Rust in the NetBSD Kernel, and other odd decisions
bentsukun.ch
bentsukun.ch
This does not paint the full picture. Rust can be bootstrapped with mrustc, which is written in C++
https://github.com/thepowersgang/mrustc
Now, mrustc supports only Rust 1.74. To build Rust 1.92, you need almost 20 builds. But this can be done from source
Guix has written about bootstrapping Rust from source (they care a lot about this). Here is how it looked like in 2018
This is not what I was expecting computer science to become, 30 years ago...
CYGWIN_NT-5.2 1.3.22(0.78/3/2)
Are there many architectures that NetBSD supports where GCC is not available?
The rustc_codegen_gcc project allows using the rustc compiler to target any architecture that GCC supports. I think it is pretty far along.
https://github.com/rust-lang/rustc_codegen_gcc
> getting a Rust compiler running in the first place is hard
> bootstrap relies on a binary package
> the compiler would also have to be part of the base system
The gccrs project aims to make Rust a fully supported language in GCC. It is fully self-hosting and can be built from source. Once accepted, it will be one of the languages you get when you build GCC.
https://github.com/Rust-GCC/gccrs
When gccrs is ready, it will address all the above objections.
Linux faces the same issues. Today, Rust is optional in Linux and can be skipped on any platform not supported in LLVM and rustc. For Rust to become mandatory in Linux, gccrs is going to be required (for some platforms).
> the release cycles of Rust are not compatible with the NetBSD ones
I do not fully grasp the objection here. Obviously NetBSD would not be forced to use any Rust features they cannot compile. And using gccrs for the NetBSD kernel would not prevent them from bundling LLVM and rustc as well if they want.
Anyway, I am not advocating the use of Rust in NetBSD. I am merely pointing out that solutions to the issues raised are being worked on.
https://github.com/rust-lang/rust/pull/151156
This means that you will be able to use either GCC or LLVM to build your Rust code with rustc (the main Rust compiler). That means you can compile your Rust for any platform that GCC supports.
This only addresses the first point in the blog post (architecture support) but it is an interesting development.
It has no formal spec, changes too fast, depends on third party libraries that change faster than I can breath, and is controlled by a foundation that is controlled by big tech corps.
What could go wrong?
https://ferrous-systems.com/blog/ferrocene-25-11-0/
Lets not forget not having a formal spec apparently wasn't an issue for C, which only got standardized in 1989, and even K&R C only specified a subset of its behaviours, which is a reason why there is so much UB, and implementation specific behaviours with YOLO C, as the Fil-C author likes to call it.
C emerged in the 1970s when there weren't many other options, and it was rapidly propagated because US antitrust laws forced AT&T to inexpensively license Unix (and by extension C) to universities and research institutions.
That was half a century ago.
I am a supporter of Rust adoption, but the line of reasoning that "C didn't need X, so Rust shouldn't either" is fundamentally flawed.
The time difference between today and C's public emergence is the same as the time between C's emergence and the 1920s. The field of computing is radically different today and any language that wishes to see wider adoption now must play by a completely different set of rules than those of 50 years ago.
C has to thank its adoption by being freely available with UNIX, that AT&T was forbidden to take commercial advantage of, for many years, the only reference implementation was whatever cc does.
Yeah, pedantically speaking there were other options but not for the average person. Languages like PL/I and JOVIAL were locked away in corporate institutions behind exorbitant licensing fees. Discoverability was limited as there was no WWW and there was no concept of open source. If you were an established company, then sure you could pay IBM $20000 for PL/I and be done with it.
But if you were a founder building something new, then you were stuck with what was available and familiar, and if you went to a university in the USA after the mid 1970s and learned anything about computers then Unix and C is what you knew. It just so happens that such companies were the ones who created much of the modern computing landscape, so C came along for the ride.
But anyway,
> C has to thank its adoption by being freely available with UNIX, that AT&T was forbidden to take commercial advantage of, for many years, the only reference implementation was whatever cc does.
This is exactly my point. The path that C took to wide adoption cannot be replicated by a modern language, so any drawing parallels along the lines of "Well C didn't need a formal standard" are not useful.
It's like thinking that you can be an internationally successful superstar by greasing up your hair and playing rockabilly music just because Elvis did it in the 1950s.
The world has changed and Rust will need to play a different game than C did.
Rust already did, it takes safety seriously, guess why Multics got an higher security score than UNIX?
The core language has been static for ages, and breaking changes are handled by the edition system so you can use a modern compiler to build code on old syntax. Since the 1.0 release ten years ago there have been four editions.
It's absolutely not changing too fast
> depends on third party libraries that change faster than I can breath
No it doesn't. The standard library is already sufficient for a lot of work; and there is an unhosted version with a "core" version of that standard library which has zero dependencies.
Modern Rust, Java, Python, TypeScript etc. developers choose to use a lot of third party libraries; but that's only because the tooling and ecosystem are both good enough to facilitate that. Nothing about the language forces it.
Linux has a wishlist of features they want for kernel development, and Rust has been working towards adding them.
Here's the paradox: Rust is very careful about compatibility and stability, the stable releases are changing slowly. But the Rust for Linux project wants to use the newly prototyped features right away, so they depend on not-yet-released features from unstable nightly versions of Rust.
This is not true. Since kernel 6.11 they have specified a minimum version that is already stable. The strategy for the Rust kernel is to use the version of Rust that ships with Debian Stable. That is very far from using "the newly prototyped features right away".
https://rust-for-linux.com/rust-version-policy
Of course, the kernel continues to inform Rust evolution. But you do not need an unstable version of Rust to compile Linux.
Rust For Linux has a hard dependency on unfinished Rust features that were not released as stable in the Rust version that Debian has.
However, every build of rustc also needs to be able to build its own standard library and a future version of itself. To do this, rustc recognizes a special private env var that exposes compiler-internal language extensions and not-yet-stabilized features meant for the compiler itself, and not for end users.
Rust for Linux relies on using this private env var to bypass feature stability checks and trick the "stable" compiler from Debian into allowing use of compiler-private and experimental features that were suppressed to stay disabled.
RfL uses an old Rust binary from Debian, but still depends on experimental features that haven't officially shipped yet in any Rust version.
Being base.dec a subleq EForth image and eforth.fth the source code, to create a new one:
./subleq base.dec < tuned_eforth.fth > new.dec
And Uxn partially. You can run the compiler as a UXN rom (Drimflim.rom).
C++ certainly isn't that special. It's a pain to implement, but so is Rust.
This argument can be pushed all the way down to build your own CPUs from sand, get to build the first hexdump monitor with electric switches, followed by writting the first Assembler in hexdump, and only then the first compiler for one of thoses languages in Assembly, and finally you get C.
Bootstrapping Rust would involve the same steps, only at some point you've got to write a Rust compiler in assembly, or write a C compiler in assembly that can compile mrustc, which probably means writing another C compiler in C so you can easily add all the needed features.
Or even better, lets follow the hype, use bootstrap strategies from the 1960's, and use WebAssembly instead, like many bytecode based compilers on those days, having written a macro Assembler instead, thus mapping the bytecodes into macros for the specific home made CPU.
Presumably the NetBSD project can bootstrap to GCC.