It’s hard to overcome the Hurd problem though. Although it would be fascinating to see how it would diverge on the design of the Linux kernel. How much can you still act like Linux while not being Linux? Or would it just be a direct algorithmic translation, basically doing the same processes under the hood with the same architecture? I’m sure there’s more than a few things Linux is doing in C that the Rust compiler would frown upon.
Main issue is drivers. One of the best places to take advantage of rust’s memory safety is in hardware drivers, and those would be hard to share between separate kernels.
That entire talk, and the complaint that Ts’o responded to was that to continue with rust, there needs to be some responsibility from the guys working on the underlying C bindings to not break downstream dependencies if they refactor code.
The answer from some of the Kernel developers, and vocally by Ts’o was: lol no fuck you and your toy language.
Yeah. The idea of an automated C to Rust replacement of the Linux kernel is fascinating.
As you say, there’s probably stuff in the Kernel that Rust’s compiler won’t allow.
I imagine it wouldn’t work at all, out of the box, but it might reduce the cost curve enough to make a dedicated team of very clever engineers able to cross the last mile, given time.
As cynical as I am of both Rust and AI generated code, it honestly feels like trying an automated conversion might be less of a long shot than expecting the existing Linux kernel developers to switch to Rust.
And I’m sure a few would kick in some thought cycles if a promising Kernel clone could be generated. These are certainly interesting times.
It’s hard to overcome the Hurd problem though. Although it would be fascinating to see how it would diverge on the design of the Linux kernel. How much can you still act like Linux while not being Linux? Or would it just be a direct algorithmic translation, basically doing the same processes under the hood with the same architecture? I’m sure there’s more than a few things Linux is doing in C that the Rust compiler would frown upon.
Main issue is drivers. One of the best places to take advantage of rust’s memory safety is in hardware drivers, and those would be hard to share between separate kernels.
That entire talk, and the complaint that Ts’o responded to was that to continue with rust, there needs to be some responsibility from the guys working on the underlying C bindings to not break downstream dependencies if they refactor code.
The answer from some of the Kernel developers, and vocally by Ts’o was: lol no fuck you and your toy language.
Yeah. The idea of an automated C to Rust replacement of the Linux kernel is fascinating. As you say, there’s probably stuff in the Kernel that Rust’s compiler won’t allow.
I imagine it wouldn’t work at all, out of the box, but it might reduce the cost curve enough to make a dedicated team of very clever engineers able to cross the last mile, given time.
As cynical as I am of both Rust and AI generated code, it honestly feels like trying an automated conversion might be less of a long shot than expecting the existing Linux kernel developers to switch to Rust.
And I’m sure a few would kick in some thought cycles if a promising Kernel clone could be generated. These are certainly interesting times.