ronjouch 5 days ago

Particularly thunderous applause for providing both backwards compatibility (published packages will keep working, won't require anything from maintainers, and we can mix old and new) and autofix features (for when maintainers are willing to make the breaking changes).

Glad Rust learned from breaking changes hiccups of other programming languages :)

EDIT see discussion with kibwen below: fixed forward compat to backwards.

  • kibwen 5 days ago

    > published packages will keep working and won't require anything from maintainers

    That's backwards compatibility: new versions of the compiler will continue to be compatible with code that was written previously. Forwards compatibility would mean that new code would be compatible with old versions of the compiler (this would be analogous to a website from 2018 working in IE5, or being able to play The Last Of Us on PS1). Rust's backwards-compatibility promise ensures the former, but not the latter.

    That said, if you stick edition=2015 in your crates, that might go a liiiiitle bit towards keeping your code compiling on old versions of rustc, allowing you to partially "opt-in" to forward compatibility for your users, if you so chose. Though this would only keep you from accidentally using things that are edition-specific; remember that when you use a version of rustc from here on out, if you compile with the 2015 edition, you're not just getting a version of the compiler frozen as of last month; code compiling with the 2015 edition will still be receiving most of the same features as the 2018 edition (a strict subset, obviously; they're still the same compiler under the hood, only the frontmost of the frontend diverges).

    • steveklabnik 5 days ago

      adding the edition 2015 key will harm your ability to compile on older rusts, as they don’t understand the edition key. Just leaving it off as the default is better.

      • kibwen 5 days ago

        Ah, very good point! I was imagining the case where some does `cargo new`, which will give them `edition=2018` automatically, and didn't consider that you could just delete that line rather than editing it. :)

    • ronjouch 5 days ago

      Right! Thanks, wasn't sure about that, that's why I detailed what I meant between parenthesis. Updating my post.

      • kibwen 5 days ago

        No problem, they're really rather unintuitive terms. :) I thought it was a bit funny that I made almost the exact same comment here a month ago: https://news.ycombinator.com/item?id=18392855

        • nindalf 4 days ago

          I always get them mixed up. From now on I’m going to remember it with the example you gave. Forward compatibility = The Last of Us on PS1

pornel 5 days ago

It's a long text, but key takeaways are:

• There are no disruptive changes. You'll just opt in to a couple of new keywords and some warnings turned into errors (most of which will be fixed for you automatically). The "old" code continues to work unchanged and can be mixed with the "new".

• In the last year, in general, Rust has fixed a bunch of papercuts in the syntax, modules/namespaces, and the borrow checker that were annoying and unintuitive. New tooling has been added. It's now easier for new users, so if you've hit a wall with Rust 2015, give Rust 2018 a try.

• Rust is finding its strengths in embedded, WASM, CLI apps, and these areas are getting more polished.

• High-performance networking is another area of focus, and async/await is coming soon.

dilap 5 days ago

> Using rustfmt ensures that all of your Rust code conforms to the same style, like clang format does for C++ and Prettier does for JavaScript.

Oh, and also this obscure little program gofmt ;-)

  • steveklabnik 5 days ago

    gofmt was certainly a huge inspiration here, and many of us are jealous that go had it so early in their evolution. Not sure why it wasn't included!

    • vanderZwan 4 days ago

      Seriously though, the Go core team has been saying from the start that even they were surprised how big of a positive impact gofmt had. It reduces bikeshedding and makes it easier for people to read each other's code because you learn to read the same basic structure fluently and quickly.

      It's a basic IxD lesson about software tooling that I think more languages should pick up on.

      Personally, I think I miss gofmt-on-save more than any other feature when I write other languages (gofmt is a lot more robust than autoformatters in most other languages). Being able to type carelessly and quickly and know the output will be corrected as soon as I hit CTRL+S lets me focus on putting my thoughts on the screen. And it even works as a soft-compiler: if the autoformatter breaks it means I have a bug somewhere.

      • dan_quixote 4 days ago

        > It reduces bikeshedding

        My team works in equal parts go, ruby and javascript. Code review for go code is SO much easier. And it's mostly thanks to gofmt. It still amazes me how fervently some will argue about a single line of white-space or an optional semi-colon...

    • renox 4 days ago

      Golang, perhaps it is good I don't know but clang format sucks: it improves slightly normal code but the price is too high as sometimes it totally butcher code making it unreadable. Macro+lambda expression makes code formatters generates ugly things (I don't know what clion use to parse code but it also goes nut thinking wrongly that some code is unreachable..)

  • saghm 5 days ago

    To be fair, one of the most useful features of rustfmt/clang-format in my opinion is reformatting things that go beyond a maximum line length, which also seems like one of the harder parts to implement in a way that's useful. At leas the last time I used it, gofmt punted on this issue.

    • dilap 5 days ago

      Personally, I think gofmt made the right call -- let the editor soft-wrap long lines. This is much nicer because you can then dynamically pick whatever window width you want. Hard wrapping defeats this.

      (But of course there are downsides, because if you're in a context where the editor/viewer isn't soft wrapping, it's worse. Still, on balance, I think not manually wrapping lines is the way to go.)

      • kibwen 5 days ago

        A lot of it depends on the properties of the language itself. Traditional C-style imperative code involves many short statements each doing one small thing, one expression per statement, one statement per line; this makes it simple to punt on wrapping long lines (which is 90% of the difficulty of writing an automatic code formatter). Go inherits this "tall-and-narrow" idiomatic style. But other languages, especially ones that try to encourage fluent APIs, will often have several expressions within each logical statement, leading to fewer lines of code overall but more action on each line. In these languages it tends to be idiomatic to break the statement across multiple lines (usually at method calls), in which case an automatic code formatter's job is to do this operation manually.

        • dilap 5 days ago

          Seems like you're talking about splitting statements over multiple lines because it aids comprehension and makes sense semantically. Code-formatters are concerned with splitting lines that exceed some arbitrary line limit, no? The former it seems like it makes sense for me to do as the human authoring the code for readability (it seems a very similar decision to, "do I assign this value to a variable or include it inline"), the latter makes sense to do dynamically, given the line-length-limit of the moment.

