Progress toward a GCC-based Rust compiler
lwn.net
lwn.net
> Cohen's EuroRust talk highlighted that one of the major reasons gccrs is being developed is to be able to take advantage of GCC's security plugins. There is a wide range of existing GCC plugins that can aid in debugging, static analysis, or hardening; these work on the GCC intermediate representation
> One more reason for gccrs to exist is Rust for Linux, the initiative to add Rust support to the Linux kernel. Cohen said the Linux kernel is a key motivator for the project because there are a lot of kernel people who would prefer the kernel to be compiled only by the GNU toolchain.
That explains why you’d want GCC as the backend but not why you need a duplicate front end. I think it’s a bad idea to have multiple front ends and Rust should learn from the mistakes of C++ which even with a standards body has to deal with a mess of switches, differing levels of language support for each compiler making cross-platform development harder, platform-specific language bugs etc etc.
> A lot of care is being put into gccrs not becoming a "superset" of Rust, as Cohen put it. The project wants to make sure that it does not create a special "GNU Rust" language, but is trying instead to replicate the output of rustc — bugs, quirks, and all. Both the Rust and GCC test suites are being used to accomplish this.
In other words, I’d love gccrs folks to explain why their approach is a better one than rustc_codegen_gcc considering the latter is able to achieve this with far less effort and risk.
Rust should learn from the mistakes of C++ and C, which are one of the longest lasting, biggest impact, widely deployed languages of all time?
It's confusing when people think language standards are bad, and instead of saying this code is C99 or C++11, they like saying "this code works with the Rustc binary / source code with the SHA256 hash e49d560cd008344edf745b8052ef714b07595808898c835f17f962a10012f964".
Literally the only reason to specify via a hash* would be if you were using such a bleeding edge feature that it was only merged in the last 48 hours or so and no nightly versions had been cut.
*Or I suppose you don't trust the supply chain, and you either aren't satisfied with or can't create tooling that checks the hash against a lockfile, but then you have the same problem with literally any other compiler for any language.
Now, none of these links are the specifications for the languages listed, but they all copiously link to the specification where applicable. In rare situations, I have gone directly to the specification. That’s usually if I’m trying to parse a subset of the language or understand an obscure language feature.
I would argue no one writes code against a compiler. Sure we all validate our code with a compiler, but a compiler does not tell you how the language works or interacts with itself. I write my code and look for answers to my questions in the specification for my respective languages, and I suspect most programmers do as well.
[0]: https://en.cppreference.com/w/
[1]: https://docs.oracle.com/javase/specs/index.html
[2]: https://developer.mozilla.org/en-US/docs/Web/JavaScript
The Project is working on a specification. The Foundation hired someone for it.
A Rust spec done purely on paper ahead of time would be the contents of the accepted RFCs. The final implementation almost never matches what was described because during the lengthy implementation and stabilization process we encounter multitude of unknown unknowns. The work of the spec writers will be to go back and document the result of that process.
For what is worth the seeming dragging of feet on this is because the people that would be qualified and inclined to doing that work were working on other, more pressing matters. If we had had a spec back in, let's say, Rust 1.31, what would have that changed, in practice?
If the compiler claims to follow the specified version of the spec, and it doesn't, you file a compiler bug.
And then use the subset that it supports, perhaps by using an older spec if it supports that fully. Perhaps looking for alternative compilers that have better/full coverage of a spec.
"Supports the spec minus these differences" is still miles better than "any behaviour can change because the Rust 2.1.0 compiler compiles code that the Rust 2.1.0 compiler compiles".
> And then use the subset that it supports, perhaps by using an older spec if it supports that fully. Perhaps looking for alternative compilers that have better/full coverage of a spec.
If you encounter rustc behavior that seems unintentional, you can always file a bug in the issue tracker against the compiler or language teams[1]. Humans end up making a determination whether the behavior of the compiler is in line with the RFC that introduced the feature.
> "Supports the spec minus these differences" is still miles better than "any behaviour can change because the Rust 2.1.0 compiler compiles code that the Rust 2.1.0 compiler compiles".
You can look at the Rust Reference[2] for guidance on what the language is supposed to be. It is explicitly not a spec[3], but the project is working on one[4].
1: https://github.com/rust-lang/rust/issues?q=is%3Aopen+is%3Ais...
2: https://doc.rust-lang.org/reference/introduction.html
3: https://doc.rust-lang.org/reference/introduction.html#what-t...
4: https://blog.rust-lang.org/inside-rust/2023/11/15/spec-visio...
How should the Rust Specification represent the type inference algorithm that rustc uses? Is an implementation that can figure out types in more cases than described by the specification conformant? This is the "split ecosystem concern" some have with the introduction of multiple front ends: code that works on gccrs but not rustc. There's no evidence that this will be a problem in practice, everyone involved seems to be aligned on that being a bad idea.
> Any divergence or disagreement is a failing of the implementation.
Specs can have bugs too, not just software.
I think this may be a large point of divergence with regard to my understanding and position on language specifications. I assume when posts have said Rust is getting a spec, that that included a formal, ‘mathematically’ proven language definition of the core language. I am aware that is not what C or C++ or JS includes in a spec (and I don’t know if that is true for Ada), but I was operating under the assumption the Rust’s inspiration from Ocaml and limited functional-esque stylings that the language would follow the more formalized definition style I am used to.
