don't think its ub after c11
edgy pilot
I don't think its ub after c11
The forward progress guarantees were added to C++11 and C11 along with a memory model.
true
it was a C++ proposal and I assumed it only applied to C++ lol
molten grotto
that proposal also says C does not make iteration statements whose controlling expression is a constant expression Undefined Behavior, whereas C++ does. so I don't think it was a thing the compiler was allowed to do before either
but anyway
surreal path
does ncurses expects read to return an error of some kind?
tw event?
tf
tf is key min
yea i know this is whats making it infinitely allocate
it thinks it doesnt have enough space for some fucking reason
idfk how to tell nano to stop
or rather ncurses to return ERR here
i hate ncurses i hate ncurses i hate ncurses
@broken depot u must agree
broken depot
ncurses suck to deal with
surreal path
ncurses suck to deal with
yes it does
and i have no idea why ncurses is not happy
broken depot
arbitrary reasons
surreal path
its not fun reading ncurses code
its not fun debugging ncurses
its not fun ncurses in general
why do we use ncurses
why cant programs just fucking implement their own shit for the tty
it isnt that hard, oooooo tcsetattr() so scary
surreal path
<@379250891838193667>
@unkempt relic i know u havent been in this thread for a long time but i wanted to show u this cause of the good progress we've been making on nyaux
also i hope keyronex is going well
let me peep into the thread actually
unkempt relic
<@391791686881247234> i know u havent been in this thread for a long time but i ...
i've been watching with considerable interest
nyaux is progressing remarkably well
surreal path
taking a break today cause im too frustrated with ncurses
as per usual
now i shall sleep
surreal path
nyaux is progressing remarkably well
thank you!
broken depot
Yeah Nyaux progress recently is absolutely legendary
kind root
Yes
When I start my os i probably won't reach nyaux levels until like a year into the project or even later 
surreal path
https://tenor.com/view/yippee-happy-celebration-joy-confetti-gif-25557730
me when i see chocolate egg
surreal path
When I start my os i probably won't reach nyaux levels until like a year into th...
is oki u will get there 
also opinion on chocolate egg
edgy pilot
When I start my os i probably won't reach nyaux levels until like a year into th...
make uacpi users to sign a contract requiring them to contribute to your os as a payment
surreal path
i feel so poop today :c
surreal path
okay so
this condition never happens
oh yeah
now that we are past userspace
i should make a wishlist for any contributors looking for work 🥺
what i really want rn is someone to at least EXPLAIN what the code does so people understand what they are looking at instead of going wtf is that in there mind
so basically comment everywhere in code lol
documentation basically
we also have some weird amd bug @tender gorge reported on
who needs to make a github issue 🥺
🥺 🙏
surreal path
what i really want rn is someone to at least EXPLAIN what the code does so peopl...
maybe fix my non existent clang-format file so everything looks readable at least
idfk i should probs do that myself but its not main priorty atm
and im not expecting any contributors
to be fair 😭 💀
also i said id be taking on break on nyaux for today cause ncurses, well turns out i lied lol
surreal path
okay so
kind root
make uacpi users to sign a contract requiring them to contribute to your os as a...
real
surreal path
okay so
we need to figure out why input doesnt return ERR
so to do that, we need to determine how ERR is decided to be returned by ncurses
this is the source of our problem
this is never true @molten grotto is this bad? because nano depends on ncurses to return error on wgetch() for it to stop reading keycodes
molten grotto
maybe, idk
surreal path
maybe, idk
surreal path
maybe, idk
ncurses gets 0 in raw mode, after reading the tty which is in raw mode
is this not allowed
does this break abi
okay i see
tender gorge
who needs to make a github issue 🥺
i keep forgoring 🥺
surreal path
i keep forgoring 🥺
is oki 🥺
flat nymph
Can I rewrite your context switching code?
surreal path
Can I rewrite your context switching code? <:trl:829316641191886888>
go ahead
as long as it doesnt break stuff
flat nymph
Ok, I'll look into it tomorrow
surreal path
issue now with bash :) it doesnt print the character until u press backspace
static size_t rw(struct vnode *curvnode, void *data, size_t offset, size_t size,
void *buffer, int rw) {
if (rw == 0) {
struct tty *tty = data;
// TODO check non blocking
if (tty->termi.c_lflag & ICANON) {
int i = 0;
for (; i < (int)size; i++) {
restart:
uint64_t val = 0;
spinlock_lock(&tty->rxlock);
int res = get_ringbuf(tty->rx, (uint64_t *)&val);
if (res == 0) {
spinlock_unlock(&tty->rxlock);
sched_yield();
goto restart;
}
((char *)buffer)[i] = (char)val;
spinlock_unlock(&tty->rxlock);
if ((char)val == '\n') {
break;
}
}
if (tty->termi.c_lflag & ECHO) {
flanterm_write(get_fctx(), buffer, size);
}
return i;
}
// return a character
int ok = size < VMIN ? size : VMIN;
int i = 0;
for (; i < (int)ok; i++) {
restart1:
uint64_t val = 0;
spinlock_lock(&tty->rxlock);
int res = get_ringbuf(tty->rx, (uint64_t *)&val);
if (res == 0) {
spinlock_unlock(&tty->rxlock);
sched_yield();
goto restart1;
}
((char *)buffer)[i] = (char)val;
spinlock_unlock(&tty->rxlock);
}
if (tty->termi.c_lflag & ECHO) {
flanterm_write(get_fctx(), buffer, size);
}
return i;
} else {
// MOST BASIC AH TTY
flanterm_write(get_fctx(), buffer, size);
sprintf_write(buffer, size);
return offset + size;
}
}
source
@elder shoal is this something to do with fflush()
?
