Next steps for Rust in the kernel

https://lwn.net/SubscriberLink/908347/da67a5162d1bc4a3/

243 235 4 years ago

sondr32• 4 years ago

Torvalds added that Rust isn't that terrible in the end; "it's not Perl".

Glowing endorsement from Torvalds right there. I'm really curious about how this will work out, Rust is my favourite programming language by a fair margin and seeing it being used more and more in places where C or C++ has traditionally dominated is very exciting. The future is looking bright (for those of us who like Rust).

TotoHorner• 4 years ago

For someone new to systems programming, should I just start with Rust?

Or should I learn C/C++ first and then learn Rust?

I've just done projects in Python,JS & Java so far.

oleganza• 4 years ago

I learned Rust because I need to write low-level high-performance stuff and I don't want to spend time learning how to do it correctly in C++. Rust compiler automates many chores for me, just like any computer technology is supposed to.

PerFartmann• 4 years ago

Keep on using python and JS.

When the time comes you can start with Rust

Twisol4• 4 years ago

Rust has a bit of a steeper learning curve at the start, but it pays off once you get used to it. C and C++ are deceptively simple at first, but as time goes on you keep discovering more nuances that affect most of your code.

There's no real problem starting with Rust, especially if you have some experience with other languages. The biggest difference is going to be getting used to dealing explicitly with memory, since all the languages you listed have a garbage collector; but this is true of all of C, C++ and, Rust. Be kind to yourself -- don't expect it all to click instantly.

sondr32• 4 years ago

I started using Rust because I wanted a programming language that compiled to native binaries, and the choice at the time was really only C++, Go and Rust. In the end I chose Rust because it was new and shiny, and I've stuck with it. Never written a single unsafe line of code or anything low-level, it's just been CLI tools and web APIs.

skywhopper2• 4 years ago

You’ll ultimately need to know both. Lots of new stuff will be in Rust, but plenty of old C/C++ code won’t be going away in our lifetimes. I’d start with the one used by the projects you are most motivated to work on immediately.

the_duke• 4 years ago

This is highly personal and depends on your learning style.

I always want to learn bottom up, and personally recommend starting with C.

Rust gives you many great abstractions that you will want to use, even if you skip the standard library. This gets in the way of really learning to work with and understanding pointers and low level mechanisms. You can certainly do that in Rust, but you have to force yourself to ignore most oft the language.

You will have to read and interact with C code eventually, and be it just calling out to libraries or wrapping libraries in Rust interfaces.

A solid understanding of C is critical to do that correctly.

There's also a lot of safety tooling that you rarely need with Rust, but should be comfortable with. (Like valgrind)

After learning Rust you will hate working with C (well, I do...), so it's better to do it early.

pizza234• 4 years ago

Unless you're doing isolated experiments, it's complicated, because systems programming inevitable interfaces and coexists with C. So one will need to have a reasonable understanding of C and its ecosystem (I'm doing this now).

Moving from high-level languages to Rust is going to be a nice punch in the face :)

Some people suggest to learn C/C++ but that ultimately depends on the resources available. If one has infinite time, certainly it's ideal, but with limited time, I think learning Rust will already be very demanding.

My suggestion: find some small projects you find interesting. Follow the usual Rust learning path (reference book first, then find other books), and apply it to the projects you've chosen; after an year or so, ask the question again.

Other small suggestion: don't start from book that pretend to teach Rust by learning another topic, because the amount of Rust that can be taught in 100 pages is useless (a lot of books pretend to do so). On the other hand, there are way more books to find fun projects/topics than there were an year ago.

thetrashcanman• 4 years ago

I would argue that if you’re really interested in systems programming, learning C is required just because the vast majority of the resources on systems programming is explained using C. The only notable exception is the Book Rust In Action! That being said, is your goal to be a systems programmer or a rust programmer? If it’s the latter, then learn Rust! But, as it seems that you are learning programming in general, learn the basics and data structures algorithms (these are rather language agnostic).

dijit6• 4 years ago

C will teach you how computers think.

Rust is it's own thing; I personally really like it but going from C to Rust is harder than just going to Rust as you have to "unlearn" C paradigms.

You'll probably be "better" if you go C first, but it will take a lot of time and energy.

In the end its completely up to you.

deckiedan• 4 years ago

It depends what you're trying to do.

If you want just to "play" and do some tiny projects as low level as reasonable - go for learning C first. You'll learn a lot - and then when you learn rust, the reasons why things are done that way will make a lot of sense. Also, so many low level utils are written in C - having a basic knowledge of it is really useful. But don't try to build a big complex thing in it unless you have a really good reason to (or you fall in love with C. Some people do!)

If you want to actually build something big and complex and for real world use - probably go for Rust.

If you want to build a big production system using an already established tech that's based on C++, learn that. (Eg. one of the C++ based game engines, or a C++ GUI framework, or whatever).

amalcon• 4 years ago

It's difficult to answer this right now because the overwhelming majority of people who use Rust now have at least some background in C or C++.

If you're coming from Java, the most difficult thing you'll need to learn is memory management, as all of those languages do that for you. The other difficult thing is threading, but that's 1) unnecessary and 2) not actually that different from Java.

Rust forces you (unsafe aside) to do memory management in one specific way, and verifies it in the compiler. C and C++ let you do whatever you want, though C++ encourages (without requiring) a Rust-like approach.

