Recent and future pattern matching improvements

栏目: IT技术 · 发布时间: 4年前

内容简介:Much of writing software revolves around checking if some data has some shape ("pattern"), extracting information from it, and then reacting if there was a match. To facilitate this, many modern languages, Rust included, support what is known as "pattern m

Much of writing software revolves around checking if some data has some shape ("pattern"), extracting information from it, and then reacting if there was a match. To facilitate this, many modern languages, Rust included, support what is known as "pattern matching".

If you are new to Rust or want to refresh your knowledge, you may first want to read chapters 6, Enums and Pattern Matching and 18, Patterns and Matching in the book, or read more about match expressions andpatterns in the reference.

Pattern matching in Rust works by checking if a place in memory (the "data") matches a certain pattern . In this post, we will look at some recent improvements to patterns soon available in stable Rust as well as some more in already available in nightly.

If you are familiar with the nightly features discussed and would like to help out with the efforts to drive them to stable, jump ahead to.

Subslice patterns, [head, tail @ ..]

Lists are one of the most basic and common data structures found in software. In Rust, lists are usually a contiguous sequence of elements in memory, or a slice .

Since slices are so commonplace, it is important that working with them is easy. To that end, we stabilized fixed-length slice patterns in Rust 1.26.0 . So now it is possible to e.g., write let [a, b, c] = my_array; to destructure an array of 3 elements. Oftentimes, however, we're working with a slice of unknown length, so given only fixed-length slice patterns, we have to provide a fallback match arm with e.g. _ as the pattern.

In Rust 1.42.0, we are stabilizing subslice patterns . To introduce a subslice pattern, we use .. which denotes a variable-length gap, matching as many elements as possible not matched by the patterns before and after the .. . For example, in a parser, we would like to error when a list of attributes, attrs , is not followed by an item,so we write:

/// Recover if we parsed attributes and expected an item but there was none.
fn recover_attrs_no_item(&mut self, attrs: &[Attribute]) -> PResult<'a, ()> {
    let (start, end) = match attrs {
        [] => return Ok(()),
        [x0] => (x0, x0),
        [x0, .., xn] => (x0, xn),
    };
    let msg = if end.is_doc_comment() {
        "expected item after doc comment"
    } else {
        "expected item after attributes"
    };
    let mut err = self.struct_span_err(end.span, msg);
    if end.is_doc_comment() {
        err.span_label(end.span, "this doc comment doesn't document anything");
    }
    if let [.., penultimate, _] = attrs {
        err.span_label(start.span.to(penultimate.span), "other attributes here");
    }
    Err(err)
}

Here we have two subslice patterns, the first one being [x0, .., xn] . In this case, the pattern binds x0 , the first element, and xn , the last element, and ignores everything in the middle, matching a slice with at least two elements in total. Meanwhile, [] and [x0] match cases with fewer than two elements, so the compiler knows that we have covered all possibilities. In the latter case, we extract the penultimate element of the slice, which, as the name suggests, also requires that the slice has at least two elements.

We can also bind a subslice to a variable. For example, suppose we want to disallow ... in all but the last parameter of a function. If so, we can write:

match &*fn_decl.inputs {
    ... // other arms
    [ps @ .., _] => {
        for Param { ty, span, .. } in ps {
            if let TyKind::CVarArgs = ty.kind {
                self.err_handler().span_err(
                    *span,
                    "`...` must be the last argument of a C-variadic function",
                );
            }
        }
    }
}

Here, ps @ .. will bind the initial elements of the slice to ps and ignore the last element.

After more than 7 years of baking in nightly, with many twists and turns, subslice patterns will finally be stable. To get here, we've had to redesign the feature, plug soundness holes in the borrow checker, and substantially refactor the exhaustiveness checker. For more on how we got here, read the stabilization report , Thomas Hartmann's blog post , and stay tuned for the 1.42.0 release announcement on the 12th of March.

Nested OR-patterns

When pattern matching on an enum , the logic for some of the variants may be exactly the same. To avoid repeating ourselves, the | separator in match , if let , or while let expressions can be used to say that the branch should be taken if any of the | -separated patterns match. For example, we may write:

// Any local node that may call something in its body block should be explored.
fn should_explore(tcx: TyCtxt<'_>, hir_id: hir::HirId) -> bool {
    match tcx.hir().find(hir_id) {
        Some(Node::Item(..))
        | Some(Node::ImplItem(..))
        | Some(Node::ForeignItem(..))
        | Some(Node::TraitItem(..))
        | Some(Node::Variant(..))
        | Some(Node::AnonConst(..))
        | Some(Node::Pat(..)) => true,
        _ => false,
    }
}

This is serviceable, but Some(_) is still repeated several times. With #![feature(or_patterns)] , which recently became usable on nightly, this repetition can be avoided:

