Distribution quotes of the week
Distributions are also working on innovative projects at the scale of the entire software ecosystem, and are dealing with bigger picture things than you need to concern yourself with.— Drew DeVault (Thanks to Paul Wise)[...] There are several areas of open research, too, such as reproducible builds or deterministic whole-system configuration like Nix and Guix are working on. You can take advantage of all of this innovation and research for the low price of zero dollars by standing back and letting distros handle the distribution of your software. It’s what they’re good at.
It's easy to be part of a community when everyone agrees. It's powerful and delightful to be part of a community when people disagree but the community still works together with respect and mutual support. Creating process that allows myself and others to do this more easily is part of how I enjoy contributing to a community.— Russ Allbery
Posted Sep 30, 2021 8:55 UTC (Thu)
by hadess (subscriber, #24252)
[Link] (9 responses)
Can somebody make a list of Drew DeVault's applications (not replacements for OS components, applications) he's been shipping in distributions? Because I'm guessing he hasn't seen them not receive any updates ever after a distribution's release.
I'd ask him on Twitter, but his account was suspended.
He couldn't get on Arch's packaging team, so I wonder how much first-hand-experience he has of distributions: https://lists.archlinux.org/pipermail/aur-general/2019-Fe...
Posted Sep 30, 2021 12:33 UTC (Thu)
by qyliss (subscriber, #131684)
[Link]
He works on Alpine Linux, I believe.
Posted Sep 30, 2021 12:48 UTC (Thu)
by ddevault (subscriber, #99589)
[Link] (6 responses)
A lot of my end-user software ships in many distributions. Sway and wlroots probably moreso than any other:
https://repology.org/project/wlroots/badges
My Twitter account was never suspended. I closed my account and it was later re-registered by a spammer. I was also not rejected from the Arch team. I was not enthusiastic about joining in the first place until a friend convinced me to (after several years of badgering), and ultimately I withdrew my application before any kind of verdict was made.
I would appreciate it if further ad hominem arguments were kept off the table on LWN. If you have substantive feedback on the article itself, I would be pleased to hear it. Cheers!
Posted Sep 30, 2021 13:33 UTC (Thu)
by rahulsundaram (subscriber, #21946)
[Link] (3 responses)
Alright, so let's try that:
As someone who has worked in the past, within the distribution world for well over a decade, the notion that all you have to do is put out a tarball is a limited view on things. If multiple different solutions exists for years and years (Appimage, Snap, Flatpak and even before that zeroinstall, Linstaller, Autopackage), it is a good idea to not dismiss it offhand as "missing the point" and atleast attempt to understand what is the problem that these attempt to address and why even people who work on the distributions themselves come up with these solutions. Then you can talk about the tradeoffs (several of these are highlighted for example in within Flatpak pages) and still come up with the same position that you do but if you don't acknowledge these tradeoffs, the discussion isn't well rounded.
Posted Sep 30, 2021 13:50 UTC (Thu)
by ddevault (subscriber, #99589)
[Link] (2 responses)
The postscript about Flatpak was a postscript. Yes, it was flatly dismissive. It was a single-line postscript at the end of a tangentially-related blog post, and at no point was intended to be, nor was it portrayed as, a comprehensive argument on the subject. I may write more about Flatpak in the future, but that's not what this article was, and I'm quite entitled to write about things which are not detailed rebuttals of the Flatpak-style packaging ecosystem.
Posted Sep 30, 2021 14:03 UTC (Thu)
by rahulsundaram (subscriber, #21946)
[Link] (1 responses)
Just to be clear, I haven't insulted you, this seems to imply I did.
>You've alluded that there's some kind of information out there which I should have acknowledged and contrasted with in the article, but you've not specifically stated what any of that information is
I am deliberately asking you to think about the disadvantages of letting the distributions package everything and acknowledge them instead of overlooking the reasons why other models exist entirely. If you can see no reason at all why those exist, that would be interesting to know as well by itself.
Posted Sep 30, 2021 14:14 UTC (Thu)
by ddevault (subscriber, #99589)
[Link]
Sorry, I had mistook you for the person at the top of this thread. I apologize for the mistake.
>I am deliberately asking you to think about the disadvantages of letting the distributions package everything and acknowledge them instead of overlooking the reasons why other models exist entirely. If you can see no reason at all why those exist, that would be interesting to know as well by itself.
Well, this post was not aiming to do that. I wasn't setting a mock debate. I'm arguing for my views as I see them, and explaining the system that I work in, rather than trying to dissect some of the upstarts in the space. The purpose of this post was not to contrast this model with others, but instead to speak to its merits in their own right, since it seems many have forgotten them, or never really understood them in the first place.
Some kind of article more specifically addressing Flatpak et al might be warranted in the future, but I really have very little to say on the subject which is not derogatory, and I have been aiming to craft a more positive tone for my blog as of late. I think this post stands perfectly well upon its own merits regardless.
Posted Sep 30, 2021 13:36 UTC (Thu)
by hadess (subscriber, #24252)
[Link] (1 responses)
Posted Sep 30, 2021 13:41 UTC (Thu)
by ddevault (subscriber, #99589)
[Link]
Posted Sep 30, 2021 13:32 UTC (Thu)
by jkingweb (subscriber, #113039)
[Link]
Different distributions have different priorities and tradeoffs, and all have _some_ lag time between upstream software being released and upstream software being packaged, not to mention they package different things. Flatpak eliminates some of those tradeoffs by introducing others. A one-size-fits-all system will never actually suit everyone, thus we have options other than the monolithic software distributor.
The pot migth be calling the kettle black, here.
Posted Oct 26, 2021 21:33 UTC (Tue)
by mathstuf (subscriber, #69389)
[Link]
The issue is when you have users that end up running stable distros (like Red Hat) that still want to run your brand-new software (because it matters to them). Red Hat (nor EPEL) is going to chase your software in their repos. So vendoring is the way to go (because you have to do it for Windows and macOS anyways). Plus it allows for a single build to be viable for multiple distros (by being just a simple tarball extraction and building against old enough libstdc++ and glibc versions).
However, you can do vendoring "right" and not conflict with existing installations. The list I go through is basically:
- mangle library names (to avoid loading the wrong one at runtime)
Yes, it's a lot more work, but it also allows us to actually get things done today. Note that this is in addition to an option to use an external copy. Granted, the software I'm thinking of isn't 100% on this (some libraries have yet to be mangled or support external copies, but these are either "not shipped anywhere" or way too complicated to manage (though I do push efforts to Do It Right when I can).
I strive for the "use distro copies" when possible (and work with packagers when they show up), but not everyone runs Fedora or Arch and actually has those deps that are new enough all the time :( .
Distribution quotes of the week
> P.S. Systems which invert this model, e.g. Flatpak, are completely missing the point.
Distribution quotes of the week
Distribution quotes of the week
https://repology.org/project/sway/badges
Distribution quotes of the week
Distribution quotes of the week
Distribution quotes of the week
Distribution quotes of the week
Distribution quotes of the week
Distribution quotes of the week
Distribution quotes of the week
Distribution quotes of the week
- relocate headers (to avoid finding the wrong thing at buildtime)
- mangle symbols (to avoid conflicts in case some other component loads the "real" one in the same process space)