I think Rust will probably be better if you want to have a constrained environment that tries to prevent mistakes. C will be better if you want to explore on your own, learn from the mistakes, and thus understand why Rust wants you to do particular things. C++ frankly doesn't have a whole lot to recommend it unless you are working with a C++ codebase, already know C, or really need one of its features.

packetlost5• 4 years ago

Depends on what you're doing, but really Rust is built with the explicit goal of being able to link to C easily. You would be doing yourself a favor to learn C (not C++) in general, but especially because you will be able to use C libraries from Rust.

kjeetgill5• 4 years ago

Start with C. C++ is fine too.

You don't need to 100% the language, a few months experience will be be handy for the rest of your computing career. It'll be a perfectly reasonable stepping stone for reading C++ without really needing to learn all the details to write it.

As much as any language can be: C is still the lingua franca, a relative common denominator. As you explore different parts of the computing environment around you, you'll run into it constantly.

Learning about python? C. Digging into how your program talks to the OS? glibc and the man pages and most examples will be C oriented. Any documentation ever explains how something is laid out in memory? It'll be explained in C structs. How function calls work? The assembly will be explained in terms of C code and C's calling convention. Any language's FFI? C. Digging into the JVM, browsers, Javascript? It's C++ but you'll know enough to eyeball the code.

This is mostly about all the things you'll want to be able to read about well. Anything you'll want to know has been done in C or C++ — Nothing big exists that you'll want to explore is mostly written in Rust or Zig yet. And there are still things that can't reasonably be yet either (oversimplification).

rust_is_dead3• 4 years ago

My advice would be to learn the language that your operating system is written in.

qwery• 4 years ago

should I learn C/C++ first and then learn Rust?

If your endpoint is Rust anyway, just start there. You can always learn C (and/or C++) later.

I don't think it matters (much) which language you start with. Any one of the suggested languages (or commonly suggested languages in general) will be fine. I think this applies in general too, at any levels of skill/experience, but probably matters more for someone with less experience.

Symmetry• 4 years ago

Learning enough C to be able to understand most C you come across is a much, much smaller lift than doing the same with C++. It can be useful later and also give you an understanding of how computers work, at least down to the level of abstraction they present to compiled code which does hide some of the underlying complexity. And I think, for me at least, having a C level understanding of things will help understand the design and rules of Rust more easily. You don't need to learn it first but spending a weekend on it before diving into Rust is plausibly a good idea.

C++ has everything C has plus many layers of complexity deposited over it that can make you more productive but also obscure your relationship with the machine. If you're going to go on to Rust I wouldn't bother with C++ at all right now.

jerf• 4 years ago

Personally I'd recommend Rust first, then C.

Rust's borrowing and ownership semantics are not bizarre and novel solutions to things that aren't problems in other languages. Rust's borrowing and ownership semantics are a compiler-level reification of problems that exist in all languages, and especially multithreaded languages. This 100% includes C. You may not adopt Rust's exact solutions in all cases, but if your are programming at a system's level at all and you're not thinking about the problems that Rust exposes directly, you're in trouble and you don't even know it. Using Rust is a great way to work in an environment where the compiler is forcing you to learn how that all works, and will train you good and hard about how to think about issues of ownership and what code is allowed to touch what.

I primarily work in Go, a garbage collected language, and I heavily use the concurrency features. That doesn't mean I don't have to think about ownership because the language lacks any support for it, or that I don't have to worry about who is responsible for deallocating things because I work in a GC'd language. It means I have to think about those things, but I lack support from the language. My personal career experience means this is not a terrible tradeoff; as a result of all my time in Erlang and Haskell it is no big deal for me to operate concurrently in a manner that I know will work, because while they are different than Rust, they were also harsh taskmasters in terms of only allowing me to do certain things that work. Other tradeoffs make Go worth it for me despite this lack of language support. But the lack of support doesn't make the problems go away. It just makes it so you can't see them immediately at compile time.

I would expect that someone who learned Rust for a year, then learned C for a year, would probably produce better C code than someone who learned C for two years. Eventually, the C programmer will, by hook and by crook, learn all the things that Rust would have taught you, but you'll be learning it in an enviromnent that lets you make mistakes, then build on them for months before they finally blow up in your face. It is far easier to learn about what mistakes you are making in an environment where it instantly tells you that you made a mistake.

(Also, I should clarify something: By "C" here, I mean programming in raw C, without any particular additional support. For the sort of purposes I'm talking about here, I actually consider "C with a strong static analysis tool like Coverity" a separate language. If you use such a tool pervasively and work in a code base that has either been cleaned to its satisfaction or started from scratch under the analyzer's support, you can also learn in a fast-feedback environment what not to do. It won't be the exact same lessons as Rust, but it'll be good enough; there are many paths to enlightenment in this case, just as I trod a completely different one as mentioned above. The important point is to use one of these, or perhaps another, and not simply stand in front of the monster that is Raw C with nothing but your own wits to guide you. I don't care how smart you are. It'll eat you alive. You do not want to be responsible for trying to tame the thing yourself under your own power and nothing else.)

NIckGeek• 4 years ago

I'd recommend learning Rust because the way Rust pushes you into writing code will lead you to write better C code, etc.

People that learn C or C++ first often end up fighting the Rust compiler a bit until they adjust their mindset.

klabb32• 4 years ago

Rust is written in the blood of fallen C/C++ projects.

If you're willing to accept, at face value, that certain easy tasks are convoluted for Good Reasons, you can go straight to Rust. If you want figure out why for yourself, start with C.

aaaaaaaaaaab• 4 years ago

Most production systems have been transitioning to Rust in the past few years, so that’s the safest bet in my opinion.

The story I’m seeing at most companies is that they have a few old guys taking care of maintaining their legacy C/C++ codebase, and there’s a diverse bunch of young folks shipping new components in Rust.

gary17the• 4 years ago

For someone new to systems programming, should I just start with Rust? Or should I learn C/C++ first and then learn Rust?

