[Timon3]

> If the filesystem corrupts data, it has to be fixed at the maintainer's discretion. This is what the users want. Tough luck it makes Linus' life harder!

Where do you get this idea? Lots of Linux users want lots of things - a big part of the reason Linux is so successful is because they don't get what they want, and the project instead focuses on stable development and release cycles. Many users want Linux to break userspace in this or that case. Do you think Linus should do that, because it's what the users want?

And lets not forget we're talking about an experimental filesystem. If you decide to use one of those, it's not asking too much of you to compile your own kernel.

[koverstreet]

Funny, because "don't break userspace" is one of the principles I've been citing.

Things always degenerate when it turns into power struggles and people are going "No, I decide!".

"Make sure thinks work" is the underlying principle, and it's based on that that the code should have been, and was merged.

But then the personality conflicts and power struggles came out, and there's no need for that.

- You don't go overriding a subsystem maintainer without a clear justification; if the patch in question has a good reason for being there and can't affect the rest of the kernel, there's a really high bar to clear. This has been an issue for the XFS folks in the past.

- We have to be able to have technical and policy discussions without it degenerating into "I don't trust you and you need therapy". That's just childish. The private discussions got really ugly on this one.

And, regarding bcachefs still being marked as experimental: I'm being much more conservative with the experimental label than btrfs or ext4 were. Your data is safer on bcachefs than btrfs, today: you're not going to lose a filesystem, repair is thorough and robust and complete.

You may still hit hiccups, which is why the experimental label is there, but robust and complete repair and rock solid multi device have been reason enough for a lot of people to switch already.

[Timon3]

> And, regarding bcachefs still being marked as experimental: I'm being much more conservative with the experimental label than btrfs or ext4 were. Your data is safer on bcachefs than btrfs, today: you're not going to lose a filesystem, repair is thorough and robust and complete.

> You may still hit hiccups, which is why the experimental label is there, but robust and complete repair and rock solid multi device have been reason enough for a lot of people to switch already.

It doesn't matter how conservative you are being, it's _still_ marked as experimental. That simply means there's no great pressing need to get a new feature into the next possible release - users can either compile their own kernels, or wait if they aren't able to. They decided to use an experimental file system.

You're making life much harder for these users by causing bcachefs to be thrown out of the kernel. No matter how you twist and turn it, you're responsible for "breaking userspace" in this case. I have no interest in trying to convince you - I've seen people much better at explaining things than I can try to do so, and I've seen you ignore each and every one of them.

[rastignack]

Maybe the FS was upstreamed too soon, causing your development velocity and the high expectations you have for your end users to be at odd with the well entrenched workflows of Linux maintainers.

In any case as a Linux user, I want to thank you for your work and for your code which taught me a lot.

I hope it didn’t take too much of a toll on you.

Let’s hope that with the recent stabilization, the maintenance will be easier.

[koverstreet]

Well, being upstream has not been a rosy experience, so that'd be an easy judgement to make in hindsight.

But consider: ext4 was done entirely in tree, and btrfs was upstreamed much, much earlier and took a lot longer to stabilize.

So compared to historical precedent, we already upstreamed much later. e.g. we upstreamed long, long after we stopped making breaking changes in the on disk format (that was the point where btrfs removed the experimental label!)

If we're saying even that is too soon, then we're saying that bcachefs shouldn't have been upstreamed until it was perfect. But, no one was ever going to fund developing a filesystem from scratch all the way to completion and perfection completely out of tree. That's far too much uncertainty, and that kind of money simply isn't being thrown around in the Linux filesystem world.

Asking a filesystem to only be merged when it's completely done and perfect is saying "we want all the benefit and none of the pain", and it's just fundamentally unrealistic.

The whole point of Linux is community based development, and that's how I've been developing bcachefs. I don't have a big engineering team - but what I do have is a large community of people doing very good QA work that I work with closely, on a daily basis. People show up from anywhere with bugs, I ask them to join the IRC channel, and we start working together and it goes from there; a lot of people see us doing productive work and stick around and find ways to help out.

If that no longer works within the development model of the Linux kernel... oi vey.

[throwaway2832]

> The whole point of Linux is community based development

You contradict yourself too much. You ignore feedback about not working well with others, and whenever someone wants to contribute, you shut them down by claiming you're the expert. This makes it seem like you're more focused on attracting investment than on actual collaboration.

[tucnak]

FWIW, investment is the name of the game in Linux development. If it takes a bunch of business to whip the old guard into place, so be it.

[throwaway2832]

> If it takes a bunch of business to whip the old guard into place, so be it.

You mean bcachefs is a plot to remove Linus Torvalds from his position?

Sorry to burst your bubble but it ain't gonna happen, Linus is way more important than bcachefs.

[rastignack]

Maybe then you should have considered this file system as truly experimental and expected your end users to make frequent backups. And advertise it as such. You could also have some kind of dkms bleeding edge module for your users to test fixes before they reach the kernel.

This way you wouldn’t be so preoccupied about getting code as fast as possible in the kernel.

[koverstreet]

No, a lot of bcachefs users are explicitly using it because of data loss, and they needed something more reliable; that's bcachefs's main reason for existing.

Besides that, if you want to make anything truly reliable, you have to prioritize reliability at every step in the process of building it. That means actively supporting it, getting feedback and making sure it's working well, and you can't do that without shipping bugfixes.

Having to ship a DKMS module just so users can get the latest bugfixes would be nutso - that's just not how things are done.

[rastignack]

Some distributions do release hotfixes before they reach mainline kernel. If you expect end users to compile Linus’ kernel head, why couldn’t they compile your branch though ? They get timely fixes.

[Brian_K_White]

There is no power struggle.

This is not a case of 2 equal entities failing to find a compromise.

One is both technically more valid and has the simple right to set the terms and processes regardless of other opinions, the other is neither. All failure to function is on Kent, not on any "struggle".