          That said, no experience with these languages, so maybe I'm way off base. I'd be curious to see code samples, if you really feel like digging deep. :-)

    • pcwalton 5 days ago

      Implementing sensible line wrapping behavior is the hardest part of a code formatting tool. It's also very important for readability, and if you don't design for it early on it can be hard to add later.

  • pjmlp 4 days ago

    Which got its inspiration from the obscure little UNIX utility called indent, being used during the 90's on CVS pre-commit hooks.

    • vanderZwan 4 days ago

      Eh, IIRC Go was somewhat designed with "easy to autoformat"-semantics in mind. Or at least it just happened to overlap really well with the goal of being fast to compile.

      • pjmlp 4 days ago

        Languages like Delphi were easy to autoformat on save during the 90's, while compiling relatively fast and being much more feature rich than Go.

        • vanderZwan 4 days ago

          Sure, but you were talking about the Unix tool indent, which as far as I can tell based on its Wiki page was written to format C.

          https://en.wikipedia.org/wiki/Indent_%28Unix%29

          • pjmlp 4 days ago

            I was, because I was making the point that there isn't anything special about gofmt other that not being aware of previous tools.

            • dilap 4 days ago

              Go's innovation was social: it was saying: "Here's the code formatter, use it, always. There is one style, use it, always."

              • vanderZwan 4 days ago

                Exactly. Saying gofmt didn't change anything is a bit like saying the invention of the camera was nothing new because the technologies to create one existed for a century before they were combined.

  • littlestymaar 4 days ago

    Well, Go is still an obscure programming language compared to C++ or JavaScript :)

reissbaker 5 days ago

Really excited that non-lexical lifetimes landed! This was the most confusing part of the borrow checker: NLLs were the main class of "this should work but it doesn't" that I encountered in Rust. This should make Rust feel easier to pick up.

danieldk 5 days ago

I just converted one smaller crate and it was only about 5 minutes of work with cargo fix --edition. Especially because it does not matter that all the dependencies are still in Rust 2015.

Besides NLL, I really like the changes in the module system. Having to use extern crate was a drag, so it's nice that that isn't necessary anymore. Importing macros with use is both nicer and more intuitive than macro_use.

Very well done Rust team!

  • Cursuviam 5 days ago

    Same here. The only real work I had to do was removing some unnecessary imports because of the new improvements to mod namespaces.

Waterluvian 5 days ago

This might seem silly but I'm struggling to get into rust because the vscode plugin for it is sometimes too buggy (but the progress on it is awesome and devs deserve a round of applause). I use vscode for 4 or 5 other languages every week and don't want to maintain multiple text editors/IDEs.

What do people usually do in these cases? Surely I'm not alone.

  • bvinc 5 days ago

    I use vscode with the rust (rls) extension. It's not perfect. Sometimes when I edit cargo files it gets stuck and isn't recompiling my code. When that happens, I just ctrl-shift-p and select "reload window" and it's fixed in a couple seconds. It's not ideal, but it gets the job done and it's better than manually compiling my code.

    • Waterluvian 5 days ago

      I've found that autocomplete is pretty janky. It often doesn't work or is incomplete. Have you been having better luck?

  • YuukiRey 4 days ago

    I write Rust in `neovim` and I don't use any kind of linter that's integrated into the editor. I also have Visual Studio Code installed and setup and here's the thing: Rust (and also Haskell to a lesser degree) can have pretty long error messages but they are also informative. Some of them even have this sort of ASCII art showing you exactly where something was moved and then borrowed and so on. If you only use VSC you might not even be aware of those errors since the formatting of error window popups is pretty poor.

    So I simply open a new tmux pane and run a file watcher (`entr`) which runs `cargo check` (and in other windows also clippy and test) and it works perfectly.

    Since VSC has an excellent integrated terminal which you can split and have multiple tabs you could just do that in VSC itself. Frankly, I don't know any editor that displays live linting (and compiler feedback) in a way that I find useful but of course your YMMV.

    I've only used Intellij IDEA, VSC and neovim though and I write JS (professionally) and now Rust and Haskell for fun.

  • ahmedalsudani 5 days ago

    Why get hung up on one tool? Just write Rust code as if it were plain text. It's a statically typed language, so the compiler will check your code before running it, and your can run rust fmt to fix formatting.

    Maybe one day you can shun the darkness and embrace emacs ;) all your problems disappear on that day

    • guscost 5 days ago

      The fancy tools can check your code as you write it and highlight problems inline so you don't have to switch context in order to fix them. This saves time in any language, but static analysis with Rust is unusually powerful so tightening that feedback loop pays serious dividends (when it works).

      I'm not saying emacs and vim aren't really cool technologies. But why would a person who cares about optimizing keystrokes and crafting macros dismiss a tool that optimizes debugging and refactoring? I spend a heck of a lot more time staring at code than I do typing it, and if the reverse is true for you then you're some kind of savant and/or (more likely) you're writing too much code and reading too little of it.

    • Waterluvian 5 days ago

      A massively important tool I use for learning a language and libraries is autocomplete.

      Emacs takes a day to master? Awesome!

    • IshKebab 4 days ago

      Robust autocomplete is probably the most important tool for a language there is though. It's a shame that Rust doesn't have it and I'm slightly concerned that in never will - type inference means in lots of places there simply isn't enough information to do autocomplete properly.

    • always_good 5 days ago

      Well, that's always an option, so it isn't really advice. It's already what you have to do when the tooling support is bad. It's the poorest when an editor can instead inline its output of static analysis.

      Integration has a lot of benefits like tell you the inference of intermediate types. "Don't care about good integration" isn't really advice.

      It's like people who brag about syntax highlighting. The 99.9% rest of us consider it a good tool that improves our workflow.

      • ahmedalsudani 5 days ago

        No dispute integration is nice, but it is not working properly from what my parent comment was saying.