IMHO, if your end-goal is Rust, learn Rust directly.

The angle I am coming from: I've been doing C++ for 20 years and Rust for over a year.

Learning C/C++ can provide great value on its own, through gaining knowledge of how lower-level programming primitives work, but many of your C/C++ skills won't be subsequently applicable to Rust, except perhaps for core concepts such as a stack, a heap, owning memory, a `struct`, etc. This is caused by the fact that Rust takes many programming abstractions much _farther_ than C and in a substantially _different_ direction than C++. For example: where C uses manually managed pointers, Rust uses compile-time checked ownership and borrowing; where C++ uses inheritance, Rust uses generics and composition; where C uses void pointers ;) and C++ uses pure virtual functions, Rust uses traits and `where` clauses; where C and C++ use... a programmer's eyeballs to guarantee multi-threaded correctness ;), Rust uses compile-time "marker" traits; and so on.

Don't get me wrong: C and C++ are extremely able, powerhouse languages worth your time in their own respect, it is just that most of the C/C++ -specific knowledge you invest your time to learn beforehand won't be directly applicable to Rust later on, because Rust just does things in a significantly different way than C/C++.

Oh, don't forget to RTFM - Rust is easy to learn by reading (in order to understand its unique concepts), but Rust is hard to learn by experimentation alone (trying 100 different syntax variations to see what compiles).

nextaccountic2• 4 years ago

Don't start with C++.

Between C and Rust, Rust has better tooling and an easy to use library ecosystem. C is a smaller and easier to grasp language.

I would start with Rust if you want to write a real world project. And probably start with Rust if you just want to write some leetcode algorithms as well (but C is fine here too)

spoiler• 4 years ago

I've personally been doing C and then C++ for about a decade before moving onto Rust 2-3 year ago. I don't think you're missing out on much by skipping C or C++, but I recommend dabbling in them a bit (doesn't matter if before/while/after learning Rust), as it will make you appreciate Rust more.

WiSaGaN• 4 years ago

Any historical background for the reference of Perl by Torvalds?

dijit6• 4 years ago

Perl is notorious for being "write only".

As in: keeping it clean and well structured is impossible and the impressive flexibility of the language is huge downside on larger projects.

I feel qualified to say such things as I've worked on a billion-line perl codebase before.

mschuster912• 4 years ago

There once was a time when the only common things you could expect to be present on a Linux system were the good old `sh` (not `bash`) and a more or less old version of `perl`.

The commands supported by `sh` are... let's keep it at "fossilized from somewhere in the 80s", and so the only alternative was perl which has a ... history of being not exactly user-friendly to code for.

werdnapk• 4 years ago

Perls big moto of TMTOWTDI always bugged me and I thought it was more of a negative than a plus which led to it being known for write-only code.

ElCheapo• 4 years ago

Linus hates Perl?

Heck, so my secret project of slowly replacing every Perl script in the kernel with C might even get approved...

tialaramex9• 4 years ago

The dislike of Perl in this context is that it's a write-only language. See Titus' recent talk about "Software Engineering languages" for some thoughts about that.

https://www.youtube.com/watch?v=yA_wUiNuhSc

C isn't a good software engineering language but it might be better than Perl. Or in this context, maybe not.

bitL• 4 years ago

That sounds like what we heard about Scala 10 years ago until complexity started to bite, and now that language is in a free fall and its parent company in financial distress...

camdenreslink• 4 years ago

Scala isn’t that complex unless you try to pretend it’s Haskell.

dblohm7• 4 years ago

Given the projects where Rust is already in use (and has been for several years at this point), I don't think that is a concern.

karmakurtisaani• 4 years ago

Scala is not that bad. I'd take it over java any day.

sideeffffect3• 4 years ago

I don't now what situation was Scala in 10 years ago. But the situation it is now couldn't be further from "free fall".

It is _the_ language where most of the Functional Programming in the industry takes place (that's not to say that Haskell, F#, OCaml or Clojure are bad, they just aren't as widely used). And that also means that there are lot of open job positions available.

I like FP and Scala comes with some awesome libraries for concurrent/async programming like Cats Effect or ZIO. Good choice for creating modern-style micro-services to be run in the cloud (or even macro-services, Scala has a powerful module system, so it's made to handle large codebases).

https://typelevel.org/cats-effect/

https://zio.dev/

The language, the community and customs are great. You don't have to worry about nulls, things are immutable by default, domain modelling with ADTs and patter matching is pure joy.

The tooling available is from good to great and Scala is big enough that there are good libraries for typical (if not vast majority of stuff) and Java libs as a reliable fallback.

avgcorrection• 4 years ago

Missing context? That came after a list of concerns and what-ifs about putting Rust in the kernel. Not an endorsement but also not a backhanded compliment, like you made it sound like with your cherry-picking.

up2isomorphism• 4 years ago

Linus loses control over kernel for quite a long time already.

bayindirh• 4 years ago

I don't think he's trying to be an all-controlling BDFL. What I see is, he acts as a trim tab which steers the kernel to a general direction rather than a precise bearing.

skywhopper2• 4 years ago

Sounds like Torvalds is taking a measured and pragmatic approach. Which is the only thing that will work. Kudos for taking it slow, allowing maintainers to opt out until the integrations are more proven, and resisting the urge to rush such a huge change. If anything, Linus is moving very rapidly, given that Rust still requires unstable features to work with the kernel.

tl• 4 years ago

Read the article, mentally replacing Rust with C++ (or terms like gccrs with g++). Interesting times for the kernel; you have to wonder whether Rust will be a success or if there will be a repeat of the 1992 attempt [1] to include C++.