There’s value in multiple implementations ensuring code isn’t bug-compatible, but at the same time in C and C++ there’s plenty of unnecessary differences and unspecified details due to historical reasons, and the narrow scope of their specs.
- My Android phone
- OpenJDK on my laptop
- leJOS for a university robotics course to run on Lego robots
- Smart cards, I have a couple in my wallet
Perhaps I'd call the Lego robot stuff obsolete, certainly not the Android userspace or smart cards though.
For example C89 and C99 are pretty old, and modern C has a lot that is different from them. But they still get targeted and deployed, and enjoy a decent following. Because even in 2024, you can write a new C89 compiler and people's existing and new C89 code will compile on it if you implement it right.
PTC, Aicas, microEJ, OpenJ9, Azul, GraalVM are a couple of alternative JVM implementations.
BTW GraalVM is based on OpenJDK so I don’t really understand your point there. It’s not a ground-up reimplementation of the spec.
Doesn't matter how popular they are, they exist because there is a business need, and several people are willing to pay for them, in some cases lots of money bags, because they fulfill needs not available in OpenJDK.
I have been following GraalVM since it started as MaximeVM on Sun Research Labs.
JavaScript itself is a very simple language with most of the complexity living in disparate runtimes and that there are multiple runtime implementations is a very real problem requiring complex polyfills that age poorly that are maintained by the community. For what it’s worth TypeScript has a single implementation and it’s extremely popular in this community.
Java is probably the “best” here but really there’s still only the Sun & OpenJDK implementations and the OpenJDK and Oracle are basically the same if I recall correctly with the main difference being the inclusion of proprietary “enterprise” components that Oracle can charge money for. There are other implementations of the standard but they’re much more niche (e.g. Azul systems). A point against disparate implementations is how Java on Android is now basically a fork & a different language from modern-day Java (although I believe that’s mostly because of the Oracle lawsuit).
Python is widely deployed & CPython remains the version that most people deploy. Forks find it difficult to keep up with the changes (e.g. PyPy for the longest time lagged quite badly although it seems like they’re doing a better job keeping up these days). The forks have significantly less adoption than CPython though.
It seems unlikely that independent Rust front end implementations will benefit it’s popularity. Having GCC code gen is valuable but integrating that behind the Rust front-end sounds like a better idea and is way further along. gccrs is targeting a 3 year old version of Rust that still isn’t complete while the GCC backend is being used to successfully compile the Linux kernel. My bet is that gccrs will end up closer to gcj because it is difficult to keep up.
It's probably a matter of time until there's a TypeScript compiler implemented in Rust. But the surface area of the language is pretty big, and I imagine it will always lag behind the official compiler.
> Forks find it difficult to keep up with the changes
That's interesting to think of multiple implementation of a language as "forks" rather than spec-compliant compilers and runtimes. But the problem remains the same, the time and effort necessary to constantly keep up with the reference implementation, the upstream.
Yes. Roslyn, Mono and some Mono-like thing from Unity to compile it into C++.
> Mono used to be a separate thing
Mono is still a thing. The last commit was around 3 months ago.
> multiple runtime implementations is a very real problem requiring complex polyfills
You can target a version of the spec and any implementation that supports that version will run your code. If you go off-spec, that's really on you, and if the implementation has bugs, that's on the implementation.
> TypeScript has a single implementation
esbuild can build typescript code. I use it instead of tsc in my build pipeline, and only use tsc for type-checking.
> [Typescript is] extremely popular in this community
esbuild is extremely popular in the JS/TS community too. The second most-popular TS compiler probably.
> [Java has] only the Sun & OpenJDK implementations
That's not true. There are multiple JDKs and even more JVMs.
> Java on Android is now basically a fork & a different language from modern-day Java
Good thing Java has specs with multiple versions, so you can target a version that is implemented by your target platform and it will run on any implementation that supports that version.
> Python is widely deployed & CPython remains the version that most people deploy. > The forks have significantly less adoption than CPython though.
That is because Python doesn't have a real spec or standard, at least nothing solid compared to the other languages with specs or standards.
> It seems unlikely that independent Rust front end implementations will benefit it’s popularity.
It seems unlikely that people working on an open-source project will only have the popularity of another open-source project in mind when they spend their time.
Roslyn is more like the next gen compiler and will be included in Mono once it’s ready to replace msc. I view it as closer to polonius because it’s an evolutionary step to upgrade the previous compiler into a new implementation. It’s still a single reference implementation.
> Mono is still a thing
I think you misunderstood my point. It had started as a fork but then Microsoft adopted it by buying Xamarin. It’s not totally clear to me if it’s actually a fork at this point or if it’s merged and shares a lot of code with .NET core. I could be mistaken but Mono and .Net core these days also share quite a bit of code.
> rebuild can build typescript code
Yes, there are plenty of transpilers because the language is easy to desugar into JavaScript (intentionally so - TS stopped accepting any language syntax extensions and follows ES 1:1 now and all the development is in the typing layer). That’s very different from a forked implementation of the type checker which is the real meat of the TS language compiler.