surreal path
```C
static size_t rw(struct vnode *curvnode, void *data, size_t offset, size_t ...
page fault when pressing a keycode in nano but one problem at a time
elder shoal
<@215823874615541761> is this something to do with fflush()
fflush is libc stuff
surreal path
fflush is libc stuff
okay
flat nymph
issue now with bash :) it doesnt print the character until u press backspace
I think it should print immediately if you're writing to terminal, and otherwise untill you press the backspace?
surreal path
I think it should print immediately if you're writing to terminal, and otherwise...
what
surreal path
i will work on nyaux maybe tmrw idk i feel like poo
surreal path
nvm
ill work on it
now no 1
why isnt bash printing my fucking characters
@molten grotto what would be the first place to check if bash is deciding not to print my characters but only if i backspace them
molten grotto
what did you change that made it not work
surreal path
what did you change that made it not work
made read blocking
int ok = size < VMIN ? size : VMIN;
int i = 0;
for (; i < (int)ok; i++) {
restart1:
uint64_t val = 0;
spinlock_lock(&tty->rxlock);
int res = get_ringbuf(tty->rx, (uint64_t *)&val);
if (res == 0) {
spinlock_unlock(&tty->rxlock);
sched_yield();
goto restart1;
}
((char *)buffer)[i] = (char)val;
spinlock_unlock(&tty->rxlock);
}
if (tty->termi.c_lflag & ECHO) {
flanterm_write(get_fctx(), buffer, size);
}
return i;
as u can see
this is if im in raw mode
which i am
with bash
@molten grotto
nvm i know the issue
molten grotto
I don't think bash uses raw mode
surreal path
I don't think bash uses raw mode
it does
molten grotto
actually wait yeah, I were thinking of the wrong thing lol
surreal path
actually wait yeah, I were thinking of the wrong thing lol
should i keep of list open file descriptors to said vnode inside a vnode or just add another arugment to my read() for fd options
?
molten grotto
for what flags
surreal path
fd flags
molten grotto
like non blocking etc?
surreal path
like non blocking etc?
yea
molten grotto
I'd probably just have an argument for them in the op functions
because open files aren't really tightly coupled to vnodes
surreal path
I'd probably just have an argument for them in the op functions
this thing is so big
molten grotto
you could move the options to a struct
surreal path
idk
rigid fable
molten grotto
btw I were looking into making mlibc glibc compat work with newer mlibc and it doesn't look good 💀 the major pain is symbol versioning that got implemented in mlibc and it means that apps linked against glibc no longer even get linked by the rtld because of mlibc not having the same versioned symbols as glibc
surreal path
btw I were looking into making mlibc glibc compat work with newer mlibc and it d...
no
dies
molten grotto
I think the easiest solution to this is to write a tool that strips versioning information from the binary linked against glibc but the annoying thing is that then I can't test this stuff on linux at all because of all the other libs and whatever linked against glibc also having versioning info (and I don't want to start modifying system libraries)
molten grotto
for the tty bash does supply the nonblocking flag when opening, it's also shown in the image that you sent where the flags in decimal are 2050
surreal path
for the tty bash does supply the nonblocking flag when opening, it's also shown ...
no it doesnt
it only supplys 2050
molten grotto
and that includes the flag
surreal path
how
surreal path
the flag in the header is in octal
its in octal????
molten grotto
yes its what the zero prefix does
surreal path
yes its what the zero prefix does
but i & it with the file descriptor flags
and it isnt non blocking cause it didnt go through
molten grotto
when I tried it it went inside the canon if and didn't enter the blocking loop
surreal path
doesnt print
surreal path
when I tried it it went inside the canon if and didn't enter the blocking loop
what
thats weird af
molten grotto
I can try again after I get back home in a bit but iirc
surreal path
I can try again after I get back home in a bit but iirc
please do when u get home
cause im hella lost as to why this isnt working lol
surreal path
@molten grotto u home?
molten grotto
yes
surreal path
yes
alright, try and see what bash is doing on ur machine
im sure its the same result
molten grotto
ah I didn't have rebuilt mlibc, after your dup2 impl it looks like that thing happens where bash doesn't display the chars
surreal path
ah I didn't have rebuilt mlibc, after your dup2 impl it looks like that thing ha...
