Discarded Wayland, it's too damn slow. The fact Xorg is a client-server architecture was a good decision. It's Wayland what is taking the wrong ones. You're creating a thousand implementations of the same thing, each one with its broken parts.
Decided to stop trying to make Sway fits, it's too damn slow. VNC is bad. Xorg doesn't need to transmit every pixel. Instead I figured how to make XRDP only work on the interface of my Zerotier network (therefore not accesible to the public IP).
Here it is so you don't have to learn it. You do not rewrite xrdp.ini, you tell configuration.nix what you want to replace from /etc/xrpd/
I managed to run Sway (not Hyprland) over VNC which is close enough to Hyprland since I can map the key bindings easily and use the same Wayland tools.
.....but, it's dreadfully slow compared to X-based Plasma6 on RDP. Plasma6 feels like a local PC
That's so strange.
EDIT: Or not so much when you consider VNC transmit the whole damn thing pixel by pixel whereas RDP on Xorg does not need to.
Notice I am being negative but with a smile.
I am giving constructive advice. This OS is salvageable, but only with MAJOR changes.
The current GCC project has to be completely abandoned, for instance. NixOS has to use ordinary GCC compilers that are compiled as the full suite, with full bootstrap, and then split, just like other distros use. And there must be no need for wrappers.
Of course the answer to my question is this notion you can run multiple glibc on the system without isolating them into chroots.
Do you see what the consequences of that are? THE SYSTEM IS SEVERELY BROKEN.
All you had to do was isolate the different libraries into chroots! That was all you had to do and you didn’t do it! Now look what a mess you have.
NixOS is garbage.
Or let’s put it this way:
It is loads of fun trying to make something that doesn’t SUCK SO MUCH out of a NixOS system.
I almost thought I couldn’t do ANYTHING with NixOS, when I started out. But now I have a full set of compilers that can build actual executables I could give to people. I just installed the latest Git of Poly/ML.
Unlike the rest of you, I have a complete GCC 15.2.
One of the things that gets to me about NixOS is, if they are so keen on linking things together by RUNPATHs, then WHY DO THEY NOT HAVE AN ld.so.cache ?
WHY DOES NIXOS NOT HAVE A REGULAR ld-linux.so?
Do you see what I am getting at? Becuase RUNPATH overrides ld.so.cache. So, in fact, there should be no problem unless the RUNPATHs are broken. So why introduce even MORE breakage?
NixOS is IMO a piece of garbage, even though it is my OS. I merely tolerate it.
...my user and /home/maikel files safely.
THIS would have taken me HOURS if not DAYS of pain without Nixos.
And sure you can install Nix in Ubuntu but it's not the same, it doesn't enforce it. With Nixos you HAVE to do Nix all the time so you get VERY productive with it very quickly.
I still don't like Flakes though. I like nothing that is experimental and I still DESPISE the manuals.
And Ansible is too slow, lacks features of Nix I use all the time.
I hate Nixos, and yet I can't live without it.
I was happy in Ubuntu.
Why can't anyone make a Nix thing that's not Nix and just works.
ANSIBLE you might say. It's just not as fast. I don't manually install anything ever again on any of my computers and having a duplicated environment costs me nothing.
Do you think I would have been this relaxed with my main PC gone without Nixos? Nope, not possible.
Yesterday I logged in, assigned my public keys to my cousin user (so .ssh/authorized_keys) via home-manager on a specific hom.nix created for him with Chrome, Steam (he's a kid needs to play) etc and erased ...
It's 2025 and we're still pretending that Wayland is AWESOME and everything that previously worked in Xorg that is 100% necesary, still works in Wayland.
In the meantime, good luck trying to set up a headless Hyprland with no monitor output (so a server) while the same server you can literally just log into KDE Plasma6 with zero issues with just these 5 lines in configuration.nix
(I had it disabled because I was just experimenting and I don't want that computer accesisble to the world, just to the Zerotier interface)
I, ah, got #NixOS working in a VM today
#nixos #qemu #virtualization
okay my home-manager and sshd both services worked
well hm's only log error is that it cailed to create to some hm prorgam's files
I am not disabling #dconf for that
I can use emacs from serial port #virtio , even vterm
and ssh to host machine
2 questions
- why is it so slow, even the #direnv and how can I speed it up
( obv it's built without a bootloader, as that's way faster)
- sharing data from host also means if I mod something in that from vm , it also gets modded on host , my point of making this vm was to test new configs for system
What re some quick and dirty ways out of it?
yes I can do with #podman maybe, but it can't share my nixos modules which enable all the display in 1st place
is there a #rust way ?
Notes
1. it doesn't include/shares nix-channels from host
2. neovim is way cooler , it can display colors, even on serial port
#linux
@ramin_hal9001
I'm back home for Diwali and about to discover just how fast #NixOS unstable moves, since the last rebuild on my home PC was on 2025-07-08 so I have just over 3 months of package upgrades to catch up on...
Isnt #nixos iso image just a VM built with bootloader ?
Okay almost almost an iso
key diff between containers and vm for #nixos
container shares the host hardwar vm doesn't
vm can easily share nixos modules
container doesn't
looks like #nixOnDroid is a container and not a vm
@niklaskorz
Soll das so? Oder bediene ich es falsch?
#nixos
Is anyone here using #helixeditor and the helix-gpt language server?
I'm on #nixos and have Helix and Helix-gpt packages in... environment variables set right, and my languages.toml says
[language-server.gpt]
command = "helix-gpt"
args = ["--handler", "openai", "--no-completion"]
and my php is set to
[[language]]
name = "php"
language-servers = [ "intelephense", "tailwindcss-ls", "vscode-html-language-server", "gpt" ]
auto-format = true
Yet, Helix acts like it has no idea what helix-gpt is
As before, my /usr/local is a ‘systemd service’.
Oh, that sounds so fancy. I loathe systemd because of the ‘vertical integration’ and the hype-laden, authoritarian culture, not because of any particular thing it does.
This ‘systemd service’ is a start script and a stop script. Ordinary init system stuff. It could have been a single script or program. It could have been integrated into an rc.d script system just as easily. Systemd is hogwash. But it works alright here.
My NixOS /usr/local is now entirely differently structured. It is because I stopped trusting fuse-overlayfs after it wouldn’t return a file to gcc, even though I could access it with other programs.
I am now using the kernel’s overlay fs with just one lowerdir and an upperdir. I construct the lowerdir mostly by untarring (two-stage install) packages on top of each other. They are compressed with zstd so decompression is fast. They are BLAKE3 hashed for update management.
Solved sekrets management for my budding NixOS pi application cluster (kinda :P)!
In this case, it's transcrypt to the rescue: https://github.com/elasticdog/transcrypt
....I really should have initialised it in aes-256-ctr instead of the aes-256-cbc default tho.......
But yeah, transcrypt + git submodules = I can open source my NixOS config in the future!