> The second most popular TS compiler probably
It’s a transpiler and not a compiler. If TS had substantial language extensions on top of JS that it was regularly adding, all these forks would be dead in the water.
> That’s not true. There are multiple JDKs and even more JVMs
I meant to say they’re the only ones with any meaningful adoption. All the other JDKs and JVMs are much more niche and often benefit from living in a niche that is often behind on the adoption curve (i.e. still running Java 8 or something or are willing to stay on older Java versions because there’s some key benefit in the other version that is operationally critical).
> Good thing Java has specs with multiple versions, so that you can target a version…
Good for people implementing forks, less good for people living within the ecosystem in terms of having to worry about which version of the compiler to support with their library. For what it’s worth Rust also has language versions but it’s more like an LTS version of the language whereas Java versions come out more frequently & each implementation is on whatever year they wanted to snapshot against.
Mono shares a lot of code with .NET (Core) but is mostly limited to the standard libraries and compiler. Mono is still its own separate implementation of the CLR (runtime/"JVM") and supports much more platforms than .NET (Core) today.
Mono has approx. 0.x% market share outside Unity. Also Mono is used by .NET Core for Blazor WASM iirc.
Let's don't compare this scenario of sane world with the mess that exists in C++ world.
I don't know if that's totally fair. I remember that it took quite a while for C++ compilers to actually implement all of C++11. So, it was totally normal at the time to change what subset of C++11 we were using to appease whatever version of GCC was in RHEL at the time.
Can be true at the same time as:
c and c++ have made mistakes, and have had issues as a result of bad choices
The later should be learned from by any language in a place to not make those same mistakes. We call this technological progress, and its OK.
I don't think language standards are "bad".
"Learn from the mistakes of X" doesn't mean X is bad, it means X made mistakes.
They are. You quoted them doing it, the care taken not to become a superset. The problem with C/C++ stemmed from compiler vendors competing with each other to be "better" than their peers.
Multiple front ends of an implementation of a language usually shakes out a bunch of bugs and misimplementations. That's the primary benefit of having multiple front ends IMO.
Yea but this is Linux and OSS. NIHisms are fucking rampant everywhere.
I give it a year before gccrs announces a new "gnurs" mode with language extensions.
Even as someone who probably will never use gccrs, I think having more than one implementation of Rust is a good thing for the language (for all the usual reasons that get cited). In the long term, I'd love for Rust to eventually to be able to specify its semantics in a way that isn't tied to rustc, which is a moving target and gets murky when you consider implementation-specific bugs.
There are some exceptions to this. Although there are possibly good reasons for such features moving slowly, a competing compiler could popularize them before they are stabilized in rustc. One example is generators[1] but there's a longer list in The Unstable Book.[2]
[1] https://github.com/rust-lang/rust/issues/43122
[2] https://doc.rust-lang.org/stable/unstable-book/language-feat...
Stabilizing things from rustc is a recipe for disaster as if rustc builds something new you are in a bad place.
Linux tends to shun unfree software which is a very different take. Buy or build is roughly money vs time (aka money). Software freedom is not the same thing.
Also why does software freedom lead to divergences? Certainly GCC partook in the arms race around C++ but recently they have been pretty good about aiming for full standard support as the goal. (The exception being proposed features but that is unavoidable when C++ prefers battle tested proposals)
My point is outside of instances where everyone agrees something has to change you need a compiler branch at the minimum. This means the compiler you changed will get that feature first.
Given it takes years to implement the full standard this leads to divergence between the compilers in standards compliance.
Honestly all totally workable but makes talking about "standard C++" hard.
I do not disagree, yet at the same time, having the same set of switches across different languages is nice, too.
Many places deliberately compile with multiple compilers as part of the build/test pipeline, to benefit from more compiler warnings, diagnostics etc.
You could probably do most of it with conservative linting and some in-house knowledge of portability issues, non-standard compiler extensions, etc.
It's typically a lot easier to do this for different compiler, same target, than different targets.
It isn't easy and honestly the idea of "just don't use MSVC for a while" is strange to me. Sure you can compile with any of them but almost certainly you are going to stick to one for a given use case.
"This release is on a different compiler" isn't something you do because of a bug. Instead your roll back a version or avoid using the unsupported feature until a fix is released.
The reason is as much as they are supposed to do the same thing the reality is bugs are bugs, e.g. if you invoke undefined behavior you will generally get consistent results with a given compiler but all bets are off if you swap. Similarly it is hard not to rely on implementation defined behavior without building your own standard library which specifically defines that behavior across compilers.
And of course plenty of OSS libraries support many compilers even beyond the main three.
If you are using intrinsics or other architecture-specific features, there is a similar practice of requiring full CI pipelines for at least two CPU architectures. Again, it occasionally finds interesting bugs.
For systems I work on we usually have 4 CI pipelines for the combo of GCC/LLVM and ARM/x86 for these purposes. It costs a bit more but is generally worth it from a code quality perspective.
(And then we could ship the product on Windows, so that was nice.)
MSVC is a bit more quirky... At times its standards compliance has been patchy but I think it's ok right now.