yea :c
i dont know why :c
try to have a look if u can cause im stuck
surreal path
oh yeah @molten grotto returning enosys in dup2 doesnt fix ze issue either
thats interesting flags is 0
yea weird
surreal path
:c
@molten grotto so i know that stdin flags are 0 because i always setup stdin per process, and i initally set stdin to flags 0 but why doesnt bash do something like fcntl the fd so its not blocking. does it expect to be non blocking on boot up??
is that what bash expects as the default?
molten grotto
I think I found the problem
surreal path
What is it
molten grotto
the termios c_cc has the fields VTIME and VMIN, there are various cases that do different things but if both of those are zero then you are supposed to do a non blocking read
surreal path
Where in the man page does it say that?
molten grotto
surreal path
I see
surreal path
where is this btw
molten grotto
surreal path
https://www.man7.org/linux/man-pages/man3/termios.3.html
im gonna add the check and see if this solves it
yea that solved the bash issue
time to test nano
again
surreal path
https://www.man7.org/linux/man-pages/man3/termios.3.html
nano is still expecting read to be blocking (or well ncurses) even tho vmin is 0 and vtime is 0 and flags is 0???
fucking hell ncurses
wtf ncurses like bro
what do u want
broken depot
Does ncurses actually want vmin and vtime to be 0?
surreal path
Does ncurses actually want vmin and vtime to be 0?
idk its set as 0
broken depot
That's what I'd figure out next
broken depot
something to look into ¯_(ツ)_/¯
surreal path
if i set vtime to 5 initially bash is no longer happy and does the same bug again
whats up with that?
molten grotto
idk tbh
surreal path
im cooked
ill ask korona or smt maybe when hes back in vc about this cause im just
surreal path
ill do further debugging in a bit as to WHY ncurses expects a blocking read
apologises for the slow nyaux development
bug after bug really gets annoying
especially when ur trying to port nano
which is a bitch to port
especially with tty bugs which are the most annoying ill admit
id rather fucking debugging rip 0x0 bugs then this shit
im not gonna have fun implementing signals after this
thats for sure
i contributed to mlibc today so that was nice
https://github.com/managarm/mlibc/pull/1348 my pr btw :3
first mlibc pr that isnt sysdeps
@alpine mulch ur next
ur ec driver is next :)
im gonna watch some ada tutorials or smthin
or just write a basici userspace ada program
idfk
also i should update uacpi
i hope the api hasnt changed
surreal path
i hope the api hasnt changed
@kind root me being absolutely clueness fr
(i havent updated in months)
edgy pilot
api has been stable after 1.0
surreal path
api has been stable after 1.0
really now?
edgy pilot
I know, crazy
surreal path
crazy indeed
surreal path
okay i see
flat nymph
yeah api is stable now
have there been any news about ROS and Haiku?
surreal path
i am the stupidest
stupidest
bro
i was returning read 1 byte even tho i read 0 bytes
💀
surreal path
yea bash exits from that, like bro what am i supposed to say
surreal path
fuck i fell asleep
tender gorge
The guy from haiku just disappeared so no idea whats happening there
Happens sometimes over there
kind root
surreal path
i dont fucking know how this read keys from function in nano works
:c
atp i dont know how to go forward
lemme try to sleep now
so i can be awake when qwinci is online
and actually ask
tender gorge
lemme try to sleep now
Goodnight 💙
surreal path
i fell asleep again :(
broken depot
Wasn't that your goal?
broken depot
Something about differing timezones
surreal path
im sorry im having really bad days atm
no development or debugging until tmre
even tmrw is a maybe
i had an awful day.
plush hearth
Take your time. Don't force yourself to work on this if you don't want to. Also, I hope you feel better soon.
edgy pilot
@surreal path how do you even make deltarune last that long
content runs out
or do you just play again but different paths
surreal path
or do you just play again but different paths
stuck on spamton neo
surreal path
Take your time. Don't force yourself to work on this if you don't want to. Also,...
thanks
i might work today
or tmrw depending
on my mood
surreal path
im ready to do nyaux development today
lets start by asking ourselves how we can solve this issue
to which my answer is no idea
@molten grotto have u reached a point where u ported ncurses before
like at all
molten grotto
not really
surreal path
not really
molten grotto
do you still have the same weird tty problem
surreal path
yes
molten grotto
ttys kinda suck ngl
surreal path
if vtime is set on tty init to 5, bash isnt happy and does that character bug again. if vtime is 0, ncurses isnt happy and nano infinitely allocates a keyboard buffer because wgetch never returns error cause ncurses was expecting a blocking read
even if i default everything to a blocking read it still dies
and i have no fucking idea why its doing this
i read the openat flags and i set them and i check them
and i do the weird ass thingy if vtime and vmin are 0
WHICH THEY ARE
and ncurses STILL expects a blocking read like bro
surreal path
i read the openat flags and i set them and i check them
and btw @molten grotto ncurses never calls fcntl or that shit so it never even tries to change the flags of the fd so i dont know where its pulling out its ass that its expecting a blocking read*
vtime and vmin are still 0 so read does polling and ncurses wants blocking
yea i remember
surreal path
vtime and vmin are still 0 so read does polling and ncurses wants blocking
even tho it never changes it
edgy pilot
I think I had ncurses
and bash worked fine
I don't know about nano
you can check the tty implementation on the old branch on my repo
surreal path
yea not until u actually RUN ncurses, which means running nano
bash is compiled with ncurses
surreal path
you can check the tty implementation on the old branch on my repo
thats not gonna solve it
edgy pilot
bash is compiled with ncurses
yes and it uses ncurses dummy
edgy pilot
thats not gonna solve it
you don't know that
edgy pilot
it doesnt call wgetch
that's not the whole of ncurses
surreal path
you don't know that
ill look at it but i doubt anything ur doing im not doing
surreal path
that's not the whole of ncurses
that where it dies in ncurses tho
edgy pilot
ill look at it but i doubt anything ur doing im not doing
hmm I remembered this file differently
surreal path
i dont do that await shit but everything else i am doing
i dont understand the code very well tho as i dont read c++ so idk
either way i dont see anything helpful here
edgy pilot
i dont do that await shit but everything else i am doing
so you don't actually use vtime?
