(!Only Arch Linux!) model issue - Codex-cli - potentially can brick Linux devices running "passwd -d James" and full-access [archlinux - glibc libc]
What issue are you seeing?
I had to format the pc
this pose an issue - because it makes things very hard --- what if your system breaks -- and you decided to a click "encrypt content on harddrive" or whatever crap linux tries to offer and then the user cant boot? because the coding agents decided to run a Sudo command
my coding agent (codex Luna) tried to install Wine using the pacman manager --- then I stopped it --- I use /premission gave it full access --- (I dont have passwords on this machine you so you just do a sudo) I rather do other things then to type my password 20 times (XD)
I have passwd -d <USER> which remove the password
citation from FreeBSD man files
````
DESCRIPTION
The passwd command changes the password or lists password attributes
associated with the user's login name. Additionally, privileged users
may use passwd to install or change passwords and attributes associated
with any login name.
````
````
In the files case, super-users (for instance, real and effective uid
equal to 0, see id(1M) and su(1M)) may change any password. Hence,
passwd does not prompt privileged users for the old password. Privi-
leged users are not forced to comply with password aging and password
construction requirements. A privileged user can create a null password
by entering a carriage return in response to the prompt for a new pass-
word. (This differs from passwd -d because the "password" prompt will
still be displayed.) If NIS is in effect, superuser on the root master
can change any password without being prompted for the old NIS passwd,
and is not forced to comply with password construction requirements.
`````
same thing with GNU (which luckily now more possix complaint so whatever -- I personally like FreeBSD because I only like their logo and i like the color red), but the folks actually speaks english over the freebsd manuals uhh
so as said my "user" has a null oassword
I asked it to debug a very old Siggraph .exe demo , which doesnt have its source code online , (very innovative paper btw regarding computer physics --- which I think this is a weak part in coding agents --- actually porting code -- or reading its asmbbly / using debug tools like Ghidria )
anyways, this is serious --- it ran a command -- on a archlinux machine --- and it kindda bricked the machine ---
I asked it to use Ghidria or Rizin I think --- it correctly decided to use Wine (I interupted it in the middle - and then switched Fullaccess) sudo has no password in it ? I thikn this is why it freaked --- its more a Model issue rather than a Codex issue --- but still, I used Codex, and I literally coudlnt Login to the computer after a restart (lol) luckily whatever I was working on - was save in the "live view" within the Fedora installtion (such a great distro - I recomend anyone using Fedora+KDE ) personally I dont get this OS fights -- the real fight is against bugs that happen with cpus that was solved completely 15 years ago - honestly and cpus are so fast now --- honestly I prompt LLMs and speaking to this robots - I feel a great joy manipulating its output -- I like its first day more - when it was praunt for errors more - and the output could be maniplated --- but with what happen 5 months with coding agents I tihnk is amazing -- I think you need a lot of messages you sometimes can get 20 turns for a simple thing --- or you assume it could related to documentation of tools which usally are heaily documented --- a lot of niche features are there for a rare exotic case (thats why mature tools - have so much paramters --- like passwd for example -- how and when you add things [to/for] [it/that])
on the topic - I used Codex, and I couldnt logged in into my computer -- because of "Arch Linux", and the "Pacman" package manager
and i was blessed and granted with "glibc 2.8somehting?" errors
/usr/lib/libc.so.6: version `GLIBC_2.38' (or GLIBC_2.3x) not found````
which i dont know things should be robust orwhatever
ironically Fedora and Windows and Mac -are very alike - because they are both a release type of software -- you assume you are up to date - or not stuck on a 2.someold.newpatch kind of software
Fedora and Archlinux actually suffer if people are trying to have Flatpak for example and then they end up installing way too many khundrants of mb -
simular to have we used to ship dx11.dll in our binaries (like 20 .dll files in our projects) linux tries to solve it by having a single binaries (fleaing macos cant get a software running for more than 4 years without breaking --- so macos does either of those --- who cares I only use email in the web browser anyways nowadays -- I trust chrome - not to give me a rnandsomware like windows-xp gave one of my old computers that I liked using ---- which is fine maybe they like ) --- so you usally with linux your issue going to be your package manager (competing dependancies --- Both Linux and FreeBSD and MacOS/XNU , is dealing with --- windows does not but you end up with a software being 200mb --- give me a break the whole of CHROME/CHROMIUM is barely 30mbs to install on Linux , its still going to be around 200mb on a fresh linux installation )
https://man.freebsd.org/cgi/man.cgi?query=passwd&apropos=0&sektion=0&manpath=SunOS+5.9&format=html
good read I recomend compareing the Manpath like SunOS and NetBSD for example --- I think its a good read, and how you should structure language or docmention (ironically coming from me, and I have a very cheotic speak pattern)
https://man.freebsd.org/cgi/man.cgi?query=passwd&apropos=0&sektion=0&manpath=NetBSD+4.0&format=html
for example
https://man.freebsd.org/cgi/man.cgi?query=passwd&apropos=0&sektion=0&manpath=Unix+Seventh+Edition&format=html
Unix 7th edition -- has only 30words on the topic (how and why computers designed? - for you to automated tasks duhh )