Also the article says otherwise > The project wants to make sure that it does not create a special "GNU Rust" language, but is trying instead to replicate the output of rustc — bugs, quirks, and all
And running into these kinds of questions, both in a single project that evolves over time (rustc) and a separate project, help to feed information back into what you need for a formal specification; which is something that is being planned out. Having a second implementation can help find these areas where you might need to be narrower or broader in your specification, in order to clarify enough for it to be implementable or broaden it enough to accommodate reasonable implementation differences.
They are not trying to diverge in language implementation, but there will always be simple compiler bugs, which may crop up in one implementation but not another. For instance, some LLVM optimization pass may miscompile some code; gccrs wouldn't necessarily be trying to recreate that exact bug. I think that "bugs, quirks, and all" really means that they aren't trying to fix major limitations of rustc, such as introducing a whole different borrow checker model which might allow some programs that current rustc does not. They're trying to fairly faithfully recreate the language that rustc implements, even if some aspects might be considered sub-optimal, but they aren't going to re-implement every ICE and miscompilation, those are places where there could be differences.
Theory is cool, but in practice other compiler has its own quirks too
How about having one compiler which is used by all developers, which increases chances of bugs getting caught faster and fixed faster meanwhile you just use workaround?
In reality if you hit a compiler bug in Rust or Go (or any other language with one main impl like Python, Ruby…) you would file a report and do one of two things - downgrade the compiler (if it’s an option) or write the code a different way. Compiler bugs in these languages are rare enough that this approach works well.
That said, for people who are really keen on a spec, there is one being worked on (https://blog.rust-lang.org/inside-rust/2023/11/15/spec-visio...).
But this GCCRS effort doesn’t get you any closer to your ideal C/C++ style workflow because they are committing to matching the semantics of the main compiler exactly. Bugs and all. And that’s the way it should be.
The Rust ecosystem becomes worse if I have to install a different toolchain and learn a different build system with a different compiler for every new project I interact with. And after all that extra effort it turns out there are subtle differences between implementations. My developer experience is just worse at that point. If I wanted to code like this, I would code in C++ but I don’t.
Perhaps because it’s currently not possible in Rust, so users don’t really see the advantages yet.
Multiple implementation are always a good thing. They bring in diversity of thought and allow for some competition in the implementation. Having multiple slightly different implementations also gives more clarity to a spec.
This isn’t the kind of stuff that users tend to think about until it’s already there.
However there isn't enough benefit to multiple language _frontends_ that I can see that outweighs the many downsides.
Not only do you have to contend with the duplication of effort for implementing two or more language frontend projects, you now create ecosystem splits when inevitably the implementations do not match their behavior.
The biggest example of multiple frontends, C++, is not one language. C++ is Clang-C++, GNU-C++, MSVC-C++, and so-on. All of these implementations have subtle incompatibilities with how they handle identical code, or they have disjoint feature matrices, or they have bugs, and so on. Multiple frontends is IMO one of the _worst_ parts of C++ because it makes writing truly portable code almost impossible. How many new C++ features are unusable in practice because one of the main 3 toolchains hasn't implemented it yet?
Rust, on the other hand, everything is portable by default. No #ifdef CLANG, no wrangling with subtly incompatible toolchains. If I write code I can trust it will just work on any architecture Rust can build for, ignoring implementation bugs and explicitly platform specific code.
I have a hobby game engine I work on in my spare time that was developed exclusively on x86-64 for years and I ported it to run on Apple Silicon ARM in two afternoons. Rust just _worked_, all the time was spent on my C/C++ dependencies and new code needed for MacOS support.
I simply don't see how nebulous, difficult to quantify benefits of competing implementations can outweigh the practical advantages of truly portable code that comes from a single implementation. The ecosystem benefits to a uniform language, and the benefits to my sanity with being able to trust my code while compile everywhere I need it to.
What we're saying is that everyone actually using Rust today is fairly happy with how the language is. That's why Rust has had an 80%+ approval rating among Rust developers since its creation and C++ is at 49% (https://survey.stackoverflow.co/2023/#section-admired-and-de...).
This suggestion of alternate frontends still has no good technical reason driving it, just people pattern matching on their past experience. If it happens, all it will do is lower Rust's approval to C++'s level.
* Libraries * ABIs * Build system(s) * Platform compiler support...
In compiled/interpreted land, most if not all of these are solved problems. Even for libraries, usually dependencies can be reliably downloaded and recompiled (if they are not compiled on the fly).
I don't really care if you can create a "portable header" library when the ecosystem is sharded, and "portable" only holds true for some subset of c++ compiler/platforms.
I'm not specifically commenting on ifdefs, because ifdefs are only a small part of the issue (caused by multiple compiler frontends).
In reality, the people wasting their time to produce a redundant ground-up rewrite of existing functionally could have spent their time and effort on something else that actually moves the world forward.
And apparently there are people who want to do the work, because we have work on a couple of implementations.
You can choose to use your one true implementation, no one is stopping you. But it’s kind of weird to assume that people working on another one are detracting from yours. That’s not usually the way open source software works.
Sincerely,
- Rust adopter
Monopolies are bad when they stifle progress, but rust's written under a free software license. At any point, anyone can take the project and fork it into a new one, can carry on the progress under a new flag.
Unlike traditional monopolies, the rust project's priorities aren't extracting profits or holding power, but rather in improving the language, and they're open to anyone (well, anyone who's tolerant of others) joining and helping.
