I have for the past half year. I don’t have numbers, but rolling with the NVK vulkan driver (context for the unaware [1]) on mesa’s main branch gets me somewhere around half the proprietary driver’s performance on average, and can be accompanied with stutters if it is a heavy bottleneck (turning down the resolution is an easy way out). Most games I’ve tried are runnable now.

It sounds like you’re looking for more performance with this post though, so you’re likely not going to see improvements taking this route. I would still suggest giving it a try for people that are able and can tolerate the sacrifice. It’s good enough for me (and better in the wayland case) that I rarely swap to the proprietary driver anymore.

Mesa has a tracker issue for games on NVK [2] [3] with reports about game statuses and issues from the past 5 months. It includes playable and unplayable games for those interested in gauging its usability.

Also, for the record, NVK is no longer considered experimental as of mesa 24.1 (May of this year) [4].

[1] https://www.collabora.com/news-and-blog/news-and-events/nvk-has-landed.html

[2] https://docs.google.com/spreadsheets/d/1RuHD3Z_nBKCp618HHC5I9hOu0lqCoFYwQ4FM69M-Ajg/edit?gid=469568508#gid=469568508

[3] https://gitlab.freedesktop.org/mesa/mesa/-/issues/11066

[4] https://www.collabora.com/news-and-blog/news-and-events/nvk-is-now-ready-for-prime-time.html

You might be interested in git-annex (see the Bob use case).

It has file tracking so you can - for example - “ask” a repository at drive A where some file is, and git-annex can tell you it’s on drives C and D.

git-annex can also enforce rules like: “always have at least 3 copies of file X, and any drive will do”; “have one copy of every file at the drives in my house, and have another at the drives in my parents’ house”; or “if a file is really big, don’t store it on certain drives”.

Slight correction, it’s available in Linux 6.7, and has to be enabled with the kernel parameter nouveau.config=NvGspRm=1. It may be enabled by default soon though (latest news I could find about that here).

NVK itself is not dependent on anything proprietary, but it’s practically required to enable NVIDIA’s GSP firmware blob if you want to see actual performance since it’s what enables re-clocking (older post by Collabora here which touches on it).

A neat trick I use to figure out what apps belong to what packages is to run realpath $(which $app_name) - this outputs the full, canonicalized path to the store item, which should include the package name.

I would do a search (e.g. using grep) through a local copy of the Guix repository to figure out what services are providing the packages, then modify the service configurations (or use modified versions of services, if needed) to remove them. This might be a tedious solution, though.

I haven’t found a need to do it, but a (modify-services ... (delete pulseaudio-service-type)) in your operating-system declaration might do what you’re asking? I don’t think this is necessary though. As far as I’m aware, applications that attempt to use Pulseaudio will be transparently rerouted through pipewire-pulse, which is already configured by home-pipewire-service-type. I am also on GNOME, and I haven’t noticed any breakage in this aspect using it.

If you’re unsure that it works the way you want, you can always try the configuration out and see how it goes (note that I had to re-log in for the wireplumber service to start properly).

I’ve never tried putting it in the system configuration, but I imagine it wouldn’t work as it depends on Guix Home services.

If you haven’t used Guix Home before, the home-environment record doesn’t have required fields like operating-system does so it’s fairly easy to get started with.

(use-modules (gnu)
             (gnu home services desktop)
             (gnu home services sound))
(home-environment
 (services
  (list (service home-dbus-service-type) ;home-pipewire-service-type needs this
        (service home-pipewire-service-type))))

There is, actually! It’s available as a Guix Home service: home-pipewire-service-type was introduced around the end of December, and you can see documentation for setting it up in the manual.

When using this service, pipewire-pulse is used by default without any extra configuration.

Cheers, glad to hear you got it working. I don’t think there’s any problem on your end; all my flatpaks are user-installed as a Guix System user, so it didn’t cross my mind that a habitually-placed --user flag would not work if something was installed system-wide!

Thanks for posting about this! I never thought to try this as an Akregator user, but it’s a great idea… I spent the past day getting this to work since I also use the Flatpaks; hope it helps.

