Comment by 0rzech | Hacker News Reader

Comment by 0rzech | Hacker News Reader

Parent Full thread

I'm replying to the whole thread at once:

The Silverblue, IoT and CoreOS logo are in the section "Related Projects". This is probably because all of them (not sure about IoT), including Kinoite, use rpm-ostree.

Both Silverblue and Kinoite are named after crystals.

Kinoite is like Silverblue with KDE instead of Gnome. They have separate deployment channels due to the nature of underlying technology.

Neither Silverblue nor Kinoite is considered primary variant by Fedora Project yet, because they're still rough around the edges. But I hope they will be in the future due to their advantages, which are mentioned on the first page of their respective docs [1][2] and which I have experienced myself.

The only thing I don't like is using Toolbox to manage CLI programs and the fact that `rpm-ostree install` always includes `Recommends`. I think Flatpak should handle CLI apps as well as GUI.

[1] https://docs.fedoraproject.org/en-US/fedora-silverblue/#intr...

[2] https://docs.fedoraproject.org/en-US/fedora-kinoite/#introdu...

I'm all-for the advantages of rpm-ostree and sandboxing - these technologies are great, and they need a platform to be experimented on. Unfortunately, I feel like relying so heavily on Flatpak is a bit of a hindrance for this distribution. Anything but the most basic of use-cases is going to cause friction, and ultimately leave your OS feeling more like an iPad than a full-featured desktop. For terminal users and developers and people who want to run background services, this is a complete non-starter.

This could have been a great opportunity for Fedora to raise the bar and lead the industry by redesigning RPM to integrate with Bubblewrap and provide unprecedented thin-sandboxing. That's a great sell to enterprise users! The execution feels a bit botched though, and there's a lot of work left on the table before it can compete with other modern distros. It's creative, but not creative enough IMO.

But the whole point of rpm-ostree is to avoid rpms where possible. You can't be for rpm-ostree and at the same time propose continuing with traditional packaging system, which is discouraged for immutable systems.

Snap and Flatpak where not only meant to enable sandboxing, but also to eliminate huge distribution repositories of packages all modifying root directory contents and requiring redundant packaging work.

I agree that neither Flatpak nor Snap are there yet, but we're all still in the transition stage.

> but also to eliminate huge distribution repositories of packages all modifying root directory contents and requiring redundant packager work.

Well... I'd argue both of them failed. Snap is well-made, but mired in Canonical politics to the point that it's unusable. Flatpak has good intentions, but is generally poorly maintained and driven by ideologues rather than project managers and top-down leadership. That's without mentioning the technical implementations of either project, both of which fail to eliminate redundancy in their runtimes and feel completely foreign to the native system.

Nix shouldn't have to come up so frequently in these discussions, but it's a much better package manager than Flatpak or Snap (isolation technologies aside). If Flatpak also worked with software maintainers instead of embracing mutual distrust, it could be faster and lighter on the system. Instead of tracking our dependencies properly, we've resigned ourselves to just targeting "Another Competing Standard" and hoping it won't also become irrelevant in 5 years when rpm-ostree-2 drops. It doesn't strike me as a resilient or reliable technology in the long-term, especially on it's own.

The redundancy Flatpak and Snap induce is nothing in comparison to hundreds of, say, coreutils packages - each for some Linux distribution and its supported versions.

Also I don't think tracking dependencies is the root problem, rather different applications requiring different dependency versions, which leads to even more redundancy on top of the aforementioned one. And if you avoid providing different lib versions, you end up with silly application version freezes until next distribution version is released along with updated repos.

Rpm/deb/etc. repositories simply don't scale. You either end up with a small well-maintained repo, or a huge one with many abandoned or version-frozen packages. Not to mention you'd have to write sandboxing rules for each distro version separately, which is why most distros do not provide complete sandboxing at all. Perhaps there are some distributions inbetween, but not in my experience.

I do understand current Flatpak's and Snap's technical limitations. I just think they will be solved, sooner or later. I'd be happy to use Guix or Nix instead of Toolbox, but currently their requirements conflict with how rpm-ostree treats system's root directory.

Great answers in this thread. I opt for distrobox [1] over toolbox.

https://github.com/89luca89/distrobox

Reply on news.ycombinator.com