The argument here is that we will end up with a better outcome if we work together as a community, rather than if we fragment just for some perceived ideal of multiple implementations.
While you're criticizing the rust monopoly, you might as well go and also yell at the group of friends voluntarily picking up trash off the sidewalk for free since they're working together as one monopoly of friends instead of creating two competing cleaning groups.
the way you are using "free" is undefined behavior. "Free" as in speech software, the F in FOSS, is associated with copyleft FSF and GNU, and not like BSD, MIT, and Apache, the open source in fOSs, which is what Rust is.
The fact that you knew I was referring to rust's license means my English was in fact not undefined, but exactly as clearly defined as one can hope english to be, i.e. we both thought it meant the same thing.
In my opinion, in context, "free software license" is also a clear and common phrase, not undefined behavior.
Essentially, I think everyone has agreed that "free software", unless you add more qualifiers, will refer to the FSF definition of free software https://www.gnu.org/philosophy/free-sw.html
The FSFs definition includes both the licenses rust is licensed under: https://www.gnu.org/licenses/license-list.html#Expat
The FSF would happily call BSD, MIT, and Apache "Free Software Licenses", and in fact do so.
"Free software means that the users have the freedom to run, edit, contribute to, and share the software."
those licenses allow derivatives which users cannot edit and contribute to thus they don't guarantee the 4 freedoms of free software.
Please read: https://www.gnu.org/licenses/license-list.html .
For example:
Apache License, Version 2.0
This is a free software license, compatible with version 3 of the GNU GPL.
Expat License [this is the MIT license]
This is a lax, permissive non-copyleft free software license, compatible with the GNU GPL.
Last I heard `open source` was used to refer to code not projects.
Every project has its "decision makers" (no matter the structure or titles) that have the power to dictate the direction and vision. You can literally see it every day in many projects, not trying to single out Rust or anything. The power is not distributed evenly, not even within the elite inner circle.
There's no 'together as a community` when it's followed by a clause like `we need to follow the one direction (X) has set`.
What power does `the community` exactly have? to open a PR and get denied? is to fork? Can the community ellect leaders and change course? You said it yourself, anyone that want to have their voice heard, they have to climb up the ladder and join the inner circle. You're descriping `Power elite` and not `community`. Even forks don't work in real life, what happens is the project usually die slowly and everyone move to the next thing.
I'm not dismissing the structure itself, it's a good pragmatic approach, it's just that you're twisting and polishing it. In short:
You: Everyone is welcome. No need to fragment. Rust is run by community.
Rust's governance: We rule. We also pick any new members. Community doesn't have much say about anything.
> The project wants to ... replicate the output of rustc — bugs, quirks, and all.
I don’t agree, mind you, but someone reading your comment with this mindset will probably be reinforced rather than swayed.
It's not plausible that that will that a similar entity will take over the HTML/CSS/JS space, because of the amount of money competing with users interests in the space due to things like advertising, the desire to invade users privacy, etc. Still, if somehow we could magic up this monopolizing organization without conflicts of interest that would be far superior to the current state of affairs...
The Rust community rationalization that they don't need / want alternatives for "reason" is self-serving and all about control. I don't care if someone is the BDFL, they aren't right 100% of the time and not always doing things for altruistic reasons. The Rust community has imbued the leadership with some godlike omniscience and altruism because it makes them feel good, not because it's sound policy.
Of course rust leadership is going to make and has made mistakes, I'm confident every BDFL ever has also made and will continue to make mistakes, the same goes for the 'herd' of clang/gcc/msvc/... and the herd of browser makers. The target is not perfection, but merely being better than the alternative.
I think that, in the absence of conflicts of interests, single source of truth models (e.g. what rust rust/python/java/... does) are likely to do better than 'herd' of competing implementation models (C/C++/javascript) at making a good language. The latter probably does better at working despite conflicts of interest, but that's not a problem with most programming languages where there is relatively (compared to the browser ecosystem) little opportunity for a powerful corporation to push their interest to the detriment of others.
I think the rust community is quite clear that the rust leadership is flawed, but that's not very interesting without a way to make leadership better. If you can convince people you have that way - you'll get a lot of interest.
And some of them are pretty nice. Life is good.
Companies love permissive licenses for a reason.
If someone will pay you to do it I would do a new frontend — it's surprisingly little work to get something you can bootstrap with.
GCC, Clang, MSVC, and other compilers complement each other, serve different purposes, and serve different markets. They also ensure that the language is robust and conforms to a specification, not whatever quirks a single implementation happens to provide. And multiple implementations avoids the dangers of relying on a single implementation, which could have future problems with security, governance, etc.
The GNU Toolchain Project, the LLVM Project, the Rust project all have experienced issues and it's good to not rely on a single point of failure. Redundancy and anti-fragility is your friend.
* A C++ codebase. At the time GCC was written in C which slowed development (it’s now a C++ codebase adopting the lessons LLVM provided)
* It had a friendlier license than GCC which switched to GPLv3 and thus Google & Apple moved their compiler teams to work on LLVM over time.
* Libc++ is a combination of friendlier license + avoiding the hot garbage that was (maybe still is?) libstdc++ (e.g. there were incompatible design decisions in libstdc++ that inhibited implementing the C++ spec like SSO). There were also build time improvements if I recall correctly.