        Your options are to wait for it to improve, fix it yourself, or change your tools/workflow.

        Some people seem to have had an allergic reaction to my comment. Maybe it's the emacs mention, which was tongue in cheek. Oh well ¯\_(ツ)_/¯ I'll leave it as it is... for this is a righteous battle!

        • always_good 5 days ago

          I don't see a righteous battle.

          You mentioned Emacs as a solution which has a Rust plugin that has problems like most other Rust tooling. Yes, Rust's tooling landscape is immature and still a work in progress.

          Obviously you can just forgo editor integration all together. But you can do that in any editor.

  • Matthias247 5 days ago

    You can use the Intellij plugin, which fares a little bit better in some cases, but also struggles with some scenarios.

    All in all IDE support is a work-in-progress for Rust. It's definitely a lot better than 2 years ago, but still has a long way to go.

    It's currently maybe the only thing which isn't that great regarding tooling in Rust. Other features, like cargo and crates.io, feel however miles ahead of C++.

    • geezerjay 5 days ago

      > Other features, like cargo and crates.io, feel however miles ahead of C++.

      C++ with conan is pretty nice, so it really depends on how you approach a C++ project.

    • Waterluvian 5 days ago

      Oh interesting. Am I right to read into your comment that this is more of a Rust problem rather than a Rust plugin for vscode problem?

      • Matthias247 5 days ago

        It's more of a "not yet implemented" thing than a non-solvable problem.

        However a thing that needs to be considered is that building good IDE support for Rust is a very hard task, since it's a pretty complicated language (e.g. with all the type interference, traits, automatic dereferencing and conversions going on).

        Another fact is that the rustc compiler hasn't been built for supporting these kinds of use-cases right from the start, as e.g. the typescript and kotlin compilers have in my understanding. That means the IDE tooling can't directly query the compiler, but must duplicate some of the parsing and interference logic.

        • branko_d 4 days ago

          Another good example of "compiler as a service" is the Roslyn compiler for C# (and also written in C#).

          What is interesting is that Roslyn was a complete rewrite compared to the original C# compiler (which was written in C++ and couldn't power an IDE, so the tools like Resharper had to implement their own compiler front-end, essentially). IIRC, the ability to evolve the language was also a factor.

          It seems that as the language matures, there comes a point where tooling (and further evolvability!) becomes important enough to justify the huge undertaking of rewriting the entire compiler. I wonder when will Rust reach that phase...

          • steveklabnik 4 days ago

            We’ve already been in the process of it.

      • timClicks 5 days ago

        No, it's a tooling problem. Rust is a static language with a good type system.

    • pjmlp 5 days ago

      Regarding C++, there is vcpkg for Windows only projects, and conan is winning the race for multiplatform projects.

      • Matthias247 5 days ago

        I know that there are several projects around, and I'm for example halfway comfortable with using cmake as a main build tool (bazel looks interesting too).

        A huge issue however is still fragmentation. Only a minimal amount of libraries are provided and can be built in this fashion, and translating and maintaining build files of others is no fun.

        Another annoyance can be getting into existing projects, where maintainers don't want to learn or invest in anything new, and claim that Makefiles and copying dependencies into subdirectories are just fine.

        There's a huge difference between a state of the art modern C++ project and the typical things one encounters.

        • pjmlp 4 days ago

          I fully agree, however I would say many existing projects are going go benefit from incrementally change into more modern C++ practices, than rebooting into Rust.

          It is great that the team has achieved the Rust 2018 milestone, and there are already a few key companies adopting it, but it needs to catch up with 30 years of tooling, e.g. Qt, C++ Builder, CUDA, ...

      • sanxiyn 5 days ago

        I am using vcpkg for a cross platform project, and it works great. vcpkg works surprisingly well on Linux and macOS.

        • pjmlp 4 days ago

          Ah thanks, I just stated it like that due to many not wanting to deal with MS tooling outside Windows.

  • staticassertion 5 days ago

    I highly recommend the intellij plugin. If you use clion you also get debugger support, I believe.

    Autocomplete works well. Macro support is improving all the time, and is pretty solid. Performance needs more work, but is generally acceptable.

  • sidcool 4 days ago

    Also try TabNine autocompletion for Rust. It's quite good.

mmastrac 5 days ago

I played around with writing a procedural macro library last night. Being able to transform functions is amazing (although my example is pretty basic - just adding some linker metadata to a function).

https://github.com/mmastrac/rust-ctor

baby 5 days ago

I'm very interested with Rust on Embedded Devices. I've recently ported my cryptographic library (www.embeddedDisco.com) to C, thinking it was the only way to support these many tiny systems, but it looks like it might become less and less true. I'm still wondering, how are young developers using something like the Arduino IDE supposed to figure out how to use a Rust library.

  • thenewwazoo 5 days ago

    Arduino has the benefit of several years of refinement and an all-in-one, multi-target library. Embedded Rust is still in its infancy, though it's absolutely usable and (imo) practical for experienced users. I think as the embedded abstractions get polished and there is more convergence of HALs, something like an all-inclusive solution may come about. As it stands, it's very easy to combine a chip HAL and device drivers, but you have to know what building blocks to grab (if they're there - there's lots to be done!).

  • keithnz 5 days ago

    This is what most interests me about Rust, using it to replace the C code on our embedded devices. Nice to see that is a first class use case now and the language will be mainted to ensure it is supported.

    I might have to try and do a proof of concept with one of our devices now :)

    • thenewwazoo 5 days ago

      To add on to what jamesmunns has already said, we're an excited group, us embedded Rust users. I recently built one of our in-house products using Rust as a proof-of-concept, and the process convinced me that Rust is the future of embedded devices. Feel free to reach out to me if you have any questions. I'd love to keep working with embedded Rust!

  • vanderZwan 5 days ago

    I wonder if there any compiles-to-C languages will "port" compilers to create compiles-to-Rust versions. Because there are plenty of those languages out there in the embedded space.

