Ubuntu Talks Up Rust Kernel Programming Potential With Ubuntu 23.04

[Veto]

x86/x86_64 and Aarch64 are the only ones well supported by the rust compiler. Even with Riscv which just works on GCC, it struggles hard to compile anything which is why Firefox is still very hard to obtain on riscv.

A RISC-V computer capable of running Firefox is still very hard to obtain.

RISC-V is a Rust tier-2 architecture (https://doc.rust-lang.org/rustc/platform-support.html) and could be tier-1 if/when there is enough critical mass behind RISC-V. However, these days RISC-V is mainly used to replace ARM in the lowest cost microcontroller CPUs.

[Siuoq]

*Urge to crusade intensifies*

- Zig is not that obscure. It leads wasm4 gamejam Game Jam 2 Results | WASM-4 (wasm4.org) , also people are rewriting Rust code in Zig, like the roc standard library roc/FAQ.md at main · roc-lang/roc (github.com)
- Rust is not "secure". The compiler forbids some things, like having two mutable references to a value, but then you can't write a bunch of totally valid programs. Zig is arguable safer than unsafe rust When Zig is safer and faster than (unsafe) Rust : rust (reddit.com)
- putting a "." before structs may not seem ideal at first, but then the rust syntax is far worse with double colons and turbofishes, also this is shorter when you would write 'this.{}'
- ((also what's wrong with javascript syntax? I like javascript, I liked old javascript language/syntax before 2000, it is arguable better at some things than say scheme syntax or python syntax. It has a lot of visual clues for what's going on when you need to read it, but not that much noise.))

All in all, Zig's advantages:
- faster
- lower level
- better c interoperability
- allows you to write a broader range of software, faster and safer than Rust
- will be finished in ~2-3 years, then frozen in time (not yet)

(While Rust might be a better if you need the compiler to prove some things about your application logic. Or when multiple people need to collaborate. Like you can fairly confidently tweak the application logic in your drivers, without having the whole driver in your mind.)

[dlq84]

It's sad to see so many people choosing to be left behind when Rust takes off. I'm glad they aren't my colleagues. Hopefully they aren't software developers.

[marlock]

Sorry, not a dev here, just an avid reader, but...

Isn't "need the compiler to prove some things about yourapplication logic" and "multiple people need to collaborate" two big desireable things for linux kernel development?

Also isn't unsafe Rust relevant (in the linux kernel context) almost entirely to write bindings to C that are unsafe where they touch C but then everything else using this minimal unsafe piece can be safe, so a lot less chance of f...ing it up everywhere the same binding is used? (please forgive me for any improper vocabulary, I hope I managed to convey enough of my meaning regardless)

[smitty3268]

In fact the AGX driver+Mesa outperforms Apples own in OGL tests

This should be considered total BS unless someone can point to some well-constructed benchmarks. And preferably actual application benchmarks, not just some micro-benchmark.

I'm not saying it's not true, but AFAIK this talking point came from some blog post which flat out admitted it was testing apps at completely different resolutions, meaning it wasn't a valid test at all. And the blog post wasn't claiming it was, but was structured so that it was easy to take out of context, which is what Phoronix and now commenters are doing. So I'm pushing back. Put up proof of this or drop it, please.

To be honest, out-performing Apple's OpenGL drivers shouldn't be that difficult. Apple's drivers suck. But AGX has a ways to go too.

[marlock]

as dr. steven strange usually says: the warnings should really come before the spell

the blog post only claims that performance is "in the same league" (aka same order of magnitude), which basically means it's usable, not competitive, and intuitively evoking an oposition to cases where an implementation is conformant but runs stuff at 1fps (such as noveau in some circumstances)

getting to the same perf range on 720p as the official driver in 1080p is really really really good news for a reverse engineered driver still in the initial bringup phase


on the other hand, I'd like to submit another disclaimer to this comparison for public appreciation, that neither the blog nor phoronix nor commenters up to this point mentioned:

it's not uncommon for incomplete implementations of a feature to have extraordinarily good fps because they skip some complex task and finish rendering faster than they should (but with incorrect output)

now i'm not claiming this is the case with the new driver... i'm just trying to point out that i didn't spot any claim of full correctness/conformance, so the results may be skewed (un)favorably to the new driver in an unfair way at this moment due to incompleteness

fortunately vulkan has an extensive conformance test suite that the driver devs are certainly leveraging to help work things out, so I'm confident they're not gonna call it a day before everything is properly polished