### What steps can reproduce the bug?
use archlinux (personal tip , do not use archlinux, never use this -- only debian/or fedora) (but fedora is better everything is smooth there )
- I used Codex, and I couldnt logged in into my computer -- because of "Arch Linux", and the "Pacman" package manager
and i was blessed and granted with "glibc 2.8somehting?" errors
I asked one of the LLMs this prompt````
on the topic - I used Codex, and I couldnt logged in into my computer -- because of "Arch Linux", and the "Pacman" package manager
and i was blessed and granted with "glibc 2.8somehting?" errors
@@mention pacman issues on github of this topic@@
````
<img width="961" height="656" alt="Image" src="https://github.com/user-attachments/assets/50b5abb7-b753-43b3-ab54-7148409ce586" />
(it gave me a response I was very pleased by and what I wanted -- sometimes it goes other than I wanted it but still pleased by the llm result) --- its funny how llms are easily can go back to what they were 20months ago - without web sourcing/ grounding with web results
### What is the expected behavior?
Not to break my linux system (but I should open this issue on Arch Linux issue tracker - which i will)
### Additional information
Do not use ArchLinux unless --- if you plan not to used a laptop after booting into the system after 2 years and then you try to get not to break your login
(cant even do this ctrl+shift+f2) to move to a terminal style tty login ---
even after tty2 -- you get like a
`glibc (libc.so.6 ` issue ----
two things you expect not to break on you , glibc and linux kernel updates and the package managers and grub --- all the same falvour
<img width="662" height="147" alt="Image" src="https://github.com/user-attachments/assets/3b917187-366c-4ce5-98b0-0e41c336a32f" />
8 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
@github-actions
<img width="642" height="105" alt="Image" src="https://github.com/user-attachments/assets/c5b1a4d6-9350-4040-b693-1401002ef2fb" />
yes this is correct
----
<img width="702" height="236" alt="Image" src="https://github.com/user-attachments/assets/29ef5031-80bc-4bf0-85c7-a359fdfe3a2c" />
thats not the same thing - my issues is I cant login into my machine (couldnt I formated the machime , I now use fedora on that ancient laptop -- which serve me well - )
-----
lastly
Codex is not to blame here
is this expected that a simple pacman install would break? no this is unexpected - when 99% of the times it doesnt happen (its 99% of the times -- but 100% of arch linux users would face it because after the 200 times you run pacman or yay install you will break it)
as per general knowledge --- I also - install
yayshorly before thathttps://github.com/Jguer/yay
which should be a default application in Arch Linux considering how often you gonna use the AUR store ---- and honestly I dont expected a user with a laptop randomally wanting to compile a very large projects that takes 90minutes --- this is not what personal computing is about
I guess you assumed if you are a British person , you would need to speak French or Spanish or Italian or German, and then access some of the ideas that only exists in this "formats" --- this is not what personal computing is about
I did used
````sudo pacman -S --needed git base-devel && git clone https://aur.archlinux.org/yay-bin.git && cd yay-bin && makepkg -si
````
the laptop was used for 2.5 years (it was running happily Archlinux-KDE-Wayland)
for example as per the issue of what you expect from a laptop - for example on Fedora - it does what happen if you use Windows-11 you can easily utalizes the utalitie on the motherboard --- like if you wanna connect a Blutooth devices (fedora support this out of the box for example --- and you dont need to [go to wiki---> decide which library is most commonly used ---> install it ---> reboot the machine / find the annoying systemd / noooo you need to have a kernel feature enabled in your GRUB or whatever step before you boot into the machine? --- we used to call that just "drivers"])
(for example if I installed bluez-tools --- it what it took for me to get bluetooth working on my laptop-archlinux ---- with Fedora its just out of the box)