    • gpderetta 5 days ago

      What would be the advantage?

      • AWildC182 5 days ago

        Probably none I'd guess. The embedded space really loves C simply because it's viewed (usually wrongly) as more deterministic than C++ but there are a lot of instances where C is not a good choice (when you no-shit need objects) so they come up with abstractions for common things (terrifying C patterns). Rust has a lot of features that C lacks so we probably won't need to do any abstraction for now.

        • vanderZwan 4 days ago

          Why do you say none, then immediately list a few advantages of Rust? To build on your points:

          - it might be a easier language to target with a compiler because it's a higher-level intermediate language

          - there would be fewer undefined behavior shenanigans

          - successfully writing a compiler that targets Rust without using unsafe (too much) would mean getting (some of) the safety benefits as a side-effect. Could especially be relevant for languages aiming to introduce zero-overhead safety to the embedded environment themselves

          - regarding previous point: discovering bugs in the process of porting to Rust

          - because Rust plays relatively nicely with C and C++, one wouldn't have to lose existing libraries along the way

          • AWildC182 4 days ago

            I'm saying "None" in response to needing compilers that compile to Rust, not to rust itself. Basically to boil down my argument, we don't need to compile to Rust instead of C because simply replacing C with Rust should be sufficient. People made "compiles to C" languages because C was not capable of doing everything they wanted easily.

            • vanderZwan 4 days ago

              Aaah, I see where you are coming from. Yes, that is a good point.

              Porting a compiler could still be useful for the sake of legacy code though.

          • gpderetta 3 days ago

            It seems to me that writing a compiler that can target safe rust would be extremely hard with little benefit.

        • pjmlp 4 days ago

          Not only that, they also love to deploy with optimizations turned off to avoid any compiler magic, so you end up with some convoluted hand written code, which could otherwise be written by the compiler itself.

      • vanderZwan 5 days ago

        Easier interop with Rust crates?

    • nicoburns 4 days ago

      Compile times might be pretty painful with that kind if setup...

2bitencryption 5 days ago

I'm a huge fan of this "edition" system.

It allows learning from past mistakes, without being bound to supporting those past mistakes forever...

The module system improvements are so nice. Same with all the other ergonomic changes. Maybe I would feel differently about Rust if I had to use it professionally, but as a hobbyist, it's such a joy.

Twisol 5 days ago

FYI, if you're on MacOS at least, `rustfmt` looks like it's still under the `rustfmt-preview` name in `rustup components list`. Running `rustup component add rustfmt` (as in the official announcement post) gives an error:

    > rustup component add rustfmt
    error: toolchain 'stable-x86_64-apple-darwin' does not contain component 'rustfmt' for target 'x86_64-apple-darwin'
  • steveklabnik 5 days ago

    Apparently you need to `rustup self update` before installing this version, and it will work. So if you've installed it, uninstall it, then update rustup, then reinstall it. Sorry about that! There was a miscommunication...

    • doodpants 5 days ago

      I'm seeing the same problem on Windows:

        $ rustup component add rustfmt
        error: toolchain 'stable-x86_64-pc-windows-msvc' does not contain component 'rustfmt' for target 'x86_64-pc-windows-msvc'
      
      I looked for how to uninstall rust, and found that "the book" says that the command 'rustup self uninstall' will uninstall both rust and rustup. So I figured, what the heck, I'll start from scratch. I ran this command, uninstalling everything. Then I re-downloaded rustup-init.exe, and then ran it, reinstalling rust. I even then did 'rustup self update' and 'rustup update' for good measure.

      I still get the same error when trying to install rustfmt. What else is there to do if completely uninstalling and reinstalling doesn't work?

      • steveklabnik 5 days ago

        Theres a bug! We’re working on it. I should have an update for you soon.

        • doodpants 5 days ago

          Ok, thanks!

          • steveklabnik 5 days ago

            This should be fixed now, please try again!

            • doodpants 4 days ago

              It works! (After uninstalling, and then re-downloading and running rustup-init.exe)

    • Twisol 5 days ago

      Ahh, you meant to reinstall this version of Rust, not rustfmt. I did that and it worked -- the components list contains `rustfmt` properly, not `rustfmt-preview`.

      Thanks!

    • Twisol 5 days ago

      (EDIT: I fixed this by reading Steve's comment more carefully. Uninstalling `stable` and reinstalling it again made `rustfmt` available.)

      I'm still seeing the same behavior, even after following these directions.

          > rustup self update
          info: checking for self-updates
          > rustup --version
          rustup 1.15.0 (f0a3d9a84 2018-11-27)
      
      Thanks for looking into this, though :)
SCdF 5 days ago

Incredibly excited for those lifetime improvements. Rust is cool and everything, but (at least for someone who comes from a memory managed runtime background) lifetimes are a hard pill to swallow. Anything that makes them go down easier is gladly received.

vvanders 5 days ago

Good stuff.

One small note on WASM, I really hope that wasm32-unknown-emscripten doesn't become an orphaned target. I've got quite a few WASM/ASM.js projects that have C dependencies which work great with WASM. Unfortunately wasm-bindgen and the like appear to require wasm32-unknown-unknown and since the cc crate doesn't work with wasm32-unknown-unknown I can't take advantage of them.

Totally love the wasm32-unknown-unknown target from a easy bootstrapping standpoint but it feels like that and the traditional target are starting to diverge pretty significantly(you can't do a cdynlib crate on wasm32-unknown-emscripten for instance).

  • steveklabnik 5 days ago

    We’re not removing it any time soon, but we’re also not focused on developing it any further. Targets based on LLVM’s native support, like the unknown target, are the future.

    • vvanders 5 days ago

      Thanks, appreciate the clarification.

      Is there any plans in the works to bring the cc crate over to wasm32-unknown-unknown? Most of the impressive stuff I've been able to do with WASM has been pairing existing C libraries with Rust in the right ways and it would be unfortunate to lose access to that whole ecosystem.

