Ubuntu Developers Seem To Be Really Pursuing ZFS Root Partition Support On The Desktop

[starshipeleven]

You are wondering over one fundamental problem with Linux, which could be described as "start with a hurrah' , lose interest half-way through without it ever getting completely finished, start something similar anew, repeat".

I was under the impression that this is a common problem with human endeavors in general.

But I'm trusting you blindly if you say that in BSD land people is completely different and they always finish what they started, they never abandon the project mid-way or develop it so fucking slowly that any potential user loses interest.

That's why BSD-licensed software rules the world... right?

'take, take, take' is largely your imagination.

It's written right there in the license, that they can take take take for free.

[rleigh]

Its kind of odd people seem to be so interested in ZFS when Linux has its own GPL filesystem called btrfs which works very well. Why doesnt Canonical start acting like they are not insane and help improve btrfs? You probably cant legally or morally do ZFS in a distro because of the licensing. I am not fond of BSD licenses because companies should give back their improvements rather than take take take.

Well, a big part of the problem is that Btrfs doesn't work very well. People like me are using ZFS because it actually works as intended. The performance, service interruptions and data integrity problems of Btrfs were unacceptable. And I say this as someone who used Btrfs intensively from the start for many years. ZFS is what Btrfs should have been had they done the design and implementation correctly from the start.

As for helping to improve Btrfs, that's absolutely an option, if you want to throw a lot of good money after bad. If the underlying design and implementation have multiple flaws, this may be an impractical approach. However, funding the integration work to make ZFS work seamlessly with the installer and the rest of the base system is a tiny fraction of that. Neither of us are likely in a position to make these decisions, but if you were a Linux distributor, the cost/benefit might well dictate the viability of supporting one over the other, and ZFS is in a much better place than Btrfs.

Your comment about not being fond of BSD licences makes no sense. ZFS is CDDL licensed, not BSD licensed. It's a weak copyleft with much in common with the GPL which requires you to distribute your changes pretty much the same as the GPL does. If you like the GPL, it's hard to fault the CDDL for requiring the same thing!

[aht0]

I was under the impression that this is a common problem with human endeavors in general.
But I'm trusting you blindly if you say that in BSD land people is completely different and they always finish what they started, they never abandon the project mid-way or develop it so fucking slowly that any potential user loses interest.

Mentality appears slightly different for me between users of two OS families with drastically different patience levels. You could describe it as one sort likes to 'hack away and see what comes out', other likes to 'build'. In BSD progress could be slow but usually shit ends up finished. Instead of being abandoned the moment something else strikes dev's fancy. Sometimes 25%, or 50% or effin' 95% done.

I'll give you few examples:

• BTRfs as a project was started with huge fanfares, bunch of devs, multiple CORPORATION's participated. Everybody, their dog and cat cheered over 'future-ZFS-killer'. 12 years later it's still only 'mostly-done', where not even BTRfs own wiki does not dare proclaim it as a 'stable'. Even worse, some of the initial backers, like Red Hat have seemingly dropped idea of adopting it completely - it's indication RH considers BTRfs a 'lost cause'. VS. Matthew Dillon started Hammer early 2007, worked out design in 9 months and finished it up by 2011. Alone. Then announced plan for better-designed successor and got it done by last year. Despite the multi-year stall in-between he finished Hammer 2. Again. alone.
• Recent Phoronix article "VK9 Project Stalls As Developer Leaves To Pursue Other Interests". Does it feel familiar behavior?
• When KDE3 became finally digestable and actually pretty darn good it was suddenly axed and KDE4 started from ground up. Pattern was repeated with Plasma5. Each time between EOL'ing the old and 'reinventing the wheel', the 'then-new-wheel's initial versions were as a rule rather buggy crap. Each time it took like 7+ point-releases before new KDE would be stable enough for daily use.
• Mir. "We do our own thing". Then axe it. Then use some of it. Waste bunch of years and developer hours while doing it.
• Unity, pretty much repeat pattern. 6 years of wasted efforts. Both read to me as 'clueless flailing around'..
• Wayland display protocol, regardless of reasons, it looks and feels like an flippin' endless road. X still reigns supreme, 10 years after Wayland's initial release, it's been employed in one distribution, Fedora. Ubuntu previous release used Wayland as well but reverted back to X for it's latest.

