Why not extend to Fedora? At the risk of starting a distro religious war…
After many years away from the RedHat ecosystem I recently tried to stand up Fedora 37 and 38 to test it with Willow Inference Server.
After hours and hours of poor and conflicting docs, random stability issues, various additional software repos of questionable quality, etc I gave up and proclaimed it officially unsupported.
Why anyone would want to waste their time being a tester for a commercial product is very strange to me when there is Debian and even Arch which (for a rolling release) seems to be substantially more stable, better documented, etc.
I understand why it exists and it has its place but absent a few specific scenarios I have no idea why it would be a first choice for anyone.
The problem with distros is that you're always picking an update cycle tradeoff. Debian has decades old versions of packages that they'll backport security fixes into till pretty much the end of time (RHEL also does this but is well, commercially backed while Debian is entirely volunteer driven, which has opened up Debian to maintainer shenanigans a couple of times).
Ubuntu is more up-to-date by virtue of simply importing the Debian backports repo every 9 months.
Arch Linux meanwhile runs on "whatever the latest upstream is", and with that you have pretty much 90% of the actively used Linux distros covered.
All of these models are untenable for someone who likes to stay up-to-date with the latest version but doesn't want to drop down into the occasional shell hell that Arch provides.
Fedora kinda... sits in the middle of that. It updates to latest version every 6 months, has a generally sane-ish release cycle that follows the most popular scripting language (Fedoras cycle is specifically staggered to allow for python releases to coincide with it every other release). That in turn makes it the best OS to recommend for Desktop Linux if you're new to Linux.
As for why not Arch/Debian/Ubuntu for beginners: Arch requires wanting to maintain the OS for the sporadic breakdown or strange default setting (not to mention the shell being an expected skill beyond running package management commands), Debian is too out-of-date to be useful and Ubuntu is only slightly better than Debian.
> Fedora kinda... sits in the middle of that. It updates to latest version every 6 months
It's even more in the middle, Fedora can update to the latest versions of packages as long as its not a major package and its an fairly standalone package. Fedora pulls new kernels as an example.
It's release timetable also suits GNOME, Fedora comes out about a month after every GNOME release, with that version of GNOME.
This middle ground of sorts is what I was alluding to with "a few specific scenarios". From what I recall Fedora is the distro of choice for Linus? So it's clearly not completely terrible and I never said it was. That said, the creator of Linux is about as far as you can get from a typical/average user...
Per usual HN there are plenty of conflicting anecdotal reports and wildly differing perspectives on my position here and as I noted in another reply I absolutely could not care any less when it comes to people selecting the best tool for their purposes/opinion/experience/worldview. Not being steeped in modern Red Hat (I have an RHCE cert for RedHat 8 around here somewhere) but having decades of experience across many other distros the difference in experience as a "new user" with Fedora was striking. Also note that "new user" isn't exactly accurate here either - the install process was great, the UI is clean, etc. However, as soon as I got off the beaten path the quality of experience (again, for me personally) dropped to among the poorest I've seen in years with Linux distros.
Back to another anecdote for me, after decades I can count on one hand the number of times I thought "wow I could really stand to have a newer system python release" (as one example) - even when using pretty long in the tooth Ubuntu 20.04.
I'm not saying that doesn't happen and doesn't have it's advantages. It's a good point generally but I'm as confused as I am because it's never been even remotely close to a deal-breaker in my decades of doing all kinds of random stuff (from embedded to desktop to server to cloud). I don't know what people are doing to call them "too out-of-date to be useful" but again that's just me - I'm sure you and others have plenty of reasons for this position. I've just never encountered anything close to it.
Python specifically has such famous and nightmarish packaging and dependency issues I more times than not just throw everything into whatever official release Docker container for the version I need and call it a day. Great? No. Hacky? Yes. Practical? Absolutely.
Just a quick note: debian stable has a new release every two years. So in the very worst case "decades" means like three years. The horror, the humanity, just remembering how software used to be three whole years ago causes me to shudder.
That can still be ages depending on the maintainer spats.
When I say maintainer spats, I'm referring to shit like how avconv was given undue credence for ages because the split happened right around a Debian release and the packager for ffmpeg on Debian was on the avconv side, which caused a false deprecation notice to be present for quite some time.
And yes, with how quick some software moves, two to three years can very much result in a mess of hacks and patches.
> just remembering how software used to be three whole years ago causes me to shudder
Well, yes. I agree with that statement even though you made it facetiously. Backporting security fixes only for bugs that have made enough noise to warrant it is a horribly janky hack. And, just thinking about all the time that has been, in my opinion, wasted on custom code that Debian maintainers have had to write in order to backport fixes makes me shudder.
Eh, there is huge demand for that static-ness, see windows ltsc, red hat, suse, debian stable, ubuntu lts etc. . Those customers actively don't want unstatic releases and pay lots of money for people to "waste" their time. So it doesn't matter whether it is a "janky hack" or what your personal opinion on that is.