      • steveklabnik 5 days ago

        Not immediate plans, but you are right that it’s a weakness of the new targets. There’s still so much work to do!

jmull 5 days ago

I was surprised by this:

"Rust has quickly become the best language for integrating with existing web applications."

Is that really true or some PR hyperbole?

  • nicoburns 5 days ago

    In terms of web-assembly, Rust really is the "market leader". C/C++ come with a lot of baggage, and the distinct disadvantage of not having a package manager, and everything else has more work to do due to having to implement a GC. Additionally, Rust has made wasm a focus, and now has bindings to most of the web APIs. And even high-level libraries that generate DOM.

  • kibwen 5 days ago

    For doing anything with WASM, Rust is (as of this moment) far and away the best experience. The Rust devs have had their eye on WASM support since long before WASM stabilized, giving them a first-mover advantage, and they benefit from the fact that Rust's model is so C-like that it's relatively little work to make it work with what WASM expects.

  • Diggsey 5 days ago

    I think it means specifically for compile-to-WASM languages. In that sense it's probably true, certainly the tooling is better than any I've seen elsewhere.

  • vram22 5 days ago

    >"Rust has quickly become the best language for integrating with existing web applications."

    In what senses? Interested to know, being both a CLI and a web dev guy, and not having looked much into either Rust or wasm yet (read a little of the Rust book, only, recently).

    • steveklabnik 5 days ago

      We’re seeing two ways so far: the first is WebAssembly. Our tooling does not assume that you’re re-writing everything in wasm, but that you’re using it alongside JS, so you can use it only where it makes sense, and it’s easy to do so.

      Second, in services. Microservice architecture means you can write parts of your applications in different languages, and we have really nice JSON handling, or whatever other format you’re using. People have moved high load services to rust and seen big gains.

      Of course, there’s still lots of work to do, and there’s always compelling reasons to use other technologies too.

      • vram22 5 days ago

        Thanks for the detailed and balanced answer.

        >we have really nice JSON handling, or whatever other format you’re using.

        Good to know. Data formats and conversion between them (often, though not solely, in the context of CLI apps) is one of my interests and areas of experience, so will check out those aspects of Rust over time.

        >People have moved high load services to rust and seen big gains.

        Cool ...

krick 5 days ago

So, is roadmap for 2019 visible? I'm really interested if any high-performance computing features are going to be introduced or not.

  • staticassertion 5 days ago

    Some HPC features (const fn) made it into 2018, and will continue to be developed.

    Out of curiosity, what is missing for you? I'd imagine int generics and const fn, simd. What else?

    • krick 4 days ago

      I wonder if better support for multi-dimesional arrays and matrix operations could be implemented.

    • luthaf 4 days ago

      Better distributed computing libraries and semantics (MPI, OpenSHMEM, ...). At the language level, it is not completely clear how RMDA and lifetimes/references should interact.

      Better GPGPU support, for CUDA and OpenCL, ideally with direct compilation of Rust code to GPU assembly.

huntie 5 days ago

I was kind of disappointed with Rust 2018. While it technically maintains the guarantee of backwards compatibility I don't feel that it meets it in spirit.

The prime example is that code for working with modules that works fine on 2015 throws an error in 2018. This means that I now have to know two languages that are basically the same but have important differences. This is a real issue because I'm sure that some crates will jump to 2018 while others want to continue compiling on old compilers.

  • Twisol 5 days ago

    My understanding is that, since the edition is a per-crate property, crates that don't jump to 2018 aren't any worse off, and will continue to compile on both old and new compilers. I'd also be interested to know whether `cargo fix` covers the module changes for you.

    (EDIT: I now realize you're talking about crates that will continue development on Rust 2015 specifically because they want to target older compilers, which is a different concern from what I was addressing. But now I'm not sure what the concern actually is. Even if you stay on Rust 2015, what reason is there to not upgrade your compiler? Both editions remain supported in 1.31 and will be for...ever, as far as I know.)

    • huntie 5 days ago

      `cargo fix` probably covers it, but `cargo new` defaults to Rust 2018, which was kind of annoying when I was getting errors on code from another project that worked fine there.

      There isn't a reason not to upgrade the compiler. My concern is that if I stay with Rust 2015 nothing changes for me, but if I need to fix a bug in someone's library there is a good chance they'll be on Rust 2018 (or vice versa). My concern is with the mental aspect of needing to remember the differences, not the technical differences.

      It's possible that there isn't much to remember. Admittedly, I haven't yet looked at what the module changes are; but it still bothers me that such basic code no longer works.

      • Twisol 5 days ago

        That's a great point! Helping devs get up to speed on the changes is definitely a significant concern for editions, which is, I'm sure, why they put so much effort into creating a dedicated edition guide: https://doc.rust-lang.org/edition-guide/

        • huntie 5 days ago

          Sure, the rust team is really good about documentation. I think that I worded things poorly though. One of the reasons that I started using rust is that the rust team said that once they hit 1.0 they would not make breaking changes (unless there was a soundness bug). While Rust 2018 technically keeps this promise since Rust 2015 will continue to compile and interoperate, I do not feel that it keeps the spirit of this promise; which is frustrating to me.

          • tikue_ 5 days ago

            Backwards-compatibly changing existing features feels qualitatively the same to me as adding new features. In either case you can ignore them until you need to work on a codebase that uses them.

  • nicoburns 5 days ago

    There doesn't tend to be that much support for older compilers in the Rust community (upgrading is so easy...). So I'd expect pretty much every crate to upgrade over the next year or so.

    • huntie 5 days ago

      I agree that most crates will upgrade, but there are some very prominent crates which maintain compatibility with old rust versions. An example is the regex crate which works on rust 1.20+.

      • steveklabnik 5 days ago

        To be clear, rust 2015 code compiles just fine on new compilers, and will indefinitely. That’s a key part of it! Nothing should change with regards to these crates; they already weren’t using new features anyway, in order to maintain that compatibility.