That's why BSD-licensed software rules the world... right?

Windows software does rule the world. From user's POV.

It's written right there in the license, that they can take take take for free.

[/QUOTE]
It comes down to want. If you want, you can use loopholes in GPL and take it for free too. Many companies do it.

Let's say your changes are not proprietary/business secret, isn't it cheaper long-term to upstream? There's your incentive right there: avoid extra work of endless patching as new releases of one or other BSD os appear. All the major corporations contributing back to BSD do it for that exact reason, it's economically more feasible in long-term.

[Weasel]

So, why didn't ZoL actually go through with this project? Until they change the license so they can be included in the kernel then yes it is still their own problem.

Maybe you like to work months for nothing. Most people don't because they value their time.

Who would give them their months back if the kernel devs still rejected or ignored it? You?

[gmturner]

Mentality appears slightly different for me between users of two OS families with drastically different patience levels. You could describe it as one sort likes to 'hack away and see what comes out', other likes to 'build'. In BSD progress could be slow but usually shit ends up finished. Instead of being abandoned the moment something else strikes dev's fancy. Sometimes 25%, or 50% or effin' 95% done.

I think this is like saying, look at this pattern, people's hearts always seem to stop beating and they die, or, look at all these buildings falling apart, that's a pattern in buildings.

Some software projects might theoretically be immortal; I will get back to you on that at the end of time. But this isn't capturing a salient social or institutional dynamic, it mixes up correlation and causation, and it ignores the life-cycle of open-source projects, which can benefit us either by being useful now, or later, or both, and dead projects may have been useful in the past.

Here's my personal take on your list:

• BTRfs: hard for me to say what went wrong here, but creeping-featurism may have played a role if I had to guess. Also, it does continue to slowly improve, maybe there is hope for btrfs yet? Maybe a lot of us may have some sort of "ptsd" from trying this in production too early. I know, for me, any tiny problem with btrfs just seems like the sky is falling and I just didn't want to feel that way anymore, so I stopped using it.
• VK9: dxvk took away all its potential market-share, leaving a lone developer who felt like he was just in a room by himself writing code
• KDE{3,4}: In both cases, the replacement projects provided migration paths and ultimately exceeded the value of their predecessors. There was some pain along the way, we must admit. They have even softened their stance on the crouton or whatever that stupid thing was on your desktop that they wouldn't let you get rid of.
• Mir: No clue/knowledge, I simply can't speak about it intelligently.
• Unity: Kind of the Windows 8 of the Linux desktop. A DE experience/paradigm that, arguably, most users simply didn't like or want, at least at first. This seems to more nearly conform to the pattern you're describing, aht0, than the others, as, like Windows 8, it was indeed abandoned before the dark cloud of negative initial reception was given time to dissipate.
• Wayland: wayland is definitely not dead but it's a huge project with a very high critical mass in the sense that end-users aren't going to want to migrate until almost every bug and limitation is fully ironed out. Perhaps wayland should have provided more of an evolutionary migration path (although this is a hard problem in itself, given that the idea of wayland is kind-of to make a clean break from legacy encumberments). I see wayland advocates bad-mouthing every fork or client-project that attempts to fill the gaps in the reference implementation. Now that's a real anti-pattern in open-source projects, everyone should stop doing that.

So is this a pattern or just a sampling of open-source software projects with no real pattern, aside from, you know, being software?

Where is the 100% done open-source software project? And even if you've found it, is it reasonable to expect all open source software projects to reach 100% completion? How should they go about achieving it?

TLDR: the failure of projects is a necessary part of how software can deliver value. Turning around and calling it a pattern tied to linux or even open-source seems unfair.

[Guest]

I like HAMMER2, I've no problem with that but you'll need to get it by religous Linux zealots like yourself because it's got the wrong license for you. "GPL onry ppl"

I happen to *like* CDDL/MPL licenses. Have you read it? File based, weak copyleft, flexible and still requires changes to be public. It's the good parts of the GPL and not the bad.

I do not wish to be associated with Linux, if anything, I feel like the ship is sinking and the kernel will start becoming irrelevant. Microsoft seems to be stagnating, and OSes with Linux might start taking over. If anything, call me a GNU GPL zealot, as that's what I care about. While I used to like Torvalds for his attitude, his support for "open source" made me dislike him.