[1] http://harmful.cat-v.org/software/c++/linus

xedrac• 4 years ago

Rust brings a lot more to the table than C++ does - it solves many of the "hard" problems, at compile time, that kernel developers have to constantly grok with. I have a feeling, that it'll not only do well in the Kernel, but may eventually become the preferred language.

daptaq• 4 years ago

I have a number of negative stereotypes associated with Rust and Rust programmers that I am trying to set aside right now and just look at this from a technical perspective. Has the Rust-portability situation been addressed? Or will this change restrict the number of platforms the Linux kernel will be usable on? Assuming that is the case (I remember reading that GCC will acquire Rust support), what will the advantages be? I would guess stability might improve due to more static analysis, but what about performance? Build-time? And will this have any effect on Rust?

dgs_sgd• 4 years ago

What are the negative stereotypes you associate with Rust and Rust programmers?

sidlls2• 4 years ago

The belief that the problem of memory safety that Rust has some compile-time protections against is the only problem (or so much more important problem that the rest almost don’t matter) of programming, and that therefore Rust is the only sane choice to use for any project. There’s a reason that memes like “Rust Evangelism Strike Force” exist.

That’s the biggest one, anyway. I picked up Rust for the first time 8 years ago. I liked it: it felt comfortable in my “worn” C++ hands. But the community outside of a very few core individuals who care deeply about the language’s success can be quite toxic and dismissive of anyone who suggests Rust isn’t basically perfect. And it was a bit of a turn off for me, personally.

chomp• 4 years ago

From my experience, the stereotype is that they are very positive within their own community, very loud and negative towards anything not in the Church of Rust. Of course, this doesn’t fit most Rust programmers, most I’ve met are great people and developers, it’s just the loud people getting the attention.

rust_is_dead3• 4 years ago

I have a word in mind, but before that I would ask: what are some safety-critical systems written in Rust? If there are any, how big a role does Rust play in their safety properties? Is it much different from any other programming language?

I can already hear faint exclamations about the language's age, but are Rust programmers justified in putting forward claims about safety? Or do they shift to using the word in the sense of "fewer bugs in this particular class" (and nevermind any other effects the language may have on development).

Even if we take such a narrow and distorting view of "safety", we can look at, say, "The Power of Ten" and ask how does Rust use fare with each rule, and is it any more advantageous than other languages in use by safety-critical systems.

MisterTea2• 4 years ago

The simple fact that they call themselves rustaceans and evangelize the language. It feels the language spread via a religious uprising rather than merit of making good programs easy to write.

gspr6• 4 years ago

I love Rust, it's the most fun I've had programming since first discovering the concept as a kid, with the possible exception of the time I was enamored with Haskell. But one stereotype that I've faced a lot is in connection with the ecosystem. In particular, if you're not satisfied with "rustup everything, then let cargo pull all your deps from crates.io, it just works" as a way of work, you'll sadly face quite a bit of hostility from parts of the community. Seriously; a programming language or its community shouldn't dictate how I assemble and build code! Luckily there are exceptions, and more importantly, the toolchain and ecosystem seem to be getting better in these regards all the time.

Still love the language though, and in general the community is very friendly and welcoming (if not always open to criticism).

Narann2• 4 years ago

Rust community is nice until you use unsafe.

More seriously, the number of peoples you can have a sane discussion about unsafe use is quiet low because when you need to use unsafe is because of a complicate/hard to explain/understand problem.

This combined with the (99% of the time valid) safety dogmas brings some to instantly tell you how to not use unsafe with blatantly bad workaround without considering why you _would_ use unsafe at the first place.

This safety/anti-unsafe dogma makes some peoples far too vocal.

I consider Rust community to be nicer with time, seriously, but I can't say this "I don't understand your problem, but I saw a unsafe, here is a code that doesn't use unsafe" is not a thing. Of course, it's changing, because Rust is more widely use, and "here is when you need unsafe" patterns are emerging. As a Rust user, I'm happy.

packetlost5• 4 years ago

I was very put off from Rust because of the people acting out those stereotypes that I've encountered whenever the topic comes up. But then I decided to just learn it (partially driven by work requirements), and it's honestly great. I don't think we should be rewriting a ton of software just because there's better compiler-level safeguards, but it does make it much harder to shoot yourself in the foot without realizing it. There's some issues, dynamic linking isn't really supported. I've had good luck with various embedded targets, having trialed STM32F4s, RP2040, ESP32-C3, and am waiting on getting an NRF52840. Surprisingly, the ESP32-C3 is the best out of the box experience I've had, but library support is just not there.

IMO Rust is a really good language, and it deserves it's place in the toolbelt of most systems programmers. It's not always the right choice, but it's usually a good choice.

consp• 4 years ago

But then I decided to just learn it (partially driven by work requirements), and it's honestly great.

I did for work as well and in my case I do not want to touch it again for quite a while. Cargo is a terrible system with all kinds of hidden and possible broken dependencies you introduce and everything depends highly on the version of rust you use. It's not even close to mature enough to consider using except for test projects. It doesn't fix programmer errors despite every rust user trying to convince me otherwise that all imported crates are perfect.

I like the idea of the language somewhat but the entire ecosystem ruins it for me.

gspr6• 4 years ago

dynamic linking isn't really supported

This is one of my few big gripes with Rust too, but it should be pointed out that it's dynamic linking with other Rust code that isn't really supported. Dynamic linking with e.g. C libraries is a breeze.

schuyler2d• 4 years ago

Rust in the kernel is starting with drivers, so if there's a Rust driver for something, it's presumably for hardware that's on a supported platform. Regardless, if you only want C-written work, you're no worse off.

