Sad to see GNU Guix struggling, they are young though and I have hopes for the future. About a year ago I made a choice between trying out GNU Guix and NixOS, since this week I now run NixOS on all my servers and desktops after gradually phasing out my old Ubuntu stack.
It has generally been smooth sailing, with the occasional hick up that required me to draw upon my general Unix skills. What caught me off guard is how having a solid configuration to revert to has made me start tinkering with my operating system again akin to how I did when I started using Linux/BSD in the early 00s. But, enough about NixOS, I hope GNU Guix continues to make progress as there needs to be more competition in this space. If you have an evening to spare, I highly recommend trying a distribution like this out in a VM to get a taste of what it feels like to use it – it is, different… but in a good way.
Something I couldn’t get about nix by skimming the documentation is how it manages application configuration.
Using traditional systems when you upgrade the package responsible for the network configuration my current configuration is carried thru or merged into the new format if the package changes it’s configuration syntax. How does nixos manage /etc ?
For the majority of changes, it "just works". For example, here's the part of nixpkgs that configures wireless networking via wpa_supplicant, as well as the actual package build of wpa_supplicant:
Because they're contained within the same git repo, if a new version of wpa_supplicant changed the syntax of wpa_supplicant.conf, the upgrade could be handled easily (assuming the user-facing options in NixOS didn't need to change).
For more complicated changes, there's `system.stateVersion`:
An example of this would be if NixOS version A ships with Postgres version X by default, while NixOS version B upgrades to Postgres version Y. stateVersion allows a sane upgrade path where you can update to NixOS version B without breaking your database immediately (because stateVersion is still A).
I don't know that much about Guix, other than "super-free Nix with scheme" (and the article wasn't particularly illuminating other than boiling down to "Guix isn't really ready for most users right now"), but it's neat that there's momentum in the declarative OS-space. I hope that both Guix and NixOS keep growing and that more people start switching.
After switching to NixOS, I'd find it very frustrating to go back to an OS with imperative-style package/service management. It's clearly the better way.
A frustrating thing about NixOS--coming from Arch Linux and being used to the AUR--is that adding packages/updating packages can be a bit slow due to the sheer volume of PRs/Issues on the nixpkgs repo. There's always the NUR[1] but it hasn't picked up that much momentum yet (probably because it's kind of hard to search the darn thing). Eitherway, the packages are just Nix expressions and repos are just collections of these expressions, so nothing's preventing a real AUR alternative other than momentum.
It's interesting that Guix uses scheme instead of a home-baked configuration language like NixOS. While I don't find Nix particularly offensive, I don't find it particularly great either. And maybe this is naivity/Nix-inexperience on my part speaking, but I really wish that Nix had a proper type system and was strictly typed.
I guess my pipe dream is "HaskellOS" that's basically NixOS configured with Haskell (just think of xmonad-style configuration for your whole OS). That'd be really nice. And my secondary-pipe dream is a baked in, extremely-comprehensive home-manager[2] so we can bid dotfiles goodbye once and for all and have your entire system-state defined in a single file.
> and the article wasn't particularly illuminating other than boiling down to "..."
Yeah, I usually don't like writing these kinds of things. When reading about a distribution, especially a new and interesting one, one (or at least I) tend to forget what could go wrong. This text (I intentionally avoid using the word "article") is a reminder to stop, think and consider if one is up for the challenge. As in my case university just started last week, my conclusion was I didn't have the time.
> It's interesting that Guix uses scheme instead of a home-baked configuration language like NixOS.
I consider that to be one of Guix selling points. An established language, with many independent libraries make it a joy to work with (assuming a set up environment). From what I head, that's not really the case with Nix, although I understand the lure of a type system.
There are significant advantages to having the entire Nixpkgs package set in a mono repository. While I think Flakes are quite interesting for a number of reasons, I hope it doesn't result in the primary package set getting split up too much. They conceptually act a lot like Git submodules, and going overboard with that is a very easy way to make your life miserable, and significantly increase development turnaround for changes that "cross" package sets. A group of packages just being "conceptually related" isn't a high enough bar to move things out.
As a long time contributor, I'd much, much rather see our bug reporting, triage, and discovery/resolution strategies improve first -- long before going as far as splitting up the mono-repository. That's the part that's frustrating for users (their reports get lost, forgotten, duplicated) -- not the fact the default set merely has a lot of packages.