surreal path
no
edgy pilot
maybe fix that first
surreal path
vtime is always 0
its never changed by bash OR ncurses
and typically that should be fine
edgy pilot
is ioctl with tcsets never called?
brisk zenith
my bootloader ignores vtime and it runs bash and nano with ncurses just fine
surreal path
it is, but VTIME isnt changed
brisk zenith
here's the tty code if you want to take a look at it https://github.com/monkuous/andromeda/blob/main/sources/andromeda-kernel/kernel/drv/console/console.c
code style is a bit unique because it's a cpu-local kernel stack so blocking uses continuations but still
surreal path
how come eagain doesnt kill bash or ncurses for you
it kills bash and ncurses for me
brisk zenith
that's the unique code style I was talking about
it returns eagain unconditionally, but if it's blocking the thread switches away before returning to userspace
and the continuation sets the actual return value
edgy pilot
that's clever
surreal path
yea i dont really see anything too obvious
other then vmin being set to 1 and u doing eagain
which is probs never returned
brisk zenith
are you sure ncurses actually wants the non-blocking read? it never sets vmin to 0 for me
what's your default value for vmin?
surreal path
what's your default value for vmin?
5
brisk zenith
wait, why 5?
brisk zenith
i just did tcgetattr from one of my host's vts and copied the values returned
brisk zenith
looks like managarm just sets it to the values from sys/ttydefaults.h by default
surreal path
looks like managarm just sets it to the values from sys/ttydefaults.h by default
yea i believe cmin is 5
brisk zenith
maybe it's this? https://github.com/rayanmargham/NyauxKC/blob/3f151fb6b875b801b250220acf912a30ed712121/kernel/src/dev/tty/tty.c#L26
you're comparing the size to the index of vmin, not its value
brisk zenith
when i do make run-kvm i get this, do you know a workaround? ```
syscall_mmap(): size 524288 flags 20
General Protection Fault with error code 0x38
CR2 0x240028
-> Function: isr_stub_14.skipswapgs2() -- 0xffffffff800bfd31
-> Function: Unknown() -- 0x0
-> Function: Unknown() -- 0x0
NYAUX Panic! Reason:
fuck
oopsie uwu :3```
oh wait tcg works fine
surreal path
when i do `make run-kvm` i get this, do you know a workaround? ```
syscall_mmap(...
that is very weird??????????????????????
huh??????????????????????????????????????????????????????????????????????????????????????????????
are u running amd
brisk zenith
yeah
surreal path
i think this might be the amd bug @tender gorge mentioned
:c so ur gonana have to sport tcg for now
brisk zenith
that's fine
surreal path
sorry i cannot debug that since im not on amd
brisk zenith
i am very unfamiliar with jinx, what's the command to rebuild a package
surreal path
i am very unfamiliar with jinx, what's the command to rebuild a package
./jinx regenerate package && ./jinx rebuild package
brisk zenith
ah ty
surreal path
np
surreal path
@brisk zenith did u find anything
brisk zenith
still looking
surreal path
alright
brisk zenith
rn i'm investigating why vmin and vtime get set to 0
because i don't observe that behavior on other systems
surreal path
yes its quite weird
i was confused by that too
that really shouldnt been happening lol
thanks for helping btw
means a lot, especially with difficult issues like these
if theres any way i can return the favour i will do so
brisk zenith
right well i can say that readline at least tries to set vmin=1
still trying to find where it gets fucked
surreal path
brisk zenith
this is my favorite kind of debugging
surreal path
this is my favorite kind of debugging
lol
brisk zenith
rn i'm investigating why vmin and vtime get set to 0
figured it out ```patch
diff --git a/kernel/src/dev/tty/tty.h b/kernel/src/dev/tty/tty.h
index 189857c..604d62b 100644
--- a/kernel/src/dev/tty/tty.h
+++ b/kernel/src/dev/tty/tty.h
@@ -121,6 +121,7 @@ struct termios {
tcflag_t c_oflag; /* output modes /
tcflag_t c_cflag; / control modes /
tcflag_t c_lflag; / local modes */
- cc_t c_line; /* don't know what this is but mlibc has it /
cc_t c_cc[NCCS]; / special characters */
speed_t ibaud; // input baud rate
speed_t obaud; // output baud rate```
kernel's termios definition didn't match libc's
brisk zenith
no clue but it's there in the abi-bits struct
surreal path
figured it out ```patch
diff --git a/kernel/src/dev/tty/tty.h b/kernel/src/dev/t...