The gccrs and rust_gcc_backend projects both are progressing well and would presumably help enable support for other targets only supported by gcc.

As far as its affect on Rust, the last year or two, a significant amount of development and affordance has gone into features for "no-std" and its likely that will only accelerate now that Rust is going to be in Linux (and have an opportunity to prove itself there) along with the number of Rust developers that are interested in kernel development.

The performance of the NVMe Rust driver in Linux persuaded some of the skeptical kernel devs. With Rust's greater static analysis (and a lot of work around formal methods), in the long run, there will be optimizations that are at least easier if not impossible in C-based codebases.

The main advantage is, as you said, the static analysis that radically reduces pointer-related bugs and security vulnerabilities. Another advantage that I'm personally a little skeptical about, but has been expressed by more than one kernel developer, is attracting new talent to becoming kernel developers as the current group is aging.

mlindner3• 4 years ago

The performance of the NVMe Rust driver in Linux persuaded some of the skeptical kernel devs. With Rust's greater static analysis (and a lot of work around formal methods), in the long run, there will be optimizations that are at least easier if not impossible in C-based codebases.

Do you have a link to some of the discussion around this? I'm curious to hear what kernel devs thought of the driver and how it was introduced to them. Is there a talk about it? Or mailing list archives? (Or both?)

__jem• 4 years ago

There are two in flight projects to bring GCC to Rust: gccrs and rustc_codegen_gcc. These both seem to be making good progress. Otherwise, right now Rust is only being included for drivers, so doesn't have as many portability concerns.

FnuGk• 4 years ago

i might be wrong but currently rustc is based on llvm so any platform llvm cant target, rust can target. For the linux kernel it seems like they want to use a gcc frontend for rust (gccrs) that is currently under development. So with gccrs rust should be able to work on any platform gcc supports thus giving it the same portability of the current c code in the kernel.

roblabla• 4 years ago

Has the Rust-portability situation been addressed?

The rust portability story is in the process of being addressed, by two different projects:

gcc-rs[0] adds a Rust frontend to GCC.

rustc-codegen-gcc[1] adds a GCC backend to rustc.

Both are progressing pretty well, but it will likely still take a while until they reach maturity.

Or will this change restrict the number of platforms the Linux kernel will be usable on?

There won't be any Rust inside the Linux core, only for drivers. So no, the linux kernel will always be usable on all platforms. However, the drivers written in Rust won't be usable on platforms not supported by LLVM until a GCC alternative becomes mature.

what will the advantages be?

There are multiple: Safety, improved stability, and developer productivity. Compile time will likely take a hit (Rust is famously slow to compile), but I don't expect a big shift in runtime performance (could be slightly better due to the extra opportunities the compiler has for optimization, but it's not super likely).

As for having any effects on Rust: it already has! A lot of unstable features got prioritized due to being necessary for the kernel, such as fallible allocations.

[0] https://github.com/Rust-GCC/gccrs - monthly reports at https://thephilbert.io/category/gccrs-status-updates/

[1] https://github.com/rust-lang/rustc_codegen_gcc - monthly reports at https://blog.antoyo.xyz/

nequo2• 4 years ago

I always have reservations about clicking LWN "subscriber links" since I have not subscribed.

Perhaps it's worth waiting a week before posting these? LWN's paywall is dropped at that point.

pabs32• 4 years ago

LWN folks are fine with subscriber links being posted here, they view it as a form of advertising and they get subscribers from it.

nequo2• 4 years ago

That is good to know. But I looked this up and what I've found is Jonathan Corbet saying in 2010 that "occasional" posting of subscriber links is fine.[1]

I am not Jonathan Corbet so I don't know what he considers occasional. But this is the fourth LWN subscriber link posted in the past 7 days, and the eighth in the past 14 days.[2]

For reference, LWN currently lists 10 paywalled posts, the oldest from 13 days ago.[3] Those posted on HN are nearly every paywalled article that they have.

[1] https://news.ycombinator.com/item?id=1966033

[2] https://hn.algolia.com/?dateEnd=1663790400&dateRange=custom&...

[3] https://lwn.net/

rascul• 4 years ago

You might consider subscribing to get more of LWN's content as it's released. Some time ago, the wait for the paywall removal for information that was relevant to something I was doing then and possibly not a week later is what finally got me to subscribe.

cozzyd• 4 years ago

In that case, why not subscribe? It's not expensive (they even have tiered subscription levels...) and it's awesome.

cesarb2• 4 years ago

In that case, why not subscribe?

Not the OP, but a possible reason: not all people have credit cards. In fact, the very first thing I did once I finally decided to get a credit card (which, frankly, is only really necessary to buy things from outside my country; within the country, there are other convenient payment methods) was to subscribe to LWN. Before that day, other than a brief year in which a friend had paid for my LWN subscription with his credit card, I mostly waited the one week to read the paid articles (and, in fact, I'm still subscribed to the LWN mailing list which automatically tells you whenever a paid article becomes free).

argulane• 4 years ago

If you have the means then I would recommend subscribing https://lwn.net/subscribe they are pretty much the only foss news outlet that published so high quality articles.

abledon• 4 years ago

I know the article is serious, but it has the writing-style of an onion/babylonbee article lol.

esp the last paragraph: >As time ran out, Matthew Wilcox asked whether kernel developers should be writing idiomatic Rust code, or whether they will be writing "C in Rust". Ojeda answered that code might be more C-like toward the beginning; adoption of more advanced features (such as async) might take longer. Gleixner asked what could be done to prevent developers from using unstable features (once the features used by the kernel are stabilized); the answer was to specify the version of the compiler to be used with kernel development.