As for CDDL, I don't even care about the content of it, as the issue is the lack of compatibility with GNU GPL. It doesn't even matter whether it provides all 4 freedoms, it's effectively as bad as proprietary software because it's incompatible with GNU.

[aht0]

So is this a pattern or just a sampling of open-source software projects with no real pattern, aside from, you know, being software?
Where is the 100% done open-source software project? And even if you've found it, is it reasonable to expect all open source software projects to reach 100% completion? How should they go about achieving it?

I presented you pretty much random list of projects. You can choose to take it literally if you wish. I wanted to use that list for illustration.

First let's get clear in base definitions:

• Is it piece of software that has met it's design goals and stays unchanged for years? As, "it works, does it well, don't further fu*k with it". If you wish, I can provide some examples.
• Or is it software that has met it's design goals, it's features are properly finished (fully working) and new features are introduced on occasion? Which also are finished before dev considers adding more.

For me either is 100% done, provided that all features are fully working.

• Software that has set of announced/planned design goals/features but of which some or more are half-finished or not working at all, is not 100% done.
• Software that ..(repeating previous sentence).. or not working at all AND where devs keep adding - is not 100% done by far and is in fact, the absolute worst.

And what I absolutely effin' hate - Lots of Linux-associated software (and many Linux distros themselves) have half-assed AND overcomplicated feel because they belong to the latter category. Introducing half-baked new stuff, then switching directions et cetera is more important for Linux developers ( & dev's developing for Linux) than finishing their work.

To put it short: Linux devs are interested more about feature-creep than tiding up loose ends. BSD devs rather bite less and finish what they had started. That's my subjective POV.

To give you another example of different philosophy: OpenBSD does not allow half-finished new features/components into their OS releases. Wanna add something to OS, finish it first - then it will be added to the next release. How often do you see something like this in Linux software-sphere?

[starshipeleven]

Mentality appears slightly different for me between users of two OS families with drastically different patience levels. You could describe it as one sort likes to 'hack away and see what comes out', other likes to 'build'. In BSD progress could be slow but usually shit ends up finished. Instead of being abandoned the moment something else strikes dev's fancy. Sometimes 25%, or 50% or effin' 95% done.

Dunno, I can't remember of anything developed in BSD that has been finished in this decade. EDIT: ah yes I remembered something, LibreSSL.

Their "systemd-like" init? Not finished yet, many years down the line, maybe not even a major project
Their "layer to run linux drivers" not finished yet, also many years down the line
ZFS? they are just a downstream.
HAMMER2? developed by someone since 2012, ETA for completion unknown

BTRfs as a project was started with huge fanfares, bunch of devs, multiple CORPORATION's participated. Everybody, their dog and cat cheered over 'future-ZFS-killer'. 12 years later it's still only 'mostly-done', where not even BTRfs own wiki does not dare proclaim it as a 'stable'. Even worse, some of the initial backers, like Red Hat have seemingly dropped idea of adopting it completely - it's indication RH considers BTRfs a 'lost cause'. VS. Matthew Dillon started Hammer early 2007, worked out design in 9 months and finished it up by 2011. Alone. Then announced plan for better-designed successor and got it done by last year. Despite the multi-year stall in-between he finished Hammer 2. Again. alone.

This is so fucking wrong.

RedHat was never an "initial backer" in any way, shape or form. They just provided it as a "preview" to see if their customers actually wanted it. Turns out their customers wanted a easy-mode tool to use what is already there, RH went for it.

Regarding HAMMER2, you have no fucking clue. Does something like this https://gitweb.dragonflybsd.org/drag...hammer2/DESIGN

look like:
-finished
-comparable to btrfs

It is a cluster filesystem, which is an entirely different type filesytem. https://en.wikipedia.org/wiki/Clustered_file_system

And yet it can't be used as cluster filesystem, as it still lacks the features to do so.

What HAMMER2 is good at now is more or less what Btrfs is also good/stable at, single drive only, no replication (RAID), no clustering. In both cases it's a far cry from the actual goal of the filesystem.

Recent Phoronix article "VK9 Project Stalls As Developer Leaves To Pursue Other Interests". Does it feel familiar behavior?