some companies only stick for java for reasons I dont get why -- they are many great options --- its not like 20 years --- with software you use the libraries --- and things used to be slower ----- barely being able to browse SVN directories for example -- things used to be very slow --- its like R&D just to find who to relay on --- but most of the times -- you usally can manipulate search engine --- and now we manipulate LLM ENGINES (because its a group of things on the stack) I wish for people who get the play/develop the consumer product like openai and other ai labs --- get to have an ever mutating technology where there is no clear specification regarding ---- per internet speeds -- you can subscribe to a 1gb connection - and give 4 of your neighbors internet access - but have 4 routers -- but we rather pay our ow internet bill - as much as the ideas seems interesting you only really need the same speeds of medium-speed-4g (120ms/40mbit/up/down/) to have a good internet expirence -- not 1000mbit which 1000/10 == 10 users ------ it is interesting -- should be treat serving infrence-llm like -- how fast we gonna "cut and cull" and go to the next user-- its currect to understand there is more correct ideas behind this stacks -- how many gpus / funds ---- as per buying gpus its importent to be moderate -- because hardware can degrade -- why 100billion transistors dont rot after 5years -- I dont know ---- we literally just hit the limit of how many transsitors can be put on a chip ---- perhaps the talent of building the chip isnt going to be gone -- think about mathematics there is much more than being a "professor' after an "approved" thesis , "peer reviewed" while you repeat other peoples frameworks and ideas ---- perhaps the history of computing isnt going to be like the history on how you make a rocket and get to space (very hard things --- yet many companies are going to space ULA and SPACEX) --- the discovery and research on how you make vaccines (and the about 3-4 general categories of vaccines) the trend and trejectory is about the same launching a rocket is very hard-- the first 30years of computing --- and the last 20 years -- the first 20 years of semiconductors - and the frameworks no one really wish to learn today --- neacular indurstry and how innefficent it was in practics -- perhaps the correct words is "revisting" in english
--- generally with operating systems - they are designed to provide Frameworks -- now llms become the framework--- and its general and its way eaiser to put building blocks - at some point its just a building , and not a "plaza"
<img width="935" height="701" alt="Image" src="https://github.com/user-attachments/assets/9124b27f-50b4-4cd1-bc92-5d193e22b4f6" />
image is attached becuase --- what do I choose? I dont care - I just want it to work like Macos or Windows
arch linux is missing the point if I want lightwight ---- thats what "Operating Systems for [embeded devices]" are for
where you can get OS that are modern yet about 200mb size total? --- Linux was mean to be general for all kind of hardwares - and not be stuck on some weird properitry ware -- for example the influence Linux later had on the FreeBSD project --- which came before linux
for example
https://github.com/torvalds/linux/blob/master/Documentation/process/stable-api-nonsense.rst
funny read
and they want to blame AMD and INTEL for what? Honestly it doesnt matter as things stands down x86 is meaningless - we are at a stage - anything meaningful can be ported if it hasnt already -- for whatever we consider novel and more efficent with ARM or RISCV or POWERPC ---- yet intel and amd effectivally killed for many years arm and others didnt pose risks (a lot of people who worked on ARM or POWERPC - now found employment and home at AMD or INTEL) -- x86 in the more mature ways done anything we could ever want from ARM or POWERPC (a lot due to the miracle of we can pack 20x transistors with the same amount of power) but undar the hood it is about the same thing
))))
honestly FreeBSD and openBSD , dont have eBPF ----despite sBPF being a lot faster --- uhh do we really need to put javascript layer over it? i wouldnt call it metastasized but I am calling it this way
eBPF is the selling points of what FreeBSD was better at the past -- now there is no point in using this unmaintained landscape?