As suggested by @progandy@feddit.de, one solution is to define a custom protocol where the URL gets passed to a script that opens Firefox Reader with the URL; here’s what I’ve done:

1. Decide on a protocol name, which the URL will be prefixed with and passed to xdg-open since that should be available to the Flatpak. I used firefox-reader as the protocol, so I put xdg-open firefox-reader://%u as the custom command (so a command Akregator would run might look like xdg-open firefox-reader://https://example.com).
2. Define a desktop entry to support the custom protocol (you can see mine below). ~/.local/share/applications is the standard place to put these, as far as I’m aware. Since the custom protocol needs to be removed from the URL, I wrote a script (also below) to do this and then call Firefox with about:reader?url= prefixed. The script can be anywhere in $PATH.
3. Add the desktop entry as a “default application” for opening URLs using this custom protocol. In my case, I ran xdg-mime default org.mozilla.firefox.reader.desktop x-scheme-handler/firefox-reader (org.mozilla.firefox.reader.desktop is the name of my desktop entry file).
4. You also might have to update some mime/xdg database stuff. I had to run update-desktop-database ~/.local/share/applications so xdg-open would find the “Firefox Reader” desktop entry.

[Desktop Entry]
Type=Application
Name=Firefox Reader
Exec=open-firefox-reader.sh %u
StartupNotify=false
MimeType=x-scheme-handler/firefox-reader;

#!/usr/bin/env bash

flatpak run --user org.mozilla.firefox about:reader?url="${1#firefox-reader://}"

If you have any other trouble or want to find more information about this since the desktop entry could probably be tweaked, here are the sources of note I used to figure this out (If I forgot a step or two writing this, they should also be present somewhere in there):

• https://unix.stackexchange.com/questions/497146/create-a-custom-url-protocol-handler
• https://wiki.archlinux.org/title/XDG_MIME_Applications
• https://wiki.archlinux.org/title/Desktop_entries#Update_database_of_desktop_entries

I’ve been having my own fun trying out NVK on Guix System so I can’t give you specific instructions (assuming you’re not using Guix), but I can tell you what you need and maybe someone else can chip in on how/if you need to do anything else on your distro:

• Linux kernel >= 6.7
• GSP firmware enabled via the kernel parameter nouveau.config=NvGspRm=1
• Mesa built with the -Dvulkan-drivers=nouveau-experimental flag.

A few notes:

• Performance and stability of games has been fairly hit-or-miss for me. Of the 10 Vulkan games I’ve tried so far: 3 run perfectly, 3 are playable with noticeable issues, and 4 are borked.
• NVK is Vulkan-only, but performance largely comes from the GSP firmware so you will still see a difference (huge for me) in games not using Vulkan.
• You can override flatpaks to use the host’s Mesa version (set FLATPAK_GL_DRIVERS=host); however, there’s a bug that causes the Steam Flatpak to not work when doing this. The mesa-git-extension Flatpak exists which can also be used to replace Mesa runtimes, but it had issues building with NVK so it’s currently disabled.

The only package I’m aware of at the moment (other than my hacked-together Guix package) is available in the AUR.

It appears that this is actually a Bash thing rather than a Guile thing:

https://superuser.com/questions/132489/hostname-environment-variable-on-linux

HOSTNAME is a variable that bash (but not other shells) sets by default. It is not exported.

Since it’s not exported, it will not be inherited by env or any C program invoked from a bash shell.

For more information about why, see the comment in this other answer which mentions that HOSTNAME is not required for POSIX compliance:

https://superuser.com/questions/132489/hostname-environment-variable-on-linux/132500#132500

https://pubs.opengroup.org/onlinepubs/009695399/basedefs/xbd_chap08.html

Curiously, I can reproduce this on my Guix System, but actually not on my Fedora system (where it is exported automatically), so I guess some distros choose to explicitly export the HOSTNAME variable?

Assuming that you are using substitutes, I recommend to try waiting a day or two between a guix pull and reconfiguring your system. Unless you have an absurdly slow internet connection, it sounds like your guix is falling back to building some packages due to the substitute servers not having the freshest packages built yet, which can happen often if you reconfigure immediately after pulling.