To be clear, I have nothing against them doing that, I'm just saying that this is how Debian generally operates. Their extended support model still allows for backporting fixes into jessie of all things, which released almost 8 years ago.
It's really good that they do that, I respect the niche they work in, but Debian has the nasty knock-on effect of having a lot of guides made for it by third parties that recommend really outdated versions of debian which can be a nasty suprise for someone new to Linux.
(It also results in notable forks like Raspbian having a poor update cycle - I remember in particular that it took about 2 years to move from the regular EOL of wheezy to jessie.)
> Their extended support model still allows for backporting fixes into jessie of all things, which released almost 8 years ago.
So your evidence that they don't update things is that they backport new fixes to an EOL version? Even if that made sense, 8 years is 0.8 decades; I think a lot of the pushback is that you said "decades old versions", which is to say "versions of packages that are at least 20 years old", which is wildly out of line with reality.
If that is what you meant in your original post, then you presented it in a very manipulative, dishonest way. You alluded to Debian only offering decades old software versions and nothing newer, while the truth is that by default, users get fairly recent versions (e.g. KDE/Plasma is only one point release behind upstream), and one has to go out of their way and jump through hoops to get a version of Debian with software that is several years (not even a single decade) old.
So you tried to run your app on it and ran into trouble, and you conclude that Fedora is an unstable hunk of shit?
Also, this attitude that Fedora is just a tester for a commercial product is absurd. I suppose this is the same logic that makes CentOS Stream just a "beta" of RHEL (but strangely, these same people don't consider RHEL a "beta" for old CentOS even though consistency would demand this).
Anecdotally, I've used a lot of distros over many years, and Fedora has been by far the most stable and "just works" of any of them. There were a couple of rough years with the Nvidia driver, but I blame Nvidia for that.
My conclusion was, essentially, that if I can't get our application to work with Fedora in a way that is practical, approachable, and sustainable for our "average" user it should likely be unsupported. I should also add I was working with at least one person who has a tremendous amount of experience with Fedora and they were taken aback by the whole experience as well. I should also note our application, specifically, (more of a long term support appliance) would likely represent an undue support burden because of the fairly rapid Fedora release cycle. As a random tangent, how is DNF /still/ so ridiculously slow? I don't get it.
Ubuntu/Debian is our main development platform so of course it "just works" there (in addition to various other Debian/Ubuntu derivates). However, when I repeated the same exercise for Arch (as one example) - a distro I have zero experience with - the process was completely painless and we had it implemented and documented in something like 30 minutes. Even though it is a rolling release the Arch wiki has substantially better documentation alone than anything I could find in the Fedora ecosystem (random hodgepodge blog posts for the last three major versions).
Again, I've been using Linux near-universally since 1997. I've had these debates (religious flamewars) plenty of times before and I have no interest in rehashing them. That said I don't know how anyone could debate my central point (let alone call it "absurd") - that Fedora is a testing ground for a stable LTS, commercially supported product (and one that with recent events is even more so). This is all right there, clear as day, in the official documentation[0]. So maybe not "absurd"?
I suppose I just don't understand the general preference for Fedora (other than momentum and familiarity) when there are plenty of other choices around where the primary goal is a free, stable, and open release.
On that note - IMO the best thing about open source is choice. Use whatever works for you, why would I care? I do not.
I agree with this assessment. Happy Fedora user here, but I am a bit nervous about the I Mification of RedHat and the potential long term ramifications for Fedora.
Same, I distro hopped for a while across various ubuntu and ubuntu derivatives, always ran into weird issues, stability problems, etc (and still do at work where we use ubuntu). Installed fedora eventually and have been using it ever since, outside of a packagekit bug like 7 years ago that didn't clear the cache and ate my disk space, it's been pretty smooth sailing.
Yes it was a "clone", but it was "downstream". In fact Alma themselves just a couple days ago in their post pointed out the CentOS Stream is "upstream" while Alma is "downstream" of RHEL. CentOS always lagged RHEL by a little bit of time. Big security things it was usually days, but other times could be weeks or even months on new minor version increases. The only way for it be just a clone would be to near-instant rebuild, which has never been the case.
Fedora has been one of the most stable distributions I've ever used (I used it as my main system for ~6 years, but moved to another distribution now for unrelated reasons). It's also mostly developed by the community — Red Hat provides the infrastructure, but not much actual development.
After many years away from the RedHat ecosystem I recently tried to stand up Fedora 37 and 38 to test it with Willow Inference Server.
After hours and hours of poor and conflicting docs, random stability issues, various additional software repos of questionable quality, etc I gave up and proclaimed it officially unsupported.
Why anyone would want to waste their time being a tester for a commercial product is very strange to me when there is Debian and even Arch which (for a rolling release) seems to be substantially more stable, better documented, etc.
I understand why it exists and it has its place but absent a few specific scenarios I have no idea why it would be a first choice for anyone.