Announcing Rust 1.27

June 21, 2018 · The Rust Core Team

The Rust team is happy to announce a new version of Rust, 1.27.0. Rust is a systems programming language focused on safety, speed, and concurrency.

If you have a previous version of Rust installed via rustup, getting Rust 1.27.0 is as easy as:

$ rustup update stable

If you don't have it already, you can get rustup from the appropriate page on our website, and check out the detailed release notes for 1.27.0 on GitHub.

Additionally, we would like to draw attention to something: just before the release of 1.27.0, we found a bug in the 'default match bindings' feature introduced in 1.26.0 that can possibly introduce unsoundness. Since it was discovered very late in the release process, and has been present since 1.26.0, we decided to stick to our release train model. We expect to put out a 1.27.1 with that fix applied soon, and if there's demand, possibly a 1.26.3 as well. More information about the specifics here will come in that release announcement.

What's in 1.27.0 stable

This release has two big language features that people have been waiting for. But first, a small comment on documentation: All books in the Rust Bookshelf are now searchable! For example, here's a search of "The Rust Programming Language" for 'borrow'. This will hopefully make it much easier to find what you're looking for. Additionally, there's one new book: the rustc Book. This book explains how to use rustc directly, as well as some other useful information, like a list of all lints.

SIMD

Okay, now for the big news: the basics of SIMD are now available! SIMD stands for "single instruction, multiple data." Consider a function like this:

pub fn foo(a: &[u8], b: &[u8], c: &mut [u8]) {
    for ((a, b), c) in a.iter().zip(b).zip(c) {
        *c = *a + *b;
    }
}

Here, we're taking two slices, and adding the numbers together, placing the result in a third slice. The simplest possible way to do this would be to do exactly what the code does, and loop through each set of elements, add them together, and store it in the result. However, compilers can often do better. LLVM will often "autovectorize" code like this, which is a fancy term for "use SIMD." Imagine that a and b were both 16 elements long. Each element is a u8, and so that means that each slice would be 128 bits of data. Using SIMD, we could put both a and b into 128 bit registers, add them together in a *single* instruction, and then copy the resulting 128 bits into c. That'd be much faster!

While stable Rust has always been able to take advantage of autovectorization, sometimes, the compiler just isn't smart enough to realize that we can do something like this. Additionally, not every CPU has these features, and so LLVM may not use them so your program can be used on a wide variety of hardware. So, in Rust 1.27, the addition of the std::arch module allows us to use these kinds of instructions directly, which means we don't need to rely on a smart compiler. Additionally, it includes some features that allow us to choose a particular implementation based on various criteria. For example:

#[cfg(all(any(target_arch = "x86", target_arch = "x86_64"),
      target_feature = "avx2"))]
fn foo() {
    #[cfg(target_arch = "x86")]
    use std::arch::x86::_mm256_add_epi64;
    #[cfg(target_arch = "x86_64")]
    use std::arch::x86_64::_mm256_add_epi64;

    unsafe {
        _mm256_add_epi64(...);
    }
}

Here, we use cfg flags to choose the correct version based on the machine we're targeting; on x86 we use that version, and on x86_64 we use its version. We can also choose at runtime:

fn foo() {
    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
    {
        if is_x86_feature_detected!("avx2") {
            return unsafe { foo_avx2() };
        }
    }

    foo_fallback();
}

Here, we have two versions of the function: one which uses AVX2, a specific kind of SIMD feature that lets you do 256-bit operations. The is_x86_feature_detected! macro will generate code that detects if your CPU supports AVX2, and if so, calls the foo_avx2 function. If not, then we fall back to a non-AVX implementation, foo_fallback. This means that our code will run super fast on CPUs that support AVX2, but still work on ones that don't, albeit slower.