Longlius• 4 years ago

I'm surprised Torvalds is so sanguine about Rust. Maybe it's age but I remember his epic gamer rant about how he would never allow C++ in the kernel. Either Rust is significantly better than C++ in his view, or he's accepted that C is likely not going to be the dominant systems programming language forever and opted for the least dangerous alternative.

Guess I should start my 5th attempt at finally getting good at Rust.

bestouff• 4 years ago

Get good at Rust, the answer will come.

vbtemp2• 4 years ago

To the OP: This is the correct answer.

pjkundert• 4 years ago

I’m a C++ guy from back in the 90’s CFront days.

I love Rust; I’d do any new project in Rust instead of C++.

It’s worth the effort to learn.

agumonkey2• 4 years ago

How did it affect your coding ? do you use different patterns / idioms ? are you more ambitious due to less bugs ? do you design / prototype differently ?

ardel95• 4 years ago

Rust is significantly better than C++. And I say that as somebody who has done many years of C++ development.

lijogdfljk• 4 years ago

I love that it draws so many different types. I'm a Python/JS/Go programmer historically. Rust is closer to Go to me. Of course it's not Go, but i'm far from a migrating C/C++ dev. Hell i only know a pinch of C++ from when i was first learning the trade.

runevault• 4 years ago

I always had the impression Linus saw C++ as C but more complicated/worse. Rust is DIFFERENT by creating guarantees that are harder to get around and if you do try by using the functionality available in unsafe blocks it is clear as day to anyone and everyone who looks at the code.

Tolexx• 4 years ago

Guess I should start my 5th attempt at finally getting good at Rust.

I guess i should do the same too but I have really and man the language is not easy to grok. These days it also feels like if you don't know Rust, you are somewhat "old school".

vbtemp2• 4 years ago

Either Rust is significantly better than C++ in his view

Do you remember the Paul Graham essay on the Blub programming language? (http://www.paulgraham.com/avg.html)

C++ is kind of like the Blub language. If you know C++ well enough, you can do anything you need to do, so something like Rust just seems too esoteric and weird and pretentious.

Fortunately, I spent many years using Rust recently (coming from primarily a C background) and when I'm on a project using C++, it just feels too weird and rudimentary and horribly over-complicated.

To end with a quote from the PG article:

_But when our hypothetical Blub programmer looks in the other direction, up the power continuum, he doesn't realize he's looking up. What he sees are merely weird languages. He probably considers them about equivalent in power to Blub, but with all this other hairy stuff thrown in as well. Blub is good enough for him, because he thinks in Blub._

Jensson7• 4 years ago

Paul Graham favored Lisp, which gives programmers limitless freedom to implement whatever they want, Rust is the opposite side of that.

C++ is full of footguns and it is easy to make mistakes, but it is also one of the most powerful and expressive languages out there.

twic3• 4 years ago

Interesting. In my experience, seasoned C++ programmers grasp the advantages of Rust very quickly. Not all of them are convinced that the advantages outweigh the practical problems, but they don't deny the existence of the advantages.

I would have said C is firmly a Blub, though.

jahewson• 4 years ago

significantly better than C++

You say that like there’s some scalar measure of “better”.

tialaramex9• 4 years ago

Surely it's implied in context that this is better for developing the Linux kernel, by Linus and Linux kernel developers ?

masklinn2• 4 years ago

Either Rust is significantly better than C++ in his view

Probably this one:

- Rust has the clear goal of memory safety, which is an active issue in the kernel (see: Kees Cook's Kernel Self Protection Project), even more so when it comes to drivers which are a lot more variable in quality and have very variable levels of oversight, which is why those are the primary use case.

- Rust has a much more expressive type system, especially without the need to go into template weeds.

- Rust has a lot less implicit behaviours.

- Rust features tend to be more orthogonal and mis-interact less with one another.

- While Rust has a fairly steep learning curve, short of unsafe it doesn't kneecap you, if it doesn't like what you're doing (right or wrong) it tells you, it doesn't go off into the weeds doing the wrong thing.

btreesOfSpring• 4 years ago

All great points. It is also a young enough language that significant kernal dev can have significant influence on the future of the language itself. CPP's future seems to be predicated upon its past adoption.

tialaramex9• 4 years ago

And while there are no shortage of C++ programmers who've said "just use C++" over the years, not too many actually lifted a finger to make it happen - the people who wanted Rust for Linux actually built Rust for Linux.

https://rust-for-linux.github.io/docs/kernel/

Out of that, core is basically just what you get from Rust's core itself, alloc is similar to Rust's provided alloc but is fairly heavily customised to meet the requirements of the kernel and of Linus. But kernel is custom code for the kernel specifically. Overall that's a lot of effort.

We didn't see a similar effort for C++. Would that have overcome Linus' reluctance? Well, it can't if it doesn't exist.

endorphine• 4 years ago

My view is he's being pragmatic. Why switch from C to C++? The expected value from such a change isn't worth the effort.

Rust though? That could make the kernel more secure in the long-term and rule out a whole class of bugs. This makes sense.

unrealhoang• 4 years ago

I think the main point is Rust takes much less time to learn to be able to contribute to a complex code base at the caliber of Linux. With safe abstraction on top of unsafe, the seasoned/guru developers can build building blocks for intermediate/beginner to confidently use and contribute on the higher level.

sam0x17• 4 years ago

Guess I should start my 5th attempt at finally getting good at Rust.