is nano running now???
surreal path
figured it out ```patch
diff --git a/kernel/src/dev/tty/tty.h b/kernel/src/dev/t...
what is the cc_t struct?
brisk zenith
cc_t is just a typedef for unsigned char
so effectively whenever the kernel read from termi.c_cc[x] it got the value termi.c_cc[x - 1]
surreal path
so effectively whenever the kernel read from `termi.c_cc[x]` it got the value `t...
i see
brisk zenith
oh btw this is also necessary for the vmin/vtime behavior to work correctly ```patch
+++ b/kernel/src/dev/tty/tty.c
@@ -33,8 +33,8 @@ static size_t rw(struct vnode *curvnode, void *data, size_t offset, size_t size,
int res = get_ringbuf(tty->rx, &val);
if (res == 0 &&
((hnd->flags & O_NONBLOCK) ||
-
(tty->termi.c_cc[VMIN] == 0 || tty->termi.c_cc[VTIME] == 0)) == false) {
-
(tty->termi.c_cc[VMIN] == 0 && tty->termi.c_cc[VTIME] == 0)) == false) { kprintf("blocking\r\n"); spinlock_unlock(&tty->rxlock); sched_yield(); goto restart1;```
surreal path
could u give the patch header too
brisk zenith
one second i'm trying to get rid of all the extra newlines added to files i didn't need to change
brisk zenith
this is the full git diff output ```patch
diff --git a/kernel/src/dev/tty/tty.c b/kernel/src/dev/tty/tty.c
index e7a8029..f68617e 100644
--- a/kernel/src/dev/tty/tty.c
+++ b/kernel/src/dev/tty/tty.c
@@ -33,8 +33,8 @@ static size_t rw(struct vnode *curvnode, void *data, size_t offset, size_t size,
int res = get_ringbuf(tty->rx, &val);
if (res == 0 &&
((hnd->flags & O_NONBLOCK) ||
-
(tty->termi.c_cc[VMIN] == 0 || tty->termi.c_cc[VTIME] == 0)) == false) { -
kprintf("blocking\r\n");
-
(tty->termi.c_cc[VMIN] == 0 && tty->termi.c_cc[VTIME] == 0)) == false) { -
//kprintf("blocking\r\n"); spinlock_unlock(&tty->rxlock); sched_yield(); goto restart1;
diff --git a/kernel/src/dev/tty/tty.h b/kernel/src/dev/tty/tty.h
index 189857c..1538213 100644
--- a/kernel/src/dev/tty/tty.h
+++ b/kernel/src/dev/tty/tty.h
@@ -121,6 +121,7 @@ struct termios {
tcflag_t c_oflag; /* output modes /
tcflag_t c_cflag; / control modes /
tcflag_t c_lflag; / local modes */
- cc_t c_line; /* don't know what this is but mlibc has it /
cc_t c_cc[NCCS]; / special characters */
speed_t ibaud; // input baud rate
speed_t obaud; // output baud rate
yes the unknown opcode is reproducable on my system
so that is another issue to look into
could be related to the amd bug but idk
surreal path
gut feeling?
brisk zenith
yeah
surreal path
fair
this kernel is all but gut feelings and hopes and dreams lol
i see this unknown opcode occurs when PRESSING a key
so it could be related to keyboard handling logic in nano/ncurses
surreal path
yea
brisk zenith
i thought it was just unconditional
tawdry mirage
c_line is the line discipline
The structure termios contains the c_line (line discipline) element, which is not used in POSIX compliant systems.
https://tldp.org/HOWTO/pdf/Serial-Programming-HOWTO.pdf
surreal path
> The structure termios contains the c_line (line discipline) element, which is ...
NOT USED
LITERARLY NOT USED
insane
brisk zenith
@surreal path on the #UD: when the read syscall unblocks and returns to userspace, it seems to corrupt the user stack
surreal path
<@787341689333350450> on the #UD: when the `read` syscall unblocks and returns t...
very weird??
how trf
tf*
brisk zenith
it's not what's causing the stack corruption and #UD but this bug results in any user page fault succeeding, even if it's not meant to be mapped ```patch
diff --git a/kernel/src/mem/vmm.c b/kernel/src/mem/vmm.c
index 3a4fca9..ab54472 100644
--- a/kernel/src/mem/vmm.c
+++ b/kernel/src/mem/vmm.c
@@ -141,6 +141,7 @@ bool iswithinvmmregion(pagemap *map, uint64_t virt) {
}
cur = (VMMRegion *)cur->next;
}
- return false;
}
void *uvmm_region_alloc_demend_paged(pagemap *map, uint64_t amount) {
assert(ker_map.userhead != NULL);```
surreal path
it's not what's causing the stack corruption and #UD but this bug results in any...
