[freeone3000]

However, it is not legal to then redistribute this combination. Which essentially means linux distros cannot ship with OpenZFS: each user must combine the two on their own.

(This doesn’t necessarily stop people, but it is read by Debian as “illegal enough” to warrant a splash screen on installing OpenZFS that you’re losing the right to redistribute.)

[Nextgrid]

Who would sue, and what damages would they demonstrate in this case? Both the Linux and ZFS source is available and you are allowed and encouraged to build and use them. Redistributing binaries can be thought of caching precompiled artifacts, merely speeding up what you can already do manually (and with reproductible builds the result should be identical). Nobody loses anything.

Let's say I built a magical compiler capable of compiling a Linux+ZFS kernel in miliseconds. I put it behind a web UI which accepts a Linux tarball and ZFS tarball and spits out the compiled kernel. Because of some mysterious bug I am still trying to solve, only specific Linux and ZFS tarballs work, so I put validation to hash the uploaded tarballs and only let through the ones that are known to work.

Let's peel back the curtain: there is no magical compiler and all this is doing is hashing the input kernel & ZFS source tarballs, using it as a lookup table in a cache of precompiled binaries, and spitting out the matching one, which is currently not allowed. But let's assume the compiler was real, in this case it should be fine, even though the functionality of the system is no different.

I would at least understand if one of the licenses explicitly restricted the right to distribute ready-to-use binaries as a way to carve themselves a moat so the authors are the only ones that would be able to do it (and thus charge for it). But that's not the case here, nobody is better (or worse) off whether I waste time building it manually vs reuse someone else's earlier effort of building it.

[freeone3000]

It would be the Free Software Foundation, for a copyright violation due to GPL licensing violation. (Technically the Linux Kernel Foundation, as they’re the copyright holders here, but the FSF is generally involved in suits to protect GPL.) It is not necessary to demonstrate damages for copyright infringement; there are statutory (assumed) damages for this tort.

WOULD they? It depends on how important it is to have a credible threat of enforcement for GPL violations. But it’s not zero, and it’s a pretty clear violation. Which is enough to scare off most major distros - if they receive a C&D, that’d be a breaking change they’d have to push retroactively. Not worth the risk.

[Nextgrid]

Would they have an incentive to sue - ie. how would that advance software freedom?

It seems like all this would achieve is make Linux + ZFS more annoying to use and overall everyone loses.

The licenses are incompatible on a technicality, but it doesn't mean there is anything for either side to be gained by suing?

[curt15]

ZFS is widely used with Linux in HPC (https://computing.llnl.gov/projects/openzfs). Is asking users to install ZFS separately really that much of a lift for ZFS's target audience?

[remexre]

Being out-of-tree means that kernel refactors break ZFS, and that it has a lot fewer hands and eyeballs available for the kinds of bugs that require internal design changes to fix (rather than paper over).

[josefx]

Most people don't blindly run the latest kernel. I don't think I ever ran into issues with out of kernel modules on a stable distro.

[pyuser583]

Asking for seperate install of a filesystem is a big deal. It severely limits how the filesystem can be used.