* LLVM provided a fresher architecture which made it more convenient as a research platform (indeed most compiler academics target LLVM rather than GCC for new research ideas).
Basically, the reason LLVM was invested in instead of DragonEgg was a mixture of license & the GCC community being quite difficult to work with causing huge amounts of investments by industry and academia into LLVM. Once those projects took off, even after GCC fixed their community issues they still had the license problem and the LLVM community was strongly independent.
Compilers don’t typically generate security issues so that’s not a problem. There are questions of governance but due to the permissive license that Rust uses governance problems can be rectified by forking without building a new compiler from the ground up (e.g. what happened with NodeJS until the governance issues were resolved and the fork reabsorbed).
It’s funny you mention the different C++ compilers consider that Clang is well on its way of becoming the dominant compiler. It’s already targeting to be a full drop-in replacement for MSVC and it’s fully capable of replacing GCC on Linux unless you’re on a rarer platform where GCC has a bit richer history in the embedded space). I think over the long term GCC is likely to die and it’s entirely possible that MSVC will abandon their in-house and instead use clang (same as they did abandoning IE & adopting Blink). It will be interesting to see if ICC starts porting their optimizations to LLVM and making them freely available - I can’t imagine ICC licenses really bring in enough money to justify things.
edit: it is both!
Yeah as someone who works in the embedded space, I do wish that LLVM had more microcontroller support. Maybe one day! It was weirdly hard to get up and running with the STM32L4 and LLVM, whereas GCC was easy as, just for a recent example. Apparently I can pay ARM for a specific LLVM implementation?
LLVM isn't the first compiler written in C++, and it isn't the first compiler framework building tool either
The most famous one, certainly.
I'd be interested to know what metric you're basing this on.
Outside the Apple ecosystem, LLVM seems to have made very little inroads that I can see. No major Linux distribution that I'm aware of uses LLVM, and Windows is still dominated by MSVC.
That's disappointing. GNU's extensions to the C language are extremely valuable features. I'd like to see what sort of contributions they would make.
Is Rust holy in some sense that’s it’s not allowed to rewrite the frontend?
It’s still a PITA to bootstrap Rust on a new architecture and there is also no working Rust compiler for architectures not supported by LLVM.
I know that there is codegen_rust_gcc but that one has the same bootstrap problems as the original Rust compiler and also requires architecture support to be landed in various Rust part which Rust maintainers have beem hesitant about (been there, done that).
So, I am absolutely glad that there will be an out of the box usable Rust compiler in the near future which will finally allow me to build rustified libraries again on architectures such as alpha without much pain.
The architecture support angle is a better one for why a GCC backend is helpful, but that will also be addressed by rustc_codegen_gcc so it isn't a relevant argument here. And by the looks of it, that will be a usable solution much sooner than gccrs, and it will remain usable with a much lower long-term resource commitment. This is an argument more in favor of rustc_codegen_gcc than gccrs.
The strongest argument I can see about bootstrapping issues here is that Rust moves much faster and Rust's own compiler code requires a relatively recent Rust version, so in the extremely specific circumstance where binaries are not available and cross compilation is also not possible, bootstrapping will take more steps than it would for a gccrs frontend limited by C++'s slower rate of evolution. I don't see this being anywhere near as much of a problem as gccrs trying to catch up to & keep pace with Rust's frontend evolution. The former we can solve with the same build farms and cross compiles we need anyway, the latter requires continuous ongoing human effort with all of the disadvantages of C++.
https://blog.m-ou.se/rust-standard/
https://rust-lang.github.io/rfcs/3355-rust-spec.html
https://github.com/rust-lang/rfcs/pull/3355
There are many organizations and industries that will not adopt Rust until it has a standard.
C, C++, C#, and even JavaScript (ECMAScript) have language standards. Why shouldn't Rust have one too?
C: https://www.iso.org/standard/74528.html
C++: https://isocpp.org/std/the-standard
C#: https://learn.microsoft.com/en-us/dotnet/csharp/language-ref...
JavaScript / ECMAScript: https://ecma-international.org/publications-and-standards/st...
... suddenly finding themselves in the "Actually, language standards are bad" camp, all because of a tribal opposition to the FSF?
Write the standard. Then argue that gccrust doesn't do it right. Don't refuse to document the language just to hamstring a competitor.
Also, please start with the borrow checker semantics. I don't think I've ever met anyone who could explain exactly what the rules are regarding what it can/can't prove.
https://blog.rust-lang.org/inside-rust/2023/11/15/spec-visio...
The argument is not that language standards are bad, it's that a C++-like ISO standard is unnecessary (when the quality documentation can exist in another form) and C++-like implementation fragmentation is bad.
(Have you read the NLL RFC? The Polonius work?)
Counterpoint: rust is doing fine without those organizations and industries. Why change what is working well?
So if you want those orgs/industries to get memory safety via rust, you either need a standard, or to convince them to not require that.
Because Rust is a game changer.
Wouldn't it be better to have Rust used for the code that runs on automobiles and aircraft? Or you would prefer that they keep using (subsets of) C and C++?
Is the security of your Internet-of-Things (IoT) devices good enough or could they be better?