        • huntie 5 days ago

          Sure, I understand that. But if I've only ever used Rust 2018 and I go to submit a PR to the regex crate, I now need to be aware of any differences between the two versions.

          It might not be a big deal, but a few weeks ago I started a new project that was going to reuse code from an existing project. The new project defaulted to 2018 (nightly compiler) and I got a bunch of errors with my module imports. It was just frustrating because it felt like it was a breaking change and one of the reasons I started using Rust is the promise of no breaking changes.

          • therockhead 5 days ago

            Is this not the situation when any new feature is added to a language?

          • steveklabnik 5 days ago

            That’s fair!

            These kinds of changes are generally on par with other “never break” languages like Java and C++. Java 1.4 feels very different than Java 10.

            It’s tough!

          • jononor 4 days ago

            If you want to use Rust 2015, just do so?

            I fail to see how Rust editions could be much smoother managed. Or should they never fix anything in the language?

  • Rusky 5 days ago

    For what it's worth you can write 2018-style modules code in 2015 crates (perhaps modulo a few edge cases?). So if you're willing to switch to that you can have one style across both editions.

    • steveklabnik 5 days ago

      You can use a 2018 crate from a 2015 crate (and vice versa) but you can’t “expose rust 2015” or something. There’s raw identifiers for anything that would be incompatible. That would require a new rustc, of course.

      • Rusky 2 days ago

        I'm... not sure you replied to the right comment here? What I'm talking about is writing 2018-style paths (starting with `crate::`, etc) in a 2015-edition crate, so you don't have to switch between two path styles when you switch between crates.

        • steveklabnik a day ago

          Ah. I just misunderstood what you meant. Sorry!

devj 4 days ago

To all Rustacenas who are running Rust in production, would you create an "Effective Rust"?

  • sidcool 4 days ago

    I would buy the book in a heartbeat.

amelius 5 days ago

Shouldn't .rs files declare the version of the language they were written for? At least optionally?

  • ekidd 5 days ago

    The Rust "edition" flag is set on a per-library ("crate") basis, not per source file. It basically just controls a few small differences at the parser level.

    I'm really happy about the ability to mix older Rust 2015 and newer Rust 2018 code in the same project, and the tools for automatically migrating code to the 2018 edition. It took us like 3 hours to migrate a large code base at work, and I think I needed to file a bug report about exactly one upstream library. Everything else was seamless. And we can still use all our Rust 2015 dependencies.

  • Tuna-Fish 5 days ago

    With [2018], rust has gone on the path of tightly integrating the build system into the language. Before, you had to declare the crates you were using with "extern crate" in your main package. Now, the only place those are found is in the Cargo.toml build script. The language version is chosen in the same place.

    • steveklabnik 5 days ago

      This imposes no additional requirements on rustc, it’s the same as it previously was. Cargo is no more tightly integrated.

      • gpm 5 days ago

        This makes using rustc without cargo much harder.

        For instance, previously I could `rustc src/main.rs -L deps` and assuming that the dependencies had all been built and had their rlibs put in deps things would just work. Now that is no longer possible, I have to manually specify every external dependency on the command line.

        • kibwen 5 days ago

          `extern crate` hasn't been made a syntax error, it's just now possible to elide it. There are plenty of use cases where I imagine people will still use it, not least of which is play.rust-lang.org, which has no direct access to Cargo or rustc, and `extern crate` is what one will continue to use to import third-party libraries there.

moother 5 days ago

Dumb question, does this mean the official Rust book I had shipped yesterday is outdated?

  • steveklabnik 5 days ago

    Not at all! It means it isn’t complete. That was always true; we release a new thing every six weeks.

adultSwim 3 days ago

I really like the changes to the type checker

pjmlp 5 days ago

Congratulations on the milestone! Updating right away. :)

nicoburns 5 days ago

EDIT: Updated to correct URL (as it changed after I posted).

Official Rust blog post: https://blog.rust-lang.org/2018/12/06/Rust-1.31-and-rust-201...

(may be inconsistently available for a bit, as they seem to be in the process of updating the site to a new look)

throwaway487550 5 days ago

Still, we want more things to be boldly and shamelessly borrowed from ML and Haskell

  * Universal pattern-matching (everywhere)
  * Obvious syntactic sugar for defining and applying curried 
    functions, partial application.
  * Type-classes (as a one single major innovation in PL 
    since Smalltalk)
  * receive (or select) with pattern matching (Erlang, protobuf).
  * first class typed channels (Go).
  * multimethods (CLOS).
Thank you!
  • paavohtl 4 days ago

    > Universal pattern-matching

    Where are you missing pattern matching? The only place I can think of where it can't be used currently is the function signature. Might be hard to retrofit to the current syntax.

    > Type-classes

    Rust's traits are type-classes.

    > first class typed channels

    Why do they need to be first class? Unlike Go, Rust's type system is powerful enough that type-safe channels can be (and have been) implemented as a library. See crossbeam-channel.

    • steveklabnik 4 days ago

      You can use pattern matches in function signatures.

      • throwaway487550 4 days ago

        How could I write this in Rust?

            fun map f [] = []
              | map f (x::xs) = (f x) :: map f xs;
        
        And boy, no TCO. What a shame.
        • steveklabnik 4 days ago

          You can only use irrefutable patterns. You’d write this by writing a match in the body. There’s been talk of maybe supporting this directy but it’s so niche there’s not a ton of demand.

          There is TCO, just not guaranteed. We’d like it; we have even reserved a keyword. Needs an RFC though.

    • throwaway487550 4 days ago

      > Rust's traits are type-classes.

      And the functor typeclass in Rust looks like...?

      • paavohtl 3 days ago

        Rust doesn't currently have higher-kinded types, which are a separate feature from typeclasses. Implementing a true Functor trait is impossible without HKT, though the lack of it and other FP abstractions generally speaking hasn't been an issue.

  • foota 5 days ago

    What do protobufs have to do with select?