I think at very least (and easy to implement) nix needs a bot that closes old tickets. The number of open issues is ridiculous and basically it makes easier for an issue to be overlooked and further increase the count. The number of open issues also causes some people think nix is very buggy and not suitable in production or even exploring.
There are probably some people who see a 3k bug report number and think that, but realistically tons of projects have thousands of open tickets for any number of things from legitimate bugs to duplicates to actual user questions. Linux, projects like LLVM, etc. It's natural when you have a project that is in active every-day use by literally thousands and thousands of people. Nixpkgs is simply a big project! But the actual numbers might tell another story.
For reference, the LLVM bug tracker has exactly ten thousand open bugs as of this writing, and I assume that number is only so perfectly rounded because Bugzilla search refuses to return any more results in a single page that already brings scrolling in Chrome to its knees. You won't find any lack of praise for LLVM's high quality and extensible nature, though -- they can all be true at once!
The nixpkgs label tracker isn't highly accurate since auto-labeling only started earlier as of this year, but as of this writing, of the 3k open issues, about 500 of them are directly marked as bugs, while 41 are marked as regressions. There are 200 enhancement requests, and 50 open packaging requests. So of 3,300 bugs, about 800 can be quickly classified, and about half of those are actual bugs. This is a very ballpark number (GitHub search doesn't allow XOR). Considering nixpkgs is literally shipping thousands of other software projects, I think this is pretty good -- even though I wouldn't argue it could be cleanly extrapolated to the whole set of bugs, because of a long tail of old bugs.
In terms of cadence, over the last month there were ~466 new PRs, and ~1,600 merged PRs. ~220 closed issues vs 215 new ones. This is done by nearly 350 contributors. I would think this is also very good and the sign of a healthy project with hundreds of active developers. IMO, if just a raw number of "3,000 open github issues" has to mean anything ("this software is unstable"), then these numbers have to mean something too. It's just that GitHub doesn't "helpfully" post them on every single page concerning your project...
But I do agree we should have better mechanisms for discovery and keeping things clean. Duplicates certainly happen and are annoying, and nobody likes drudging through 1,000 of the oldest open reports to close old stuff, etc. Some of this is also limited by GitHub and other things, unfortunately.
Personally as a developer (not a user!), I think bots that close old tickets are often very annoying (to everyone) and rarely do much other than feed a human desire to hit "inbox zero", but this desire is not (to my knowledge) based on any actual scientific criteria that leads to better software development or happier, better cared for users. It is more a way of externalizing a deep human desire for "cleanliness" and using it as a yardstick for "quality" rather than evaluating the numbers for yourself. You can also find tons of 0-issue repositories on GitHub, but this doesn't mean they're actually good, or they do what you want -- and I doubt most of them have as much momentum as we do.
There are, that was response from one of my colleague after mentioning Nix to him. Having 3k open bugs, don't help with anything there's no way all of them will ever be addressed and important ones could get lost between them.
I don't think closing such bugs is "feed a human desire to hit "inbox zero"" even with that you'll never get there. From my perspective it is all about not losing track of important issues. Having bots close tickets makes sure that only tickets are opened are ones that someone cares about. There are tickets that are from 2013, a lot changed since then, they might no longer be relevant anymore.
I personally hate it when a bot closes an issue I wrote. What's the point of closing unresolved issues? Just assign them a label (it could be a "stale" label like some projects do). Otherwise it just feels like hiding your mess under the bed. Other sorting tools can provide the same list as automatically closing bugs would achieve. The opposite isn't true, unless assigning a specific label when manually closing an issue.
Guix in general just needs a lot more contributors and lot more work to come anywhere close to being a competitor to the mainstream (or even near-mainstream) distros and package managers.
I'm coming from Gentoo myself, and I've tried to get Guix working on Gentoo a couple of times over the years, and have always been frustrated by how much work and hacking was involved in doing so. I'm no Linux novice, either. I've been a professional Linux/UNIX admin for decades, I know and love Scheme, and am no stranger to putting in a lot of work to get something to run on Linux, but I failed in two major attempts with Guix, despite having lots of thorough assistance from Guix developers on #guix on IRC.
So back to Gentoo's package manager I went.
A typical user would not have put in nearly that much work to trying it out, nor would they have had the skills or knowledge to get nearly as far as I did. Though, to be honest, I did have some requirements (like doing my own compilation of everything on my own old, slow laptop, just as I'd been doing with Gentoo, and not trusting their binary distribution) that a typical user would not.