What do you have against Rust being used in more places and for more purposes?
This happened without a standard.
They created a specification for Ferrocene:
While it is technically not a Rust language standard, it serves a similar purpose for Ferrocene.
It happens so often it became a meme at this point.
No, the meme is kept alive by C++ people who say that this happens without it actually happen. It happened a long time ago a few times, since then it's either an active discussion about languages, where for some reason talking about Rust is a problem, but every other language is okay or it's someone feeling attacked by the mere idea that projects which only could be done in C++ could now also be done by another language and starts crying about how the Rust community would flood every topic.
There's no longer an area where C++ is the only available option. Get over it. The rest of the world did a long time ago.
Progress has been disappointingly slow, but the project is alive, and has potential to speed up next year.
https://blog.rust-lang.org/inside-rust/2023/11/15/spec-visio...
> Progress has been disappointingly slow,
I don't think there's ever been a more concise summary of Rust.
[1] https://blog.rust-lang.org/2023/12/18/survey-launch.html
It can be moving too slow on important things, and too fast on unimportant or volatile things.
Hopefully Rust has people with vision and drive in places to let it escape its niche status.
Since she created the RFC for standardizing Rust (https://github.com/rust-lang/rfcs/pull/3355) and is also on the team that is working on Rust standardization (https://blog.rust-lang.org/inside-rust/2023/11/15/spec-visio...), I think she was making the point that Rust has good controls in place for adding features while compatibility, not that "Rust does not need a standard".
If she really believed that Rust does not need a standard, why would she create the RFC and join the team working on the effort?
Rust is a great language. There is no reason why it should not have a standard to better formalize its requirements and behaviors.
See the Ferrocene compiler which has been qualified for ISO standards. It’s essentially a standard Rust 1.68 compiler with a lot of added documentation. If you need a Rust compiler for safety critical environments, it’s reasonably priced and requires essentially zero changes to the Rust compiler that they didn’t just upstream. Without a standard.
Yes, it would be nice to have a standard for reducing ambiguity. But does the language need a standard? And if so, then for what purpose?
>> But does the language need a standard?
Yes, Rust needs a standard.
>> And if so, then for what purpose?
For the same purpose that all standards have--to formally define it in writing.
Ferrocene's web site (https://ferrous-systems.com/ferrocene/) shows that it meets the ISO 26262 standard (https://en.wikipedia.org/wiki/ISO_26262).
Why does ISO 26262 matter? What purpose does it serve? Couldn't a vehicle manufacturer just say "our vehicles are safe"?
Which would you trust more: a vehicle that is verified to meet ISO 26262 standards, or a vehicle whose manufacturer tells you "it's safe" without formally defining what "safe" means?
I stated it above, but I will re-state it here: Without a language standard, there are many organizations and industries that will not use Rust. Not because Rust is not a fantastic tool for the job, but because laws, regulations, etc. require standardization and qualification of components.
This means that I can use a qualified C compiler and toolchain to write safety-critical code, but I can't use Rust despite the fact that Rust is a better choice and will help to prevent problems. Standards do matter. Rust needs a language standard.
This is tautological. It's equivalent to saying "it needs a standard to be written because it needs a written standard."
I mean what use case is there for Rust language users that isn't already met by the Ferrocene project? And the Ferrocene project is not a standard as in "other implementations will be found lacking," but a description of the 1.68 compiler as-is. That is a specification, not a standard. Ferrous Systems did not need Rust to have a standard in order to qualify the compiler for ISO 26262 and IEC 61508.
Yes. And if the law in your country requires it to be standardized for specific use cases, then a language standard is needed.
>> what use case is there for Rust language users that isn't already met by the Ferrocene project?
Can you legally use Rust for the control software in aircraft? (https://en.wikipedia.org/wiki/DO-178C)
What about the safety systems for railroads? (https://ldra.com/ldra-blog/software-safety-and-security-stan...)
What about the control systems for nuclear reactors? (https://www.nrc.gov/docs/ML1300/ML13007A173.pdf)
Since you edited your post…simply having a standard won’t immediately qualify the language for those industries. There is only a tenuous link between having a standard and qualifying the language for industrial use.
Not having a language standard disqualifies Rust.
Administrators say that it shows that Rust is "not serious" and "not ready" for critical work.
Then explain how Ferrous Systems qualified a stock Rust compiler.
Ferrocene is qualified for some safety critical work and plans to have more qualifications soon.
Ferrous Systems wrote a blog post about the process: https://ferrous-systems.com/blog/qualifying-rust-without-for...
> Ferrocene is upstream rustc but with some extra targets, long term support, and qualifications so you can use them in safety critical contexts. This is what was stopping things like automotive companies from moving to Rust for things like engine control units, etc.
> It basically costs some money for the support and the qualification documents, but they will be all you need to prove qualification to any pertinent regulatory body so that your software can be certified for use in a real vehicle or whatever.
> ...Ferrocene is just unmodified rustc
https://old.reddit.com/r/rust/comments/17qi9v0/its_official_...
Basically the value add was to expand the support and documentation, which was required for qualification.
Again...no "standard" needed.
I think you are conflating standards and specifications. Ferrous fleshed out the specification, the description of the 1.68 compiler as-is. That means Rust 1.68 as-is was good enough for ISO qualification. Without a standard.