    • throwaway487550 4 days ago

      Protobuf does implicitly, via code-generation, what could be done explicitly via defining algebraic types and pattern-marching on receive, a-la Erlang.

krobolt 5 days ago

Yay Rust empowering people of something to diversify and something no real reason safe Mozilla sjw crap only really used on Firefox [which is shit].

2.5 years of Rust and this is about all I can say for it. Personally I just spend my time learning C.

Animats 5 days ago

Why should they need "edition"? Just bump the first number of N.N.N semantic versioning.

  • lalaithion 5 days ago

    Because they haven't actually broken backwards compatibility, but rather allowed people to opt-in to incompatible changes by setting a flag.

    • deathanatos 5 days ago

      > haven't actually broken backwards compatibility

      My understanding from reading the article is that they have broken backwards compatibility; things using the new keywords that were introduced (e.g., "await") will no longer compile.

      I think part of the confusion here is that there are essentially two things: there's the compiler, and there is the language itself. It appears (from the article) the compiler can presently support compiling either "edition 2015" or "edition 2018" of the language. (So, it's not the compiler's major we're bumping, rather, it's the language's.)

      That flag which says which "edition" to use could just as easily specify which SemVer major version to use. So, instead of saying "use Rust edition 2015" or "edition 2018" it would just be "Rust (the language) v1" or "Rust (the language) v2". There's no difference aside from the naming.

      The year doesn't particularly add anything, and obscures if we're jumping over multiple "barriers" of breaking changes or not. I'm not actually sure that that really matters.

      My only thought is that having completely different looking things might be better in that it just visually distinguishes the two better. I've repeatedly seen devs muddle the difference between the language, and a implementation of the language, and what the language guarantees and what the implementation happens to do.

      • oconnor663 5 days ago

        > things using the new keywords that were introduced (e.g., "await") will no longer compile

        That's not backwards compatibility, it's forwards compatibility. Whenever any language adds a new feature, using that feature breaks the build on old compilers. Setting `edition=2018` is no different. But the important thing is that old code continues to build unmodified on new compilers. That's what the edition system is preserving.

        • deathanatos 3 days ago

          Forwards compatibility would be old code compiling in the newer "language", and I mean this from the grammar perspective: that the changes to the grammar are such that all things that were previously valid in the language remain as valid, and their meaning does not change.

          This is not the case.

          That the compiler has the user opt-in to the never version by setting a flag is irrelevant, for the discussion of versioning. (Now, for an end-user, I think it's great: you do not want to automatically break working code. I'm not saying the flag shouldn't exist, I'm saying the flag exists because the change is breaking.)

          • oconnor663 2 days ago

            You're right, I totally misread the above.

      • steveklabnik 5 days ago

        You have to add in a flag to get those new changes, so it is backwards compatible. All existing code compiles and will continue to compile with no changes into the future.

        • deathanatos 3 days ago

          The flag's mere existence is proof that the change is not backwards compatible. If it were, you wouldn't need the flag.

          > All existing code compiles and will continue to compile with no changes into the future.

          My point, again, is that it won't compile without changes in the new edition. "Edition" is the word that was chosen (and it's not necessarily a bad word), but what it is is a version for the language. Whenever a backwards incompatible change (or set of changes) is going to be made to the language, that's a new "edition", but you could just say "version" and it would be as accurate.

          Again, consider the language and compiler as separate entities, each with their own API, and I think it becomes clearer. Changes have been made to the language (e.g., the addition of new keywords) that render strings previously valid in the language now invalid, or valid with a changed meaning. In any system, this is a breaking change. This change could, were SemVer used, be identified with a major version bump.¹

          That the compiler (which itself has a version number, separate from the language now) is capable of recognizing a flag and switching internally which version of the grammar it uses is great, but again, proof that there exists a version of the grammar (called "edition").

          ¹But honestly, that ship has sailed for Rust. And that's fine, and I think it's just a different term, and that it has a good chance of helping better convey the meaning by simply being different. But the argument that, under that, it's essentially SemVer, is nonetheless true.

          (That at present the compiler and the language are very much intermixed makes this much more muddled. Were the language's definition more formal, I think this would all get much clearer, as we could ignore rustc, and talk about the language.)

          • steveklabnik 2 days ago

            All existing code still compiles, just as before, with no changes. That’s backwards compatibility.

    • dj-wonk 5 days ago

      I'd like to see more projects consider this approach for updates.

      Would anyone like to share other examples of this sort of thing (such as in other compilers or tools)?

      • steveklabnik 5 days ago

        We took a lot of inspiration from C++ and Java, both of which have some sort of flag that says “compile this against a particular version”.

        • mlindner 5 days ago

          Except C++03 isn't getting any more features added to it and neither is C++11 (which also use the year, not version number). 2015 edition will continue to get new features.

          • steveklabnik 5 days ago

            Yes, the non-2015 editions are direct analogues, but 2015 is special.

  • int_19h 5 days ago

    As I understand it, the N.N.N version number for Rust is compiler version. Editions are language versions.

  • andoriyu 4 days ago

    Imagine if python3 didn't break your shit and you were able to use python2 packages in your python3 code without thinking twice? Well, that's what they did.

v_lisivka 5 days ago

Incompatibility with C enums in repr(C) enums is still not fixed.

In C, enums are more like:

  #[repr(C)]
  enum Foo {
    BAR=0,
    BAZ=1,
    Unknown(i32), // other possible values of int
  };
but Rust doesn't allow to write enums like that, so it leads to bugs in FFI.
  • kibwen 5 days ago

    It's true that C allows manually setting an enum to an integral value other than those explicitly specified, but no project that I've ever seen regards such an action as anything other than a bug. Can you give an example of code that's using such a thing, especially for FFI?

    • olliej 5 days ago

      The Mac and iOS design uses explicitly valued enums in all APIs (aside from anything else it makes it much harder to accidentally break ABI).

      JSC also uses them extensively in the lexer and parser.

      • kibwen 5 days ago

        Rust can do explicit values for C-style enums; what it currently doesn't support is the ability to set such an enum to an arbitrary (non-declared) value as C does.