good fine
find*
brisk zenith
which delayed the exception until rip collided with a page that had actual data
surreal path
very weird tho
could it be related to elf.c?
i heard leo say before it was gonna cause issues down the line
the way i load elfs
i dont know why though, dont remember what he said
but might be good to investigate elf.c
brisk zenith
trying to figure out if i can use watchpoints to track down the moment the stack gets corrupted first but i'll take a look
surreal path
hold on could u try allocating everything with backing pages, might help to see
maybe theres different behavior
oh yeah and bash bug with chars not appearing until u press backspace is back weirdly
idk why tho
might just be a quirk on my machine ™️
brisk zenith
i get that as well
surreal path
weird
brisk zenith
ah it's not user stack corruption, it's register corruption
rsp gets changed from 0x1003e8 to 0x100a38
brisk zenith
syscall_waitpid does cpu->cur_thread->syscall_user_sp = cpu->arch_data.syscall_stack_ptr_tmp;, which is missing from all other syscalls - is this required for syscalls to block?
surreal path
`syscall_waitpid` does `cpu->cur_thread->syscall_user_sp = cpu->arch_data.syscal...
yes
brisk zenith
ah that's what's happening then
read blocks, so bash gets scheduled, which does waitpid and changes the value that syscall_entry loads into rsp
syscall_read upon returning doesn't restore it
i think it'd be less error prone if the scheduler did this save/restore
brisk zenith
yeah i'll add it to syscall_read and see if it fixes it
brisk zenith
well the panic is gone but the chars don't get displayed and it occasionally prints cannot stop wont stop
surreal path
well the panic is gone but the chars don't get displayed and it occasionally pri...
if u check the patch of nano
youll see where this appears from
it seems to be stuck with the keyboard buffer
brisk zenith
ah
surreal path
and reallocating to infinitely
brisk zenith
well no it's not like every x seconds it prints it, it's every x keypresses
so i assume it just doesn't pop from the buffer for some reason
surreal path
u can look at the source in nano and patch it accordinly to investigate
sources/nano-workdir
make sure to run ./jinx regenerate nano
which will regenerate the patch
brisk zenith
i have a gut feeling this is related to the bash prompt issue
surreal path
same
brisk zenith
anyway it's like 11 pm where i am so i'm going to sleep
here's the patch that fixes the stack issue ```patch
diff --git a/kernel/src/arch/x86_64/syscalls/syscall.c b/kernel/src/arch/x86_64/syscalls/syscall.c
index a01140b..c5df7c9 100644
--- a/kernel/src/arch/x86_64/syscalls/syscall.c
+++ b/kernel/src/arch/x86_64/syscalls/syscall.c
@@ -99,6 +99,8 @@ struct __syscall_ret syscall_read(int fd, void *buf, size_t count) {
if (hnd->node == NULL) {
return (struct __syscall_ret){.ret = -1, .errno = EIO};
}
- struct per_cpu_data *cpu = arch_get_per_cpu_data();
- cpu->cur_thread->syscall_user_sp = cpu->arch_data.syscall_stack_ptr_tmp;
if (count > (hnd->node->stat.size - hnd->offset) &&
hnd->node->stat.size != 0 && hnd->node->v_type != VCHRDEVICE) {
count = hnd->node->stat.size - hnd->offset;
@@ -110,6 +112,7 @@ struct __syscall_ret syscall_read(int fd, void *buf, size_t count) {
return (struct __syscall_ret){.ret = -1, .errno = res};
}
hnd->offset += bytes_read; - cpu->arch_data.syscall_stack_ptr_tmp = cpu->cur_thread->syscall_user_sp;
return (struct __syscall_ret){.ret = bytes_read, .errno = 0};
}
struct __syscall_ret syscall_close(int fd) {```
surreal path
thanks
ill investigate the bash and nano keyboard buffer issue
thank you for your help
surreal path
again, if theres anything i can do to repay you. like maybe contribute to andromada or somt lmk
i already have a bucket list of projects to contribute to lol
ive contributed to mlibc
and im doing ironclad next
i may not be a great programmer but hopefully i can provide help or smt, also same goes for u @molten grotto if u ever want me to contribute something :)
brisk zenith
it's fine, debugging this was pretty fun imo, you don't have to repay me or anything
surreal path
i wouldnt call that fun but everyone has their own versions of pain
brisk zenith
plus most of my own projects are the kinda thing I wouldn't accept any contributions to, I do them because I want to write everything myself
brisk zenith
ty
surreal path
np
tender gorge
this girl kinda just forced her on me
If you don't like it/don't consent then tell them that
surreal path
If you don't like it/don't consent then tell them that
i feel kinda numb so idk
tender gorge
It's not worth it risking being raped
tender gorge
i feel kinda numb so idk
It's a thing you have to decide
surreal path
It's a thing you have to decide
im fine with it i think?