A standard is a minimum bar for languages to meet in order to be considered compliant. That's not a problem right now because there is, for all intents and purposes, a single canonical compiler and that is not likely to change.
So yes, a given vanilla release is compliant, you just can't do anything with that in a way that an audit will allow until you include all the justifications as to how that makes the language safe (defined by whatever safety standard you need).
Rust absolutely does not need a standard. Having a standard is a completely outdated way to design software in the internet era. Having a specification is a great idea and is what most people actually mean.
Without that, you won't have real-life alternatives.
That seems like a purely semantic argument? Rust is adding features extremely rapidly! You're just saying it's "development" if it's done by one entity but "creep" if it's done by someone else?
That used to be the common wisdom (especially because of C/C++), but it's a lot more debated these days.
The consensus in the Rust community is that the current situation (one canonical-by-definition compiler, lots of documentation, a minimal spec for safety-critical industries, and specs for some modular sub-parts) get most of the advantages of multiple implementations without the drawbacks.
And no, the official compiler isn't canonical by definition. If it were, it would mean it has no bugs, and if there's a crash or a wrong result that's what the language is supposed to do.
For now, anyone who finds that “crate foo works fine on rustc but doesn’t on gccrs” can just report a bug to gccrs.
By who specifically? I only ever see arguments against standardization and multiple implementations from the Rust community.
Not just ideologically, I mean literally - I don't understand how one sets up a development environment for GCC itself. I've had the misfortune of bootstrapping it a handful of times and it's the single worst behaved piece of software I've ever seen.
https://github.com/NixOS/nixpkgs/blob/master/pkgs/developmen...
(there are a ton of infrequently-used options in there; if you ignore them what's left is quite simple. replace all the options with `false` and do the dead-code elimination.)
Hint: don't use glibc. The real problem comes from glibc and the way it depends on gcc internals, which depend on libc (i.e. circularly). Bootstrapping GCC on Musl is quite easy.
glibc is the real dumpster fire.
This is delightful.
For example, I recently discovered that with RISC-V, GCC supports the RV32E target but LLVM doesn't.
I was pretty sure that llvm has supported RV32E for years now. https://reviews.llvm.org/D70401?id=395048
In any case, there are a lot of other compiler flags that are exclusive to gcc.
https://github.com/rust-lang/compiler-team/issues/648
This is the biggest problem with the LLVM mentality: they use architecture support as a means to extract support (i.e. salaried dev positions) from hardware companies.
GNU may have annoyingly-higher standards for merging changes, but once it's in there and supported they will keep it for the long haul. It's like buying vs renting. It takes a lot more dev hours to get support into GCC, but once it's there, it stays there.
I have a hard time seeing this as a bad thing. Hardware companies seem like the most logical people to pay for maintaining support for the architectures they sell.
Their customers pay full cost up front. Vendors can pay full dev cost up front too. GCC's model encourages this.
Be aware that GCC doesn't have a similar level of specificity about the state of each platform they support. There are packages in Debian that "compile" but when trying to execute them on "exotic" platforms you encounter bugs immediately. Supporting a platform in a static codebase of a compiler is easy, but in codebases actively being worked on, like GCC and LLVM, keeping things working is not trivial.
In my experience, the part in italics is a significant mistake.
Rust does not have a specification; there is a reference but it is explicitly not normative. A language udocumented except for a single reference implementation (as is tpday's fashion) have a long term weakness. What is the motivation for slavishly trying to maintain compatibility with the bugs and accidental quirks of another implementation? To guarantee that existing code will work in both implementations, which sounds sensible. And it is a sensible goal, but it does so at enormous cost by enforcing it in the wrong place.
The problem is that sometimes decisions are wrong, and sometimes bugs are written. But when you promise that all implementations will be bug compatible as part of compatibility you are also signing up to fossilize these bugs whether you want to or not.
A good example of someone who embraced this (to their credit!) is Microsoft: they spend a lot of person-power making sure that old programs continue to run while trying to fix security and reliability bugs. Rust need not and should not sign up for this burden so early in its lifespan. They should learn from history.
If they want the language to evolve they should embrace QA and QC. Famously "you cannot test quality into a product". You need QA: architecture, design, design and code reviews, etc to ensure that things will work properly and when not, that "failure heads in the appropriate direction". Then later in the development cycle QC (test cases) tries to see if you missed. This doesn't just apply to product development -- it applies to language development especially.
The strong standards (e.g. ComonLisp, C++, FORTRAN) embraced this belief. The weak, de facto ones (Most notably Python, but plenty of others) can still become popular, but change is difficult. Look at how long the Python 2->3 transition took, and how few python implementations there are.
I have a system now where much to my shock I am getting rid of gcc because it is not needed but keeping upstream golang and rust for updating existing software, which is by the way a nightmare when I updated golang few months back for a CVE and had like four places where the go based apps stored their own go environment.
Linux can already be compiled with clang if you want an all LLVM based toolchain. The duplicate effort of developing and maintaining this does not sound worth it to have GNU "purity."
A GNU compiler could add convenient feutures that the hardcore Rust users don't want other Rust users to be able to have.
Like ideological purity.