> The stated goal of the project was to allow drivers to be written in Rust. Adding Rust bindings to the kernel oversteps that goal. It's a legitimate concern.
You do recognize that all drivers will need to bind to some C interfaces? So -- your argument (or the argument you suppose Hellwig has) is that it is better that each driver author recreate each such interface for themselves? Now, when these interfaces break as a result of a change in the underlying C code, instead of fixing that breakage at possibly a single place, that one true binding, now a maintainer might have to fix that breakage in a dozen such places? And this is preferable? This will cause less work for the overburdened maintainer?
You are aware this patch introduced no code into the main kernel tree?
It doesn't have to. By becoming a single point of failure for all Rust drivers that depend on it, it becomes the responsibility of all maintainers of the kernel to avoid breaking it when they change the C interfaces. It's a foothold into a world where all kernel maintainers need to run and test Rust builds, something Christoph does not want the headache of dealing with.
When your teenager brings home a puppy and promises you he'll never let the puppy leave his room, you know that's not true and it won't be long before you're the one taking care of it.
Ultimately it's about motivations. Long-term kernel maintainers are motivated to protect and promote the kernel as a maintainable and successful project. R4L developers, on the other hand, seem more interested in promoting Rust than promoting Linux.
>> You are aware this patch introduced no code into the main kernel tree?
> It doesn't have to.
Ah, it's one of those other kinds of Trojan horses that don't enter the city walls.
> By becoming a single point of failure for all Rust drivers that depend on it, it becomes the responsibility of all maintainers of the kernel to avoid breaking it when they change the C interfaces.
So -- I'll ask what the Rust for Linux people asked Hellwig -- what is your suggested alternative? Where do we go from here? Is it Rust drivers not be allowed to common interfaces ever? In that case, what are the Rust for Linux team doing?
Or is it that you would like Linus rethink his decision re: adding Rust to the kernel? And if so, why didn't Hellwig make that case directly to Linus? What's with all this performative bellyaching on the LKML?
Ah, it's one of those other kinds of Trojan horses that don't enter the city walls.
The kind that have to be invited in, yes.
So -- I'll ask what the Rust for Linux people asked Hellwig -- what is your suggested alternative? Where do we go from here? Is it Rust drivers not be allowed to common interfaces ever? In that case, what are the Rust for Linux team doing?
That's not the kernel team's problem. They provide a common C interface. The fact that there's an impedance mismatch with binding to them from Rust code is a Rust problem.
Or is it that you would like Linus rethink his decision re: adding Rust to the kernel? And if so, why didn't Hellwig make that case directly to Linus? What's with all this performative bellyaching on the LKML?
I don't know what Linus's goals are, apart from keeping his maintainers happy and keeping the kernel rolling along smoothly. That's not a small thing. From what I can see, Christoph has been a maintainer for over 25 years.
Does Linus want to have his cake and eat it too? Sure. But I think he earned that right by building Linux into what it is today. The R4L team hasn't paid their dues. As someone else mentioned, it took 10 years for Clang to become a supported compiler for the kernel.
You are aware this patch introduced no code into the main kernel tree?
See: https://lkml.org/lkml/2025/1/8/801> The stated goal of the project was to allow drivers to be written in Rust. Adding Rust bindings to the kernel oversteps that goal. It's a legitimate concern.
You do recognize that all drivers will need to bind to some C interfaces? So -- your argument (or the argument you suppose Hellwig has) is that it is better that each driver author recreate each such interface for themselves? Now, when these interfaces break as a result of a change in the underlying C code, instead of fixing that breakage at possibly a single place, that one true binding, now a maintainer might have to fix that breakage in a dozen such places? And this is preferable? This will cause less work for the overburdened maintainer?