Yes, Bcachefs is experimental and thus needs more fixes. Everyone understands that, and bugfixes are indeed totally fine to be merged at all times.
The problem are new features, general improvements, and fixes that are actually features or require nontrivial refactorings. I understand the temptation, but these carry significantly higher risks than a well-thought out bugfix and are thus not welcome outside the merge window. Most prior rows between Kent and Linus have been about such patches, sometimes surreptitiously mixed in with more benign fixes.
This time his argument is "think of the distro users", but this won't fly - those either accept that using an experimental FS has consequences, or Kent learns how to work with the kernel community instead of testing the boundaries of the rules in the name of all the oppressed kernel developers gatekept by Linus. Shipping bugfixes for bugfixes would be kinda embarrassing for everyone involved (Kent, Linus, and the distros!), and that's why Linus rejects his PRs.
The problem are new features, general improvements, and fixes that are actually features or require nontrivial refactorings. I understand the temptation, but these carry significantly higher risks than a well-thought out bugfix and are thus not welcome outside the merge window. Most prior rows between Kent and Linus have been about such patches, sometimes surreptitiously mixed in with more benign fixes.
This time his argument is "think of the distro users", but this won't fly - those either accept that using an experimental FS has consequences, or Kent learns how to work with the kernel community instead of testing the boundaries of the rules in the name of all the oppressed kernel developers gatekept by Linus. Shipping bugfixes for bugfixes would be kinda embarrassing for everyone involved (Kent, Linus, and the distros!), and that's why Linus rejects his PRs.