edgy pilot
I am too much of a nerd
surreal path
be skinny and short
surreal path
Talk with her about it
like she was my previous ex
tender gorge
Don't be worried about losing her by talking with her, if you raising concerns makes her leave, it wasn't going to work anyway
tender gorge
like she was my previous ex
I would be SUPER careful
surreal path
ill talk abt it with her tmrw dw
tender gorge
The amount of red flags here genuinely concerns me 🥺
surreal path
The amount of red flags here genuinely concerns me 🥺
ill take care dw 🥺
tender gorge
🥺 💙
orchid dawn
be skinny and short
this one doesn't work for me
can't reproduce that behavior
- 95lbs
- 5ft 2"
- no bitches
surreal path
i feel so sick
tender gorge
i feel so sick
🫂 💙
broken depot
i feel so sick
tender gorge
🫂 💙
frigid cliff
❤️
ebon needle
surreal path
being sick is not absolute cinema :(
edgy pilot
relative cinema
worn rain
While I do find these jokes hilarious maybe consider making fun of him elsewhere considering it's a serious matter 💀
surreal path
It’s fine I know they are being light hearted about it
ebon needle
@surreal path i saw you on the vinegar server, can you show me your places please 🥺
surreal path
fix ncurses memes
kind root
nice
median goblet
@surreal pathquestion
median goblet
have you ever dealt with a bug where the PTE is valid accessible and can be executed from yet for some whatever reason it doesnt want to execute
and just keeps page faulting
cause i am
and i have no idea why it would keep PFing
surreal path
not really sorry :c
median goblet
rip
ebon needle
rayan_thevrman
I'll send to you friend request
surreal path
so im ready to work on nyaux now
nvm i got worse
ebon needle
nvm i got worse
😔
rich trellis
broken depot
Get well soon (threateningly)
surreal path
i know why the issue happens
i did this by just reading logs i didnt write shit cause im still sick
old
so basically
thats in decimal
and fd 0 and fd 1 and fd 2 flags are all set to 0
they arent changed by the openat syscall
causing this issue
no i do
nvm
if tty is opened with these flags does it expect stdin and stdout to have those flags too?
cinder plinth
so basically
ayu dark mention 🗣️ 🗣️ WTF IS A BAD THEME?
surreal path
if tty is opened with these flags does it expect stdin and stdout to have those ...
like how should this even work
brisk zenith
like how should this even work
most of the flags are for the file description, so all file descriptors to it share the same flags
surreal path
most of the flags are for the file description, so all file descriptors to it sh...
atp how am i supposed to attach fd 0 1 and 2
getting it to follow the flags
brisk zenith
taking a quick look at the code it seems you already do, through the dummy and realhnd fields
surreal path
taking a quick look at the code it seems you already do, through the dummy and r...
not really?
brisk zenith
like, fd 1 and 2 should be dup'd from fd 0, which should be open'd from the tty device
so fd 1 and 2 point to the same handle and same flags as fd 0 through the realhnd thing
surreal path
yes but fd 0 flags are zero
brisk zenith
oh right
brisk zenith
right so on process init fds 0..2 are allocated but don't point to any vnode
surreal path
yep
brisk zenith
idk what assumptions bash makes about fd state but most kernels open the tty for the init process and makes fd 0..2 point to that file description
surreal path
idk what assumptions bash makes about fd state but most kernels open the tty for...
i can make an init process that does this
brisk zenith
https://github.com/monkuous/andromeda/blob/5e06d74ea49295333386191dab2db77dc7a062d5/sources/andromeda-kernel/kernel/drv/console/console.c#L545 for reference this is where Andromeda does that
allocassoc is kinda like dup2
edgy pilot
files can be opened
surreal path
files can be opened
by that it could mean a creation of a new file descriptor
pointing to a vnode
edgy pilot
yes
surreal path
but can i actually have a vnode->open() op as well
edgy pilot
that would just return a file descriptor
edgy pilot
vfs designs have left the chat
surreal path
how does ur kernel setup fd 0 1 and 2
and its flags
for example
say /bin/bash opens /dev/tty with flags
do u just copy them over to the stdin stdout and stderr
how does that even work
@edgy pilot ?
everyone goes offline
:c
edgy pilot
how does ur kernel setup fd 0 1 and 2
current one doesn't yet
old branch did
honestly, I don't remember how things worked
it's been years
kind root
how does that even work
Bash does itself
With dup2 etc
You can assign any fd to any other fd
surreal path
right so on process init fds 0..2 are allocated but don't point to any vnode
no this is not true, on devtty_init
it is associtated
the problem is /dev/tty is opened with flags
and those flags arent reflecting on 0..2
kind root
Which flags tho
brisk zenith
openat returns a new file description, existing file descriptions (like the one that fds 0..2 point to) keep their initial flags
edgy pilot
me when no manpages
surreal path
openat returns a new file description, existing file descriptions (like the one ...