I use git-annex and Guix (particularly Guix Home in this case) for managing all of my data, including dotfiles. git-annex handles syncing (and backups via delivery to a Borg repo) and version management as git does, while Guix takes care of installing programs and setting up configuration files.

I previously wrote a custom Guix service that utilized Stow as well for managing writable files, but have since replaced it with another custom Guix service that handles some cleanup processes better.

Depending on how much you want simplicity of restoration, this approach might be on the heavier side since it’s concerned with a lot more than just dotfiles. You could replace git-annex with git to simplify the syncing part if you’re only interested in managing configuration files, though. Here’s what my Guix config looks like; the readme file shows how I would set up a system from scratch.

Was not aware of this, thanks! Looks like it does, with a notable difference being that Cryptomator has better cross-platform support in exchange for not having file size obfuscation.

The CryFS developers have a comparison page here that might help you decide what to use. There’s a summary table at the bottom that gives a comparison of features between encryption filesystems if you don’t feel like reading through it all.

I personally use and would recommend CryFS because it’s the only one (that I’m aware of) that plays nice with data synchronization software (i.e. doesn’t store the container as a single file) while keeping the directory structure encrypted.

They’re talking about this Squad.

I haven’t gotten around to trying the stable release out, but there’s one ProtonDB report - presumably for 6.0 given the post date - so far that says it works flawlessly.

I’m guessing the situation is the same as pre-6.0, though. I participated in some of the playtests, where I got the same performance issues that I got before with the game slowing to a crawl after some time. The user on ProtonDB also has pretty beefy specs, so I couldn’t say if the performance issues were fixed, either. I’m not sure if you’re asking if it’s playable or has gotten better, so I will say that (at least for pre-6.0) it technically works regardless as long as your computer is beefy enough.

The Proton GitHub issue for Squad might be nice to bookmark to make or check up on every once in a while; usually any issues, fixes, and updates end up there.

Does Guix fit your criteria, perhaps? If you haven’t heard of it, you can think of it as Nix with a Lisp frontend.

I unfortunately am not very experienced with containerizing packages so I can’t say much, but I know you can do it; the Nonguix channel employs containers for some proprietary software.

Like Nix, Guix has all that building-from-source stuff you’d want from Gentoo. There’s recently been work on making parameterized packages (the Guix equivalent of USE flags) a thing, but it’s still work-in-progress.

Ignoring the steep upfront cost of learning it, I’d say Guix makes it incredibly easy to add your own packages. Here’s the custom packages I currently have in my dotfiles repository. I can import one to my main config file, add the package, and it gets included in my environment the next time I reconfigure it.

As for patches, I can’t make any comparisons since I’m not familiar with Gentoo, so I think a code snippet is probably better for you to judge if you’d like it.

(define-public custom-pkg
  (package
    (inherit pkg)
    (name "custom-pkg")
    (version (package-version pkg))
    (source (origin
              (inherit (package-source pkg))
              (patches
               (list (string-append (dirname (current-filename))
                                    "/fix-some-thing.patch")))))))

(define-public custom-pkg
  (package-with-patches
   pkg
   (list (string-append (dirname (current-filename))
                        "/fix-some-thing.patch"))))

Not sure if this addresses your concern about multi-architecture support, but the Foreign Architectures section of the manual discusses what you can build to.

EDIT: So I was curious after posting this because usually the CLI often has much less verbose options (like --with-input for replacing inputs), and I started wondering if there was any procedure that would make this simpler. Turns out there is :) I’ve included it under the example. Although, I suppose I should have mentioned you could write your own if you really wanted to.

Could you elaborate on why you think Guix will attempt to pull systemd? As far as I’m aware, Guix doesn’t impose any hard requirement on systemd when installing it on a foreign system; the manual (particularly 2.1 Binary Installation) simply provides instructions for setting it up on systemd and Upstart.

I just found this blog post about setting up Guix on Void Linux if it helps at all.