// Any local node that may call something in its body block should be explored.
fn should_explore(tcx: TyCtxt<'_>, hir_id: hir::HirId) -> bool {
    match tcx.hir().find(hir_id) {
        Some(
            Node::Item(..)
            | Node::ImplItem(..)
            | Node::ForeignItem(..)
            | Node::TraitItem(..)
            | Node::Variant(..)
            | Node::AnonConst(..)
            | Node::Pat(..),
        ) => true,
        _ => false,
    }
}

Previously, when using | in a match expression, the | syntax was part of match itself. With or_patterns , this is now part of patterns themselves, so you can nest OR-patterns arbitrarily, and use them in let statements too:

let Ok(x) | Err(x) = foo();

An OR-pattern covers the union of all the | -ed ("or-ed") patterns. To ensure that whatever alternative matched, all bindings are consistent and initialized, each or-ed pattern must include the exact same set of bindings, with the same types, and the same binding modes.

Bindings after @

When matching on a certain substructure, you sometimes want to hold on to the whole. For example, given Some(Expr { .. }) , you would like to bind the outer Some(_) layer. In Rust, this can be done using e.g., expr @ Some(Expr { .. }) , which binds the matched place to expr while also ensuring that it matches Some(Expr { .. }) .

Suppose also that Expr has a field span that you would also use. In ancient times, that is before Rust 1.0, this was possible, but today, it results in an error:

error[E0303]: pattern bindings are not allowed after an `@`
 --> src/lib.rs:L:C
  |
L |         bar @ Some(Expr { span }) => {}
  |                           ^^^^ not allowed after `@`

This was turned into an error in #16053 , mainly due to the difficulties of encoding borrow checking rules in a sound way in the old AST based borrow checker.

Since then, we have removed the old borrow checker in favor of one based on MIR , which is a simpler, and more appropriate data structure for borrow checking. Specifically, in the case of a statement like let ref x @ ref y = a; , we would get roughly the same MIR as if we had used let x = &a; let y = &a; .

So now that having bindings to the right of @ is handled uniformly and correctly by the borrow checker (e.g., the compiler won't allow ref x @ ref mut y ), we have decided to allow them under #![feature(bindings_after_at)] , now available on nightly. With the feature gate enabled, you may for example write:

#![feature(bindings_after_at)]

fn main() {
    if let x @ Some(y) = Some(0) {
        dbg!(x, y);
    }
}

Our hope is that with providing this feature, we remove one surprising corner of the language.

Combining by-move and by- ref bindings

For similar reasons as noted in the case of bindings after @ , Rust does not currently allow you to combine normal by-move bindings with those that are by- ref . For example, should you write...:

fn main() {
    let tup = ("foo".to_string(), 0);
    let (x, ref y) = tup;
}

... you would get an error:

error[E0009]: cannot bind by-move and by-ref in the same pattern
 --> src/main.rs:3:10
  |
3 |     let (x, ref y) = tup;
  |          ^  ----- by-ref pattern here
  |          |
  |          by-move pattern here

At the same time, however, the compiler is perfectly happy to allow...:

fn main() {
    let tup = ("foo".to_string(), 0);
    let x = tup.0;
    let ref y = tup.1;
}

... even though there is no semantic difference between these programs.

Now that we have moved to the new borrow checker, as outlined in the previous section, we have relaxed this restriction on nightly as well, so under #![feature(move_ref_pattern)] , you may write:

#![feature(move_ref_pattern)]

fn main() {
    let tup = ("foo".to_string(), 0);
    let (x, ref y) = tup;
}

How can I help?

To recap, we have three unstable features, all improving pattern matching in different ways:

To help us transition these features over to stable Rust, we need your help to ensure that they meet the expected quality standards. To help out, consider:

  • Using the features in your code where applicable, if a nightly compiler is something you are OK with, and reporting any bugs, problems, deficiencies in diagnostics, etc. as issues.
  • Looking through the reported issues under the feature gate labels (e.g., F-or_patterns ) and seeing if you can help out with any of them.
    • In particular, if you can help out with writing tests, that is appreciated.

Thanks for reading, and happy pattern matching!


以上就是本文的全部内容,希望本文的内容对大家的学习或者工作能带来一定的帮助,也希望大家多多支持 码农网

查看所有标签

猜你喜欢:

本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们

Agile Web Development with Rails, Third Edition

Agile Web Development with Rails, Third Edition

Sam Ruby、Dave Thomas、David Heinemeier Hansson / Pragmatic Bookshelf / 2009-03-17 / USD 43.95

Rails just keeps on changing. Rails 2, released in 2008, brings hundreds of improvements, including new support for RESTful applications, new generator options, and so on. And, as importantly, we’ve a......一起来看看 《Agile Web Development with Rails, Third Edition》 这本书的介绍吧!

在线进制转换器
在线进制转换器

各进制数互转换器

XML、JSON 在线转换
XML、JSON 在线转换

在线XML、JSON转换工具

HEX CMYK 转换工具
HEX CMYK 转换工具

HEX CMYK 互转工具