VK9 is a library supposed to be run on Windows, or through Wine (which is multiplatform and can run on MacOS or BSDs too).
What the fuck makes it a "linux project" in your mind?

https://github.com/disks86/VK9

When KDE3 became finally digestable and actually pretty darn good it was suddenly axed and KDE4 started from ground up. Pattern was repeated with Plasma5. Each time between EOL'ing the old and 'reinventing the wheel', the 'then-new-wheel's initial versions were as a rule rather buggy crap. Each time it took like 7+ point-releases before new KDE would be stable enough for daily use.

More stupidity. KDE is not a Linux project, it is also running on BSDs, and most of their libraries use LGPL or BSD/MIT licenses.
KDE core (the libraries) allow to make cross-platform applications for Android and even Windows.

But it is a "linux project", because reasons.

--Mir. "We do our own thing". Then axe it. Then use some of it. Waste bunch of years and developer hours while doing it.
--Unity, pretty much repeat pattern. 6 years of wasted efforts. Both read to me as 'clueless flailing around'..

This is Canonical's work. You know the ones that ALWAYS reinvented the wheel when given the opportunity. Hardly big enough to be the flag of all things Linux though.

Wayland display protocol, regardless of reasons, it looks and feels like an flippin' endless road. X still reigns supreme, 10 years after Wayland's initial release, it's been employed in one distribution, Fedora. Ubuntu previous release used Wayland as well but reverted back to X for it's latest.

It's in a pretty decent shape, and for what it actually is, it's not doing as bad as you think. It will be ready to be a default within a few more years as even now it "mostly works".

To the contrary of Btrfs where I can't see the light at the end of the tunnel, and I'm subscribed to their mailing lists too.

Windows software does rule the world. From user's POV.

That's completely tangential to the point. Proprietary software also does exactly the same as your "idea of linux software".

It comes down to want. If you want, you can use loopholes in GPL and take it for free too. Many companies do it.

It's objectively harder to exploit GPL than working with a project where you have no such limitations to begin with.

Let's say your changes are not proprietary/business secret, isn't it cheaper long-term to upstream?

Maybe? It all depends on how much you care about tracking closely upstream, and how complex it is to upstream.

Let's not forget that upstreaming something that "WorksForMe"(tm) isn't a painless process, especially if upstream disagrees with your "time-sensitive coding practices" and calls them "horrible hacks" and NAKs them.

[k1e0x]

I do not wish to be associated with Linux, if anything, I feel like the ship is sinking and the kernel will start becoming irrelevant. Microsoft seems to be stagnating, and OSes with Linux might start taking over. If anything, call me a GNU GPL zealot, as that's what I care about. While I used to like Torvalds for his attitude, his support for "open source" made me dislike him.

As for CDDL, I don't even care about the content of it, as the issue is the lack of compatibility with GNU GPL. It doesn't even matter whether it provides all 4 freedoms, it's effectively as bad as proprietary software because it's incompatible with GNU.

I think the GPL maybe was important in the 90's when Microsoft and other propitiatory software looked like it was going to rule the earth.. but I feel business gets it now and want their code to be open so they don't have to maintain a fork. Due to that I like more business friendly licenses. It's the terms of the GPL that makes it incompatible, not anything in the CDDL. This is why you see ZFS on every other OS except Linux. Solaris, BSD's, Mac OS even Windows and Haikau have a port. It's positioned to be a universal file system.. maybe even a replacement for fat32. (I want checksums on my usb stick, don't you? having glitches in that freebird mp3 really sucks lol)

It's unclear also if the CDDL is *actually* incompatible. It was Debian that originally found fault in it, Sun believed it was (at least the developers that have spoken upon it) pretty much the FSF and GNU say it isn't but others (such as Canonical and a Law professor at Cambridge say it is, because of a few reasons. (the fact that it's not a derivative work and it's file based and it's intent is compatible in spirit with the GPL) A real legal test should be put forth because if it is.. maybe we can all agree to work on bcachefs or something.

And.. if you really feel this way.. that's fine.. but you'll have to live without Firefox, Libreoffice, VirtualBox, Syncthing.. and a ton of other stuff you probably use every day is similarly licensed as ZFS.

[Guest]

but you'll have to live without Firefox

The only problem with MPL is about logos and the name, there are plenty of forks that sidestep the issue.