🍌

“KDE Linux” (codenamed “Project Banana”) is a work-in-progress name of a KDE-owned general-purpose Linux distribution proposed at Akademy 2024. Not to be confused with KDE Neon.

Warning

This page serves as a design document, thus information presented here should be considered a snapshot of the ongoing discussion, not final decisions.

This page has lots to talk about, please consult the table of contents on the left :)

Goals

TL;DR: Create a bulletproof OS showcasing the best of KDE that we can proudly recommend to users and OEMs, with a coherent "here's how you get it" story.

Goals in detail:

Be "The KDE operating system"
User-friendly; high-quality UX
Doesn't break, or at least easy to recover
Keeping security in mind
No packaging knowledge needed to develop for it
Focus on modern technologies
Attractive for our hardware partners
Can be used as the main system by our developers for internal dogfooding purposes
Exercise codepaths for containerized apps and immutable base systems, to improve KDE software deployed using these technologies in other environments

Non-goals

Does not have to support the runtime installation of kernel modules. This will prevent the out-of-the-box installation of, for example:

Proprietary NVIDIA kernel driver (NVIDIA GPUs must either be new enough to use the open-source kernel modules that can be distributed in-tree, or else use Nouveau)
VirtualBox
Vendor-specific VPNs that require custom out-of-tree kernel modules that cannot be redistributed with the kernel due to license incompatibility

Does not have to support the use case of developing low-level system components like the kernel, drives, systemd, etc., as this can be troublesome with an immutable base OS.

Target audience and use cases

It should have multiple editions suitable for different kinds of users. Ideas:

Developer edition: built from git master and released daily, including debugging tools and KDE dev environment. Like Neon Developer.
Enthusiast edition: ships released software, and releases to users on upstream KDE's schedule, like Neon User. Additionally, when there are any beta releases, ships the beta.
Stable edition: ships only released software on a delayed schedule, based on TBD quality metrics.

Architecture

Original architecture ideas for the project included the following:

Reproducible builds, must-pass CI, automated UI testing
Base OS is Arch-based. OS updates are some degree of rolling; snapshot based releases with relatively recent libraries
Systemd-boot as the bootloader with nice boot theming
Btrfs as the filesystem
Encryption of all mutable data (e.g. user homedir, and cache locations on /)
Included recovery partition
Read-only base system, like SteamOS, Kinoite, and MicroOS
Atomic image-based A/B updates with rollback functionality
Manual package installation happens transparently using a per-user or systemwide overlay
Apps are from Flatpak (and maybe also Snap if it's not too hard and the UX is okay)
Wayland by default
Automatic user data backup system using Btrfs snapshots, with a nice GUI around it like Apple's Time Machine
DConf-like configuration management UI suitable for enterprise and managed environments leveraging KConfigXT for everything
Simple input method configuration for CJK and more
"Troubleshooting hub" app

TODO: hardware support, software separation, security model, deployment, OEM mode; proposed solution, alternatives, trade-offs for each section

updates

systemd-sysext

systemd-sysext allows us to overlay developer content on top of /usr without impacting the base system.

Setup

# create directories
mkdir -p ~/kde/usr/lib/extension-release.d/
# create an extension-release file
cp /usr/lib/os-release ~/kde/usr/lib/extension-release.d/extension-release.kde
# make the ID ignored so updates don't break the extension
sed -i s%^ID=.*%ID=_any%g ~/kde/usr/lib/extension-release.d/extension-release.kde
# owned by root so it can't be removed
sudo chown root:root ~/kde/usr/lib/extension-release.d/extension-release.kde
# enable the extension
sudo mkdir /var/lib/extensions/
sudo ln -s $HOME/kde /var/lib/extensions/kde
sudo systemd-sysext merge
sudo systemd-sysext

Use

Use DESTDIR=~/kde to install stuff and then restart systemd-sysext. Beware that when changing polkit/dbus stuff you also want to restart those services as they don't necessarily pick up changes.

DESTDIR=~/kde ninja install && sudo systemctl restart systemd-sysext.service

Prototype

The code is currently located here. Note that it is not representative of the final product and exists as an experimental playground for now.

Installation

Download the latest .raw file
Attach a USB drive
Use lsblk to find the right /dev/node. e.g. /dev/sda
sudo dd if=kdeos.raw of=/dev/sda bs=4M
sudo sync
Reboot into the USB stick
no password on SDDM
run Calamares via Kickoff/KRunner

Updates

Until discover gets support the following needs running

git clone https://invent.kde.org/sitter/kde-linux
cd kde-linux
sudo ./update.sh update

VM

virt-manager

File -> New VM
Import existing disk image
[Forward]
Select from disk
Set arch as OS
[Forward]
Set resources
[Forward]
[x] Customize configuration
[Finish]
Config window opens
Make sure at the bottom it says Firmware: UEFI
[Add Hardware]
Add a storage of some reasonable size
[Finish]
In the boot options item:
check VirtIO Disk 2 and move it above 1
[Apply]
[Begin installation]

Local Development

In order to speed up local builds, you can create a `mkosi.local.conf` file in the root of the repository with the following content:

[Content]
Environment=LOCALE_GEN="en_US.UTF-8 UTF-8" # replace with your locale`
Environment=MIRRORS_COUNTRY=us # replace with your country code`
Environment=PARALLEL_DOWNLOADS=50 # if your internet connection is fast

# Only uncomment this after you have done a complete build once
#Environment=KDE_BUILDER_ARGS="--no-src --install-only"

Then (assuming you have docker with the BTRFS storage driver) you can run:

./build_docker.sh --incremental

Related projects

Differences from other immutable distros

(e.g. Kinoite, MicroOS, SteamOS)

Principally, that it is distributed by KDE. This has several advantages:

The chain of responsibility is never gated on a third party
KDE and KDE e.V. can have a direct relationship with third parties using it, e.g. hardware OEMs
KDE can explicitly recommend it without "picking favorites" from among other distro partners

Secondarily, KDE Linux relies almost exclusively on systemd tools. This means for example that updates use systemd-sysupdate rather than something like RPM-OStree.

Tertiarily, making the base OS exclusively image-based with content installed using Arch packages and kde-builder means that no packaging knowledge will ever be required by KDE Linux developers.

Prior art

KDE Neon, KDE's first version of a self-made OS. Neon fulfills the "distributed by KDE" requirement, but fails on the reliability angle due to the Ubuntu LTS base that ironically becomes unstable because it needs to be tinkered with to get Plasma to build on it, breaking the LTS promise.

Roadmap

TODO (milestones)

Long-term maintenance

TODO (team and infrastructure requirements for long-term sustainability after release; update cycles; testing infrastructure; architectural future-proofness)

Governance

TODO

Promotion

TODO (name and branding, public image, effect on relations with other distros and hardware partners)

🍌