If all of this seems a bit low-level and fiddly, well, it is! std::arch is specifically primitives for building these kinds of things. We hope to eventually stabilize a std::simd module with higher-level stuff in the future. But landing the basics now lets the ecosystem experiment with higher level libraries starting today. For example, check out the faster crate. Here's a code snippet with no SIMD:

let lots_of_3s = (&[-123.456f32; 128][..]).iter()
    .map(|v| {
        9.0 * v.abs().sqrt().sqrt().recip().ceil().sqrt() - 4.0 - 2.0
    })
    .collect::<Vec<f32>>();

To use SIMD with this code via faster, you'd change it to this:

let lots_of_3s = (&[-123.456f32; 128][..]).simd_iter()
    .simd_map(f32s(0.0), |v| {
        f32s(9.0) * v.abs().sqrt().rsqrt().ceil().sqrt() - f32s(4.0) - f32s(2.0)
    })
    .scalar_collect();

It looks almost the same: simd_iter instead of iter, simd_map instead of map, f32s(2.0) instead of 2.0. But you get a SIMD-ified version generated for you.

Beyond that, you may never write any of this yourself, but as always, the libraries you depend on may. For example, the regex crate has already added support, and a new release will contain these SIMD speedups without you needing to do anything at all!

dyn Trait

Rust's trait object syntax is one that we ultimately regret. If you'll recall, given a trait Foo, this is a trait object:

Box<Foo>

However, if Foo were a struct, it'd just be a normal struct placed inside a Box<T>. When designing the language, we thought that the similarity here was a good thing, but experience has demonstrated that it is confusing. And it's not just for the Box<Trait> case; impl SomeTrait for SomeOtherTrait is also technically valid syntax, but you almost always want to write impl<T> SomeTrait for T where T: SomeOtherTrait instead. Same with impl SomeTrait, which looks like it would add methods or possibly default implementations but in fact adds inherent methods to a trait object. Finally, with the recent addition of impl Trait syntax, it's impl Trait vs Trait when explaining things, and so that feels like Trait is what you should use, given that it's shorter, but in reality, that's not always true.

As such, in Rust 1.27, we have stabilized a new syntax, dyn Trait. A trait object now looks like this:

// old => new
Box<Foo> => Box<dyn Foo>
&Foo => &dyn Foo
&mut Foo => &mut dyn Foo

And similarly for other pointer types, Arc<Foo> is now Arc<dyn Foo>, etc. Due to backwards compatibility, we cannot remove the old syntax, but we have included a lint, which is set to allow by default, called bare-trait-object. If you want to lint against the older syntax, you can turn it on. We thought that it would throw far too many warnings to turn on by default at present.

Incidentally, we're working on a tool called rustfix that can automatically upgrade your code to newer idioms. It uses these sorts of lints to do so. Expect to hear more about rustfix in a future announcement.

#[must_use] on functions

Finally, the #[must_use] attribute is getting an upgrade: it can now be used on functions.

Previously, it only applied to types, like Result<T, E>. But now, you can do this:

#[must_use]
fn double(x: i32) -> i32 {
    2 * x
}

fn main() {
    double(4); // warning: unused return value of `double` which must be used

    let _ = double(4); // (no warning)
}

We've also enhanced several bits of the standard library to make use of this; Clone::clone, Iterator::collect, and ToOwned::to_owned will all start warning if you don't use their results, helping you notice expensive operations you may be throwing away by accident.

See the detailed release notes for more.

Library stabilizations

Several new APIs were stabilized this release:

See the detailed release notes for more.

Cargo features

Cargo has two small upgrades this release. First, it now takes a --target-dir flag if you'd like to change the target directory for a given invocation.

Additionally, a tweak to the way Cargo deals with targets has landed. Cargo will attempt to automatically discover tests, examples, and binaries within your project. However, sometimes explicit configuration is needed. But the initial implementation had a problem: let's say that you have two examples, and Cargo is discovering them both. You want to tweak one of them, and so you add a [[example]] to your Cargo.toml to configure its settings. Cargo currently sees that you've set one explicitly, and therefore, doesn't attempt to do any autodetection for the others. That's quite surprising.

As such, we've added several 'auto' keys to Cargo.toml We can't fix this behavior without possibly breaking projects that may have inadvertently been relying on it, and so, if you'd like to configure some targets, but not others, you can set the autoexamples key to true in the [package] section.

See the detailed release notes for more.

Contributors to 1.27.0

Many people came together to create Rust 1.27. We couldn't have done it without all of you. Thanks!