    • v_lisivka 5 days ago

      Protobuf specification explicitly states that protobuf compatible implementation must support values outside of the enum range, to be forward compatible with future updates to .proto files.

    • bonzini 5 days ago

      You could define flags as an enum with values 1, 2, 4, 8, ... and OR them together.

      • auscompgeek 5 days ago

        Such a scenario is more appropriately represented with the bitflags crate.

  • pornel 5 days ago

    It's not fair to call it "still not fixed". That behavior is entirely intentional and working as designed (even if it's not useful for all C enums), and no changes to it were planned for this edition.

    • v_lisivka 5 days ago

      Behavior of #[repr(C)] enum doesn't match behavior of the C enum, so it's bug for #[repr(C)] enum's.

  • gpm 5 days ago

        #[repr(C)]
        #[derive(PartialEq, Eq, Clone, Copy)]
        struct Foo(i32);
    
        const FOO_BAR: Foo = Foo(0);
        const FOO_BAZ: Foo = Foo(1);
    
    I remember writing a macro to handle this properly, even using modules for namespacing, ages ago...
    • huntie 5 days ago

      You can also use associated consts which feels nicer in my opinion:

          #[repr(C)]
          #[derive(PartialEq, Eq, Clone, Copy)]
          struct Foo(i32);
          
          impl Foo {
              const BAR: Foo = Foo(0);
              const BAZ: Foo = Foo(1);
          }
      • gpm 5 days ago

        Good point!

        • kibwen 5 days ago

          Though to be fair, if you wrote your macro version "ages ago", there's a good chance associated consts were still unstable back then. :P

    • v_lisivka 5 days ago

      It doesn't play nicely with tooling like debugging output(#[derive(Debug)]), introspection, serialization, match statements, etc.

      My solution to this problem, for i8/u8 enum's only, is just to list of unused values as enum Foo { ..., RESERVED_2=2, RESERVED_3=3,.. RESERVED_255=255,}.

      • gpm 5 days ago

        These days it would be perfectly possible to have a macro that looked like

            cenum! {
                Foo: i32,
                Bar = 1,
                Baz = 2
            }
        
        That expanded to the above `struct Foo(i32)` code and implemented Debug printing `Foo::Bar`, `Foo::Baz`, and `Foo::Unknown(3)` or something.
        • v_lisivka 5 days ago

          Of course, we can use i32 directly, but when project has hundreds of enum's, when enum's are constantly updated, when different crates are using different incompatible approaches to represent repr(C) enum's, it hurts. Rust compiler can pack Option<NonZeroU8> into single byte, so it can do that for repr(C) enum's with e.g. `Other(2..255:u8),` variant too (see my proposition above). Technically, they are not different, just have more specific cases.

  • Twisol 5 days ago

    Is there an issue posted somewhere on this?

    From a glance, I would imagine you could mediate across the FFI using an `i32`, and provide an `impl From<i32> for Foo` that neatly separates the discriminated and excess cases.

    • v_lisivka 5 days ago

      Yes, I can use numbers to represent anything, but I will lost compiler support. It's not a solution but workaround.

      I saw number of tickets about that in Rust ticketing system. This one seems to be most relevant:

      https://github.com/nikomatsakis/rust-memory-model/issues/41

      Then these ones:

      https://github.com/rust-lang/rust/issues/36927

      https://internals.rust-lang.org/t/c-style-enums-in-ffi-and-a...

      https://internals.rust-lang.org/t/pre-rfc-enum-from-integer/...

      • Twisol 5 days ago

        For the first one, which I see you've made a comment on:

        > This behaviour affects Prost: i32 type is used instead of enum type, because Protobuf defines that enum variable must be able to hold values outside of enum range to be compatible with future versions. As alternative solution, it's proposed to use _Unknonwn_(i32) variant, but this solution cannot be implemented in current version of Rust, because C-like enums cannot contain tags.

        In principle, a protobufs library (like Prost) could map a protobuf enum `X` into a Rust enum `ProtobufEnum<X>` with `Known(X)` and `Unknown(i32)` fields. Or autogenerate a Rust enum with the known variants inlined, plus the extra Unknown field.

        It is incumbent on the library author and/or FFI user to map semantic notions appropriately across the interface. Rust enums do not have C semantics, even if the underlying machine representation is annotated to mimic C. If something across the interface allows its flavor of enum to take on unnamed values, then that semantics needs to be preserved faithfully in Rust -- and that may mean using something slightly different from what Rust itself natively provides.

        • v_lisivka 5 days ago

          > Or autogenerate a Rust enum with the known variants inlined, plus the extra Unknown field.

          Unfortunately, support for this extra Unknown() field is not added to rustc yet even in Rust2018 edition. Compiler can do that, because it already used to save space for Option<>.

          > Rust enums do not have C semantics, even if the underlying machine representation is annotated to mimic C.

          Yes, it's the bug I'm talking about. Solutions are: a) remove repr(C), b) fix repr(C).

  • int_19h 5 days ago

    In vast majority of cases, C enums are used in the same way as Rust enums - i.e. only the named enum variants are supported by any code that touches enum values (and if you're lucky, it might check for other stuff and reject it explicitly, but more often people forget and things just don't work or silently misbehave). So in practice this behavior is correct.

    For those very rare cases where someone used open-ended enums in their API, ascribing some meaning to unnamed values that must be exposed, you can always use i32 etc.

strgrd 5 days ago

Crazy to think how far Rust has come since 2013. Very excited for the electricity update coming later today -- I've already seen a few videos of people implementing basic computer systems. I imagine there won't be many people falling for bases with open doors this wipe... Still frustrating to see basic optimizations in the pipeline. Many users have been suffering from microstutters for years. Regardless, there is nothing else out there like Rust, and for that I wish the Facepunch team a happy 5th birthday!

  • reificator 5 days ago

    This is in regards to the programming language, not the survival game.

  • h8hawk 5 days ago

    In hacker news "Rust" is a programming language!