Timothée Besset, a software engineer who works on the Steam client for Valve, took to Mastodon this week to reveal: “Valve is seeing an increasing number of bug reports for issues caused by Canonical’s repackaging of the Steam client through snap”.
“We are not involved with the snap repackaging. It has a lot of issues”, Besset adds, noting that “the best way to install Steam on Debian and derivative operating systems is to […] use the official .deb”.
Those who don’t want to use the official Deb package are instead asked to ‘consider the Flatpak version’ — though like Canonical’s Steam snap the Steam Flatpak is also unofficial, and no directly supported by Valve.
This might be an unpopular opinion but I really don’t get this trend of wanting to containerized just about everything, it feels like a FOTM rather than doing something that makes sense.
I mean, containers are fantastic tools and can help solve compatibility problems and make things more secure, especially on servers, but putting everything into containers on the desktop doesn’t make any sense to me.
One of the big advantages Linux always had over Windows is shared components, so packages are much smaller and updating the whole system is way faster, if every single application comes with its own stuff (like it does on Windows) you lose that advantage.
Ubuntu’s obsession with snaps is one of the reasons I stopped using it years ago, I don’t want containers forced upon me, I want to be free to decide if/when to use them (I prefer flatpack and appimage).
Debian derivatives that don’t “reinvent the wheel” is the way to go for me, I’ve been using Linux MX on my gaming desktop and LMDE on laptop for years and I couldn’t be happier, no problem whatsoever with Steam either.
Shared components work brilliantly in a fantasy world where nothing uses new features of a library or depends on bug fixes in new versions of a library, and no library ever has releases with regressions or updates that change the API. That’s not the case, though, so often there’ll exist no single version of a dependency that makes all the software on your machine actually compile and be minimally buggy. If you’re lucky, downstream packagers will make different packages for different versions of things they know cause this kind of problem so they can be installed side by side, or maintain a collection of patches to create a version that makes everything work even though no actual release would, but sometimes they do things like remove version range checks from CMake so things build, but don’t even end up running.
Shared containers work beautifully for a lot of things, though, many programs aren’t all that sensitive either. Making snaps for the tricky ones makes sense. Having snaps for all of them is ridiculous.
I can count the software requiring repo-pins on one hand on my desktop. For those, snaps make sense, replacing the need for any pins. Snaps are less confusing than pins. IMO.
It reminds me of Python programming, with requirements pinned to version ranges. Some dev-teams forget, and their apps won’t work out of the box. Sometimes, software still works ten years later, if they only use the most common arguments and commands from the packages.
Snaps <==> Virtualenv.
I agree with a lot of your points but I do think containers a great solution.
I’ve been a really big fan of Universal Blue lately. It presents a strong argument for containerizing everything. Your core is immutable and atomic which makes upgrades seamless. User land lives in a container and just gets layered back on top afterwards.