eBPF effectively took BSD's historical superpowers and turned them into Linux's modern ecosystem advantage
almost every path and commit into linux you see REDHAT , literally they are the last people I expect to screw off my package manager like what ARCH LINUX just did (which I dont care I love everything and every software bad or good equally ----) --- Fedora is the only good linux distro - this is a fact because of the high threshold of software that support this kind of eco system natrually/out-of-the-box-Cle en main
as for "Arch Linux" its not even a software that worth evaluation -----
use Fedora -- its the only thing that properly get maintained (even windows or macos nowadays dont get the same treatment ---- beside getting a captcha because your user-agent in your browser has "linux" in it)
linux always had a crappy feeling because of lack of frame vsync ---- but now Wayland fixed it out of the box --- literally the same year of Wayland - we see people using linux way more , and more ---- so it was x11 that caused the issue that people didnt likeed it ---- and you know whatever Windows and MacOS had when freebsd and linux personal computers had --- was really really really better than anything else --- I thikn its always importent to also look at the first itiration of softwares - and what you got today and compare --- and see what is robust and working ---- we are only 70 years into the computing world ---- and not 200 or 400 or 600 into this world of computers ---- who cares really , I didnt had pertussis/whooping-cough -- you didnt had it --- I personally think its halirous people at the linux kernel and on Red Hat using MacOS as their daily driver while being a part the linux kernel or Fedora --- speaking of what we had at the early days of Linux 3.0 and linux 2.0 --- now we are a few years after the manner ------ you can ask any LLM if you need some task done with the terminal if you are lazy for something or your memory isnt sharp so you dont need to hammer a search engine to a find man/documentation page --- if you are omiting a serious of general serious of general porpuse commands --- on another layer every system would be using systemd --- its not that "its the genius of linux" its just happen natrually people used other people work (like tool calling today by coding agents) making all of this Linux is just a kernel you have dozen of distros ---- linux for example serve a large scope technicals things ---- yet you should be bond to a mainstream framework --- I personally think Fedora is the only home - it should bare the name linux --- I think its funny the eBPF comes from the realms of busybox thats the taste I get from it atleast - a flavour of everything like what gpus aim to be ---- a chip and a computer- -- on a PCI EXPRESS ---- kind of what you viewed a cpu in with instruction -- a gpu saves you the time for the tasks you dont need to do --- a PCI EXPRESS and a direct HIGHWAY to communicate with the CPU -- a sanbox for --- a gpu is like another computer , we see a PCI EXPRESS moment -- like what you see with this UNIFIED memory -- where a GPU would bare the name of the same level of a CPU ... we all aim about this hardware --- so an OS is literally software - and its not coded in some magic , just clever ideas and clever people very knowledgable in the C standards -- and applying and creating specification
--- linux adopted to be something so great despite its limitations and "mal->p"ractices ...
Why GNU I dont know I dont like this FSP type of people - there is a general okayness - of preducing and maintaining market -- in a sociologal importance level ---- like you should play the consumer part --- not like a "developer/creator" part
we all consume dependentices --- none of us reinvent the wheel --- (for example this is why I like python/zig so much) and as per rust , uhhh it has its place - its proven to be great - it aims to generalize dependenices which is something I recognize as a good thing in my world view ----- you buy a second hand car - and you expect it to work --- you may be very happy with what you got ---- (for example,,, all of our houses and all the houses are second hand-- not really -- but a most common thing)
for example I like the language used in FreeBSD pages --- and when you read a manual file in Linux not always
for example the https://xfree86.org/
is meaningless because of the luck of adoptation ----- what gonna happen you gonna look at the commits and patches linux does---- and slowly relize Linux is the best... afer all ... but its importent that its a skill that isnt being lost to "dune sands of time"
--- what linux is extracted from the idea of -- maybe host some of the documentation in your project --- linux was successful - it was at the right time --- but also knew to copy and migrate software with it ---- with the "basic" task to sastify a user -- so he can browse his email and youtube -- without the screen flickering
as per https://github.com/openai/codex/issues/35492
the coding agents --- tried to install Wine - and then do PE debugging over it ---- the "point of failure" / "root cause" was "Pacman" it was pacman , the package manager of Arch-linux ---- not having intel drivers out of the box or have wayland ----- is one thing
(just for the sake of porting a project --- cited in literature)
despite that I dont see a world where a designer sole job is to be a designer --- a designer shal be an artifact of another craft - thus relating of what the user would feel useful - not provoking/eleminating the purpose -- not wanting to more creative going on-forward.
as per the ZIG language --- I dont see the reason its a fancy idea to fight with GCC or LLVM or msvc -- Fortran Compilers --- compilers nvidia makes and other dozen ---- as per RUST and I think it iterates from the late stages of the c++ standards foundations -- and it understood we needing something different-yet-general --- but zig is very compact and nice --- so is python with cffi --- unless you dont serve 20 billion internet users - you dont need something that is not python -- you can use typescript or javascript - just be careful with oral hygine (if not javascript - for basic amount for user joyment and enhancement -- clicking buttons or whatever )
but then the coding agent literally bricked my computer --- which is not getting breached --- but you know -- I couldnt reboot my personal-computer(it was a laptop)
as per "ai factorys" they are still called data centers , you spin 20000pb - ever nano seconds - and eBPF is the reason nvlink is even a thing
the so smart model - would try to use bpftrace -- on a none sudo
when you must be root to use it................. also ptrace