The thing that finally made Rust click for me was doing a usb HID project. Suddenly all the machinery around ownership and multi-threading made sense and I was able to get things done, and I could suddenly also do all the things I previously attempted to do in Rust.

brundolf2• 4 years ago

I think it's both. I remember reading at one point (not sure if it was quote or speculation) that the difference is C++ adds a lot of abstractions that:

- Have hidden costs

- Have sneaky failure-modes

C just... doesn't add any abstractions. And then Rust adds abstractions, but makes a concerted effort to avoid ones that have these two problems

nyxtom• 4 years ago

I think a lot of it comes from the fact that he realizes there is a demand for kernel maintainers in the future and the C expertise isn't there as well. If you can get maintainers to come in that are Rust devs then they can work on things without introducing significant issues due to some of the default safety guarantees and guardrails that Rust provides while at the same time having approximate to near performance parity

keehun• 4 years ago

Does anyone have good resources to learning rust? I have a fairly good working knowledge of C, C++, and Swift.

I understand there's the Rust Book, Rustlings, and Rust By Example at https://www.rust-lang.org/learn. Are there other good resources? Does anyone have a strong suggestion on which of those official resources I should start with?

surrTurr• 4 years ago

If you‘re willing to spend money on a great book that teaches you all the basics of Rust, I can only recommend Programming Rust, 2nd edition (OReilly). It includes tons of examples, code snippets and is also a good resource for looking up stuff once you‘re done reading it. It also includes many passages comparing C++ with Rust, so that might be useful for you. It‘s a great book for anyone, who already knows a few programming languages. Programming beginners, should look elswhere tho.

topbanana• 4 years ago

I think there's a missing book - 'Thinking In Rust'. The mechanics of borrow-checking, lifetimes, std lib, tokio can all be learned quite quickly. It's the muscle memory of techniques and design patterns that takes the longest to re-learn.

dkryptr• 4 years ago

I started with the book to learn the fundamentals. I didn't get very far before I decided to rewrite an existing API I had in Python into Rust.

It definitely took some time and learning but now I have 6 Rust API microservices, a few scheduled/queue-reading services, and a shared library for common models/utils/providers.

atoav• 4 years ago

A few things. The hardest thing about Rust is to forget many of tthe paradigms or aesthetical ideas other languages hammered into your brain. Programming Rust like you would program Python, Javascript or C++ is going to give you a hard time, but each time in a different way. If you program Rust in the Rust way, suddenly things become easy.

That means what the best way to start is depends on your background. In my eyes you can do nothing wrong with going through the rust book you already mentioned. Having read 4 different Rust books I have to say this introduction in combination with "Programming Rust" by Jim Blandy, Jason Orendorff, Leonora F. S. Tindall is probably the best way to get started. https://www.oreilly.com/library/view/programming-rust-2nd/97...

masklinn2• 4 years ago

Does anyone have a strong suggestion on which of those official resources I should start with?

The rust book. Reading it start to finish provides a very good basis.

And if you're tempted to "learn by doing" with the usual linked lists, go and read "learning rust with entirely too many linked lists": https://rust-unofficial.github.io/too-many-lists/

Rust has enough rare or novel concepts that trying to learn on the job with no book learning whatsoever is really reserved for the rarefied few. Though I understand that a lot of C and C++ concepts port quite easily, it also differs from those languages (especially C++ for advanced features) that doing a bit of a reset will avoid future pains e.g. Rust's concepts of copy, move, references, ... are very different from C++'s and going in half-cocked will be frustrating.

Jensson7• 4 years ago

As you already know C and C++ this one is great: https://web.mit.edu/rust-lang_v1.25/arch/amd64_ubuntu1404/sh...

Just go through everything in "3 Syntax and Semantics" and "4 Effective Rust", you already know how to program C and C++ so that is all you should need to know. Took me a few days.

d12bb• 4 years ago

I learned it combining the Book and Rustlings. There's a table in the exercises directory mapping the exercises to the chapters of the book, so I'd every day do the exercises for yesterdays chapter first before todays chapter to have some sort of spaced repetition. For more material, check out https://github.com/jondot/rust-how-do-i-start

8jy89hui• 4 years ago

Start with the Rust book. Then "Rust for Rustaceans" by John Gjengset is a fantastic resource to learn about more than just the basics. He posts hour long videos on his youtube where he explains the concepts that he mentions in his book. I found that reading a chapter and then watching one of his videos really taught me about /why/ the language is as it is, rather than just learning to get the compiler to be happy. (It also includes chapters about API design and what to look out for if you are going to publish your own crate)

mfashby• 4 years ago

I enjoyed the rustlings puzzles. They reference the book a lot, and I tend to learn better by doing (at least a little bit of doing).

tialaramex9• 4 years ago

So, with "good working knowledge" of C++ and Swift, you probably aren't going to have too much trouble with the basics, if anything you might worry that there's some subtlety which you're missing when actually nope, whatever just seemed easy maybe actually was that easy.

[ For example move really is as simple as Rust makes it look, it is a headache in C++ because they added it to a finished working language ]

Several people suggested books/ web pages let me suggest some videos:

Jon Gjengset's "Crust of Rust" Youtube videos are good once you reach the point where you can write more than "Hello, world!" with some confidence but certain specific things aren't clicking.

https://www.youtube.com/watch?v=rAl-9HwD858&list=PLqbS7AVVEr...

For example, Jon does a whole video on lifetime annotations, which are something you won't see in most popular languages, but also one on Rust's iterators, which are a familiar concept from both C++ and Swift but don't work quite like either (they're very different from C++ iterators, your Swift experience will help more).