The other lesson I got from my last attempt was that trying to compile everything in Guix from scratch on an old laptop is just way too slow. Doing this on Gentoo is slow and painful enough, but it's way worse on Guix, and simply unworkable. You really have to have a fast, modern system (or offload your compilation to another machine which will crank away for weeks on end and that you can then download the binaries from) to constantly compile everything you need for Guix.
That said, Guix has really come a long way over the years, and it's impressive to see how far they've come, and I'd absolutely love to see them succeed, but they've got a long, long, long way to go.
> trying to compile everything in Guix from scratch on an old laptop is just way too slow. Doing this on Gentoo is slow and painful enough, but it's way worse on Guix, and simply unworkable
Would you elaborate on why you think compilation on Guix is slower than Gentoo? Thanks.
It's not that compilation itself was slow with Guix, but (to my memory) it was because all Guix packages are completely separate by design and have dependencies which are not shared, so you might have to compile, say, five different versions of guile to satisfy those five different dependencies in five different packages.
Compiling guile from scratch was, by the way, insanely slow, and because I struggled to get Guix working right on my first try, I had to compile it over and over again, which took about a day each time. A "guix pull" took four hours on my system, and I had to do it twice each time, once for root and once for my own user. My first "guix pull" took two weeks due to all the errors, toubleshooting and restarts the process took as I tried to get Guix working as it should.
In addition, it seems that Guix packages don't have optimizations enabled, which is a deal breaker for me, as optimizing the packages I install is a significant reason I compile them from scratch in the first place, as I'm running on an old, slow machine which could use all the performance boosts it can get.
Finally, I might have a broken environment on my system which makes compilation even slower still than it otherwise would be, but until I get it fixed I have to live with it, and so for all of these reasons compiling everything from scratch on Guix is just not practical for me right now.
Maybe with a newer, much faster machine I'd be ok with it.
I understand and perhaps admire the GNU project's insistence on being blob-free, but that also means Guix won't run out of the box on 99% of all people's computers. The options are buy (old) hardware that doesn't need blobs (which sometimes isn't even possible), or compile the default kernel to include the binary firmware yourself (which the Guix documentation won't explain for ideological reasons). Both of these will sadly limit the appeal of this distribution to enthusiasts-only, which is unfortunate since Guix probably has the most advanced package manager of any Linux system.
Right, but now you have two separate operating systems to manage which are different at the kernel level (as your host is probably linux, bsd, esxi, etc.) This means you often have to manage twice the amount of tooling, documentation, scripts, etc. and makes it impractical for many cases.
There's already hundreds of distributions with binary blobs. It is very important that at least a few respect user's freedom. So it's great that Guix exists and that it has a serious stance against secret blobs.
I would recommend the author to try NixOS instead. It is not like that you wouldn't have issues however mostly things works (never had the kind of issue the author reported in this post, for example).
I think the difference between NixOS and Guix, excluding the language of choice of course, is that NixOS has the community to make it work. For example, I had a issue last night that I opened a issue in GitHub, and in a few hours the maintainer of the problematic package answered me and the problem was fixed.
I'm hoping you're joking, because petition for him to step down, because he took part in with hunt to have himself removed sounds ridiculous. I'm still interested in a real answer.
Thank you for shedding light on a distro I didn't know existed!
One tip: The contrast on your code blocks is terrible with white text against a light white background. Consider changing the `background` of line 3 of style.css to `purple`, `darkorange`, or some other dark color.
Oops, sorry for that. I don't have a browser that supports dark mode, so I just tried something that I thought may work. Then I changed some CSS earlier this morning that probably messed something up.
It's broken for dark mode due to a `body { color: white }` rule in style.css. The code block background is not overridden so it remains light yellow while the text is white.
I love the style of packets organization that they describe. I think it has a name, I have seen it promoted as an alaternative to the typical mess that is "install everything in various parts of /usr/ and /opt/, put things in /etc/ or ~, silently overwrite old versions"
It has generally been smooth sailing, with the occasional hick up that required me to draw upon my general Unix skills. What caught me off guard is how having a solid configuration to revert to has made me start tinkering with my operating system again akin to how I did when I started using Linux/BSD in the early 00s. But, enough about NixOS, I hope GNU Guix continues to make progress as there needs to be more competition in this space. If you have an evening to spare, I highly recommend trying a distribution like this out in a VM to get a taste of what it feels like to use it – it is, different… but in a good way.