I know
But bash is opening tty with ononblock
brisk zenith
yeah so the new fd returned by openat had o_nonblock, fds 0..2 don't
surreal path
But it reads from fd 0 expecting a non blocking read
brisk zenith
oh hmm
ohh wait
do you have a stub implementation for pselect?
if so, it might be expecting a non blocking read because pselect returned it as readable
surreal path
Yes
brisk zenith
yeah for ttys bash will use pselect before reading
and a stub implementation of pselect would always say "hey all the fds you asked me about can be read without blocking"
surreal path
how does pselect even work
I’m not home on the pc so can’t check man pages
like what does pselect do
brisk zenith
Pselect checks a list of fds and returns whether each one is ready for reading/writing
And it blocks until at least one fd is ready for at least one thing
surreal path
and that can include telling bash that an fd can be read without blocking??
brisk zenith
Yeah, if pselect says "hey this fd is ready for reading" that means it can be read without blocking
brisk zenith
I believe stdin is the only fd bash passes to it, so you're supposed to block until it's ready
surreal path
I believe stdin is the only fd bash passes to it, so you're supposed to block un...
In pselect?
brisk zenith
yeah
surreal path
how do u implement it?
brisk zenith
File descriptions have three related functions: poll, which returns all the events it's ready for (so ready for reading and writing would be POLLIN | POLLOUT); poll_submit, which adds the current thread to a list of threads that will be woken up if the poll status changes, and poll_cancel, which removes the current thread from the aforementioned list. I first set up a pselect context by resolving all the fds. I then do a single "pselect iteration", which runs poll on all the file descriptions, and I return if there are any matching events. If not, and the caller hasn't requested non-blocking operation, I run poll_submit on each file description, then go to sleep. Every time the thread's awoken, I run a poll iteration again, and if it doesn't find any, go to sleep again. If it does find an event, I run poll_cancel on each file description, and return
kind root
same
surreal path
I believe stdin is the only fd bash passes to it, so you're supposed to block un...
is there a way i can stub this in a way to get it to work or do i really have to implement pselect
frigid cliff
surreal path
because it looks like a really complicated syscall
brisk zenith
is there a way i can stub this in a way to get it to work or do i really have to...
I don't think so unfortunately. You might be able to use a hacky implementation that only works for the specific params bash passes, but I believe bash just exits immediately if pselect errors, so that isn't an option
surreal path
I don't think so unfortunately. You might be able to use a hacky implementation ...
im going to struggle with the implementation of pselect then
because it looks very complicated
brisk zenith
It's complicated yeah, but it's better to get it out of the way early imo
Every poll-capable file and device needs to be written with it in mind, so that things blocking in a pselect/poll of that file get woken up when appropriate
It's easier to take that into account and test when you already have it
surreal path
can i just spin in pselect instead
instead of having to do the whole thing of blocking the thread and putting it in some queue or smt
brisk zenith
Like just literally polling the files and yielding if you don't get anything? I'm pretty sure that's a valid implementation yeah
surreal path
Like just literally polling the files and yielding if you don't get anything? I'...
okay
now i have to learn what polling is
brisk zenith
In this context, it's asking the files "hey are you ready to do <x> without blocking"
surreal path
In this context, it's asking the files "hey are you ready to do <x> without bloc...
so basically checking if data is ready
brisk zenith
Yep
surreal path
ppoll() is cringe
poll is cringe
pselect is cringe
ban bash for ever trying to pselect shit
time to implement this the vinix way ™️
vinix
i swear to god
if u weren't written in V ill make u C pain
get it?
C pain?
hahahahahaha
dear mintsuki
wtf
signed rayan
median goblet
@surreal path i have a question
are you going to take care of some glaring bugs i found in your syscalls?
surreal path
for example
<https://github.com/rayanmargham/NyauxKC/blob/fe245b527bc5ee4e78ec35...
whats wrong
median goblet
not only does this allow you to crash the kernel, it also allows you to read from arbitrary kernel memory
buf isn't sanitized or checked for validity at all
surreal path
thats on purpose
median goblet
nyauxkc: not a secure kernel?
surreal path
because i trust userspace with all my heart
userspace would never do something bad ™️
median goblet
thanks
i have a paranoia with userspace so i employ measures to stop it from abusing anything
surreal path
just end game ur local at&t employee
finished
blame everything on unix design then implement it anyway
median goblet
but are you going to fix them, ever?
or just leave them wide open for exploitation like that
surreal path
the security issues?
nah leave em WIDE open
let userspace abuse the shit out of it idfc
surreal path
yea
i dont care about security in this kernel
not like anyones gonna run this kernel as their main os
median goblet
would be a noble goal
surreal path
i was actually going to santizie and verify the buffers userspace gives but @thorn bramble pushed me in the right direction considering i dont care about security anywhere else in the kernel
median goblet
almost nobody cares about it
surreal path
almost nobody cares about it
almost ™️
median goblet
i was actually going to santizie and verify the buffers userspace gives but <@79...
i wouldnt say "pushed me in the right direction"
surreal path
i wouldnt say "pushed me in the right direction"
blame him for it