I recommend watching specifically a handful on topics you don't feel you understood well, although you could watch all of them especially if you just enjoy Jon's style and have the free time. They're not fast paced, if you wanted "Rust in 60 minutes" this is not that, but each one is actually writing carefully chosen code that runs and talking through what's going on, not just clicking through slides.

davidatbu• 4 years ago

This is my journey so far:

(1) Go through the Rust Book,

(2) Do "exercises" (small side projects, advent-of-code, rustlings. ..etc),

(3) Go through "Rust for Rustaceans". It's specifically targetted at "intermediate Rust developers". If you have "good working knowledge" of C and C++, I will wager that you will absolutely love this book, because, to me, (1) it explained so much of Rust's design choices, and (2) you will learn so much so quickly.

javier_e06• 4 years ago

Rust covers an infinitesimal footprint of the C/C++ code out there running the world. Rust won't replace anything substantial or meaningful in the embedded world for the time being.

mlindner3• 4 years ago

I would say that that statement is a tautology of any new language and was true also of C and C++ when they were respectively introduced.

badrabbit• 4 years ago

I love this, but also very surprised Trovalds is ok with a language that changes so much. But with git and all it maybe easier to manage just as if you would C89/99/11.

w10-1• 4 years ago

I appreciate the write-up representing the flow of discussion, not just the outcome, for the insight it carries about how decisions move through the group.

It appears Linus has said it is happening, and a bit how it should happen, but has not given strong guidance or even parameters on how long any experiment would last.

Instead, he says the maintainers can decide to accept or reject or use Rust however they like. He doesn't even say they should state their policy, so they can let people try and still reject them.

Giving maintainers the ultimate discretion I think tracks the incentive system in the kernel: maintainers, who do boatloads of work, get to be deciders. That also gives companies a strong incentive to employ maintainers.

On the tools front, Linus expressly said he wants not just an anointed compiler on kernel.org, but compilers from the distributors. He's driving Rust normalization and platform adoption as a condition of use, even though the kernel historically adopts a fairly narrow (if not archaic) set of tools.

Somehow this all seems like a bunch of rafts tied together with (platform) boats pushing at the edges -- somewhat tenuous, hard to drive, but vaguely heading in the right direction.

As others have mentioned, it's interesting that Linus and many maintainers are getting older, and instead of getting freedom and flexibility in their maturity, they can look forward to more low-level bouts of increasing complexity. When they talk about the integration of Rust, they're clearly anticipating it might not fully happen until after they're gone. That, too, could change the incentives.

Animats2• 4 years ago

Whatever happened to that effort to write a QNX-compatible kernel in Rust. The QNX kernel is tiny. That should be working by now.

That's what you really want in Raspberry Pi sized projects.

Kukumber• 4 years ago

Instead, Bottomley suggested that, rather than bringing in Rust, it might be better to just move more Rust-like features into C. Ojeda said that he has actually been working with the C language committee to push for that to happen, but any such change will take a long time if it happens at all. Christoph Hellwig said that this sort of change will have to happen anyway unless the plan is to rewrite the whole kernel in Rust; he was not pleased at the idea of rewriting working code in a new language. Perhaps the sparse static analyzer could be enhanced to do more Rust-like checking, he said. Ojeda answered that the result of such efforts would be like having Rust — but much later.

Some reasonable people in this sea of sectarian crabs, glad to hear that

klysm• 4 years ago

You’ve added nothing to the quote but vitriol.

phendrenad2• 4 years ago

I believe that in 10 years, we'll look back at Rust in the kernel and see it as a big disaster. Some people will blame the kernel team (they didn't adopt Rust well enough), and some will blame Rust itself (writing a non-toy kernel in it hasn't been tried before).

But heck, we should TRY it. So I'm glad this is finally happening.

oxplot• 4 years ago

For those with better understanding of where zig is at, why not zig for kernel development? Is it merely because it's still in heavy dev?

mlindner3• 4 years ago

I think the answer to your question is also the answer to the question of how Zig achieves memory safety at compile time.

Jensson7• 4 years ago

Yeah, Rust is C with a sane version of C++'s RAI, Zig is C with a sane version of C++'s templates. Having Turing complete duck typed template metaprogramming is very fun and useful, but I understand why you wouldn't want that in huge security critical codebases.

menaerus3• 4 years ago

If this happens, and it seems inevitable at the moment, I personally see this resulting in a potential decline of a Linux kernel in the next few years.

Just a few questions off top of my head:

* How much will it affect the development velocity? Existing devs (and maintainers) will at least need to learn how to read and debug Rust code. New devs will likely have no good competences in C which is 99% of the codebase so how good/quickly will they be able to blend in?

* How many existing kernel developers will community potentially loose because of the increased complexity?

* How much new talent will it actually be able to attract? Now people will need to learn two completely different languages and their vastly different ecosystems.

* Does Rust have high enough entry bar to produce competent kernel developers? Sometimes high entry bar acts as a very good filter to end up in a pool with a highly skilled (and motivated) engineers.

* Coordination of development efforts is now going to become much more complex. This includes writing new features (C or Rust or both?), maintaining the old ones (rewrite pieces in Rust or not?) but also code-reviews which will now pose a bigger challenge given that it will consist of mixed Rust+C code.

Having worked in codebases which mixed C and C++, I already understand what kind of issues mixing C and Rust is going to produce. And I expect them to be much more accentuated because distance(C, C++) << distance(C, Rust).

Next steps for Rust in the kernel

Related Stories