Things have taken a bad turn for Bcachefs as Linux supremo Linus Torvalds is not happy with their objections.
“no features in RCs” is a very basic rule, this was a forgone conclusion. If these features were so integral to data integrity, he should have kept bcachefs out of the kernel until they were ready.
Sort of the ultimate fuck around and find out.
There is no reason that Kent Overstreet needed to do this.
I love bcachefs but I am so angry at him for making this happen.
There is in this case, and why Linus did accept the patch in the end. Previous cases less so though which is why Linus is so pissed at this one.
The reason for this new feature is to help fix data loss on users systems - which is a fine line between a bug and a new feature really. There is precedent for this type on thing in RC releases from other filesystems as well. So the issue in this instance is a lot less black and white.
Honestly, not even in this case (especially given the history). Kent has a kernel tree people can pull if they need to. If it is an emergency, point people there.
My uneducated kernel take. Flexibility is acceptable and desirable in small projects or low impact projects.
When the majority of the internet and a good chunk of PC are dependent on your project, predictability and stability is much more important than flexibility.
Reading all the comments (between Kent and Linux), the problem is that the bcachefs dev thinks that his project (the filesystem) is the critically important one and that the Linux kernel needs to bend to his will.
I am a bcachefs user but it is pretty damn obvious to me that the production Linux kernel is more important than an experimental filesystem.
Yeah, this one is on Kent… again.
He posted on Patreon that there’ll be a DKMS module. In my opinion, this should have been the option from the very beginning and upstreaming at a later point in time. It would have avoided a lot of drama. And now bcachefs is kind of tainted. The only way I ever see it back in mainline is there is an independent downstream of Kent’s kernel that has no connection to him whatsoever.
Shame because I had very good experience with the filesystem. Definitely better than when btrfs was new. But Linus is unfortunately right; Kent is unable to follow agreed collaboration rules.
Unfortunate situation that could have been avoided entirely. Though I don’t want to be too harsh on Kent. He spent a lot of time and work on bcachefs and it’s his most important project. As such, he’s more passionate about all of this. But the same can be said for Linus and the kernel on the other side.
Took me way too long to figure out this stuff is called B Cache FS and not BCA Chefs.
I still read it as BCA Chefs every time.
Well I didn’t…
Thanks, I too thought about the worst cooking software ever😂
Good for him. It’s a reminder that the rules apply to everybody.
Ah, sorry to read - I like the idea of Bcachefs and would have been happy to have it ready for production eventually.
OTOH it seems the recent years I read more about the drama about Bcachefs commits to the kernel, than about any technical parts of Bcachefs.
Can’t wait for a new Brodie-video on this topic. Stay tuned for some comments.
Good I didn’t choose bcachefs
It’s an interesting filesystem, but you shouldn’t use it at this point unless you know what the hell you’re doing. You’ll need to be able to notice, report and help resolve bugs, and under no circumstances use it for production or where you can’t afford to lose some or all of the data on the partition.
…which makes Overstreet’s argument for urgency even weirder.
True
