commit | 62bdd84672151e5b5f0a67b91b92bb2aa5cccf9a | [log] [tgz] |
---|---|---|
author | David Tolnay <dtolnay@gmail.com> | Sat Jun 08 06:03:08 2019 -0700 |
committer | David Tolnay <dtolnay@gmail.com> | Sat Jun 08 06:03:08 2019 -0700 |
tree | 40e78068645c60d1d41da742577e29b690f7390a | |
parent | 7ee1d633feb63b871dfb7662a391228fd1fbc03a [diff] |
Fix token order of async unsafe fn
Syn is a parsing library for parsing a stream of Rust tokens into a syntax tree of Rust source code.
Currently this library is geared toward use in Rust procedural macros, but contains some APIs that may be useful more generally.
Data structures — Syn provides a complete syntax tree that can represent any valid Rust source code. The syntax tree is rooted at syn::File
which represents a full source file, but there are other entry points that may be useful to procedural macros including syn::Item
, syn::Expr
and syn::Type
.
Custom derives — Of particular interest to custom derives is syn::DeriveInput
which is any of the three legal input items to a derive macro. An example below shows using this type in a library that can derive implementations of a trait of your own.
Parsing — Parsing in Syn is built around parser functions with the signature fn(ParseStream) -> Result<T>
. Every syntax tree node defined by Syn is individually parsable and may be used as a building block for custom syntaxes, or you may dream up your own brand new syntax without involving any of our syntax tree types.
Location information — Every token parsed by Syn is associated with a Span
that tracks line and column information back to the source of that token. These spans allow a procedural macro to display detailed error messages pointing to all the right places in the user's code. There is an example of this below.
Feature flags — Functionality is aggressively feature gated so your procedural macros enable only what they need, and do not pay in compile time for all the rest.
If you get stuck with anything involving procedural macros in Rust I am happy to provide help even if the issue is not related to Syn. Please file a ticket in this repo.
Version requirement: Syn supports any compiler version back to Rust's very first support for procedural macros in Rust 1.15.0. Some features especially around error reporting are only available in newer compilers or on the nightly channel.
The canonical custom derive using Syn looks like this. We write an ordinary Rust function tagged with a proc_macro_derive
attribute and the name of the trait we are deriving. Any time that derive appears in the user's code, the Rust compiler passes their data structure as tokens into our macro. We get to execute arbitrary Rust code to figure out what to do with those tokens, then hand some tokens back to the compiler to compile into the user's crate.
[dependencies] syn = "0.15" quote = "0.6" [lib] proc-macro = true
extern crate proc_macro; use proc_macro::TokenStream; use quote::quote; use syn::{parse_macro_input, DeriveInput}; #[proc_macro_derive(MyMacro)] pub fn my_macro(input: TokenStream) -> TokenStream { // Parse the input tokens into a syntax tree let input = parse_macro_input!(input as DeriveInput); // Build the output, possibly using quasi-quotation let expanded = quote! { // ... }; // Hand the output tokens back to the compiler TokenStream::from(expanded) }
The heapsize
example directory shows a complete working Macros 1.1 implementation of a custom derive. It works on any Rust compiler 1.15+. The example derives a HeapSize
trait which computes an estimate of the amount of heap memory owned by a value.
pub trait HeapSize { /// Total number of bytes of heap memory owned by `self`. fn heap_size_of_children(&self) -> usize; }
The custom derive allows users to write #[derive(HeapSize)]
on data structures in their program.
#[derive(HeapSize)] struct Demo<'a, T: ?Sized> { a: Box<T>, b: u8, c: &'a str, d: String, }
The token-based procedural macro API provides great control over where the compiler's error messages are displayed in user code. Consider the error the user sees if one of their field types does not implement HeapSize
.
#[derive(HeapSize)] struct Broken { ok: String, bad: std::thread::Thread, }
By tracking span information all the way through the expansion of a procedural macro as shown in the heapsize
example, token-based macros in Syn are able to trigger errors that directly pinpoint the source of the problem.
error[E0277]: the trait bound `std::thread::Thread: HeapSize` is not satisfied --> src/main.rs:7:5 | 7 | bad: std::thread::Thread, | ^^^^^^^^^^^^^^^^^^^^^^^^ the trait `HeapSize` is not implemented for `std::thread::Thread`
The lazy-static
example directory shows the implementation of a functionlike!(...)
procedural macro in which the input tokens are parsed using Syn's parsing API.
The example reimplements the popular lazy_static
crate from crates.io as a procedural macro.
lazy_static! { static ref USERNAME: Regex = Regex::new("^[a-z0-9_-]{3,16}$").unwrap(); }
The implementation shows how to trigger custom warnings and error messages on the macro input.
warning: come on, pick a more creative name --> src/main.rs:10:16 | 10 | static ref FOO: String = "lazy_static".to_owned(); | ^^^
When testing macros, we often care not just that the macro can be used successfully but also that when the macro is provided with invalid input it produces maximally helpful error messages. Consider using the trybuild
crate to write tests for errors that are emitted by your macro or errors detected by the Rust compiler in the expanded code following misuse of the macro. Such tests help avoid regressions from later refactors that mistakenly make an error no longer trigger or be less helpful than it used to be.
When developing a procedural macro it can be helpful to look at what the generated code looks like. Use cargo rustc -- -Zunstable-options --pretty=expanded
or the cargo expand
subcommand.
To show the expanded code for some crate that uses your procedural macro, run cargo expand
from that crate. To show the expanded code for one of your own test cases, run cargo expand --test the_test_case
where the last argument is the name of the test file without the .rs
extension.
This write-up by Brandon W Maister discusses debugging in more detail: Debugging Rust's new Custom Derive system.
Syn puts a lot of functionality behind optional features in order to optimize compile time for the most common use cases. The following features are available.
derive
(enabled by default) — Data structures for representing the possible input to a custom derive, including structs and enums and types.full
— Data structures for representing the syntax tree of all valid Rust source code, including items and expressions.parsing
(enabled by default) — Ability to parse input tokens into a syntax tree node of a chosen type.printing
(enabled by default) — Ability to print a syntax tree node as tokens of Rust source code.visit
— Trait for traversing a syntax tree.visit-mut
— Trait for traversing and mutating in place a syntax tree.fold
— Trait for transforming an owned syntax tree.clone-impls
(enabled by default) — Clone impls for all syntax tree types.extra-traits
— Debug, Eq, PartialEq, Hash impls for all syntax tree types.proc-macro
(enabled by default) — Runtime dependency on the dynamic library libproc_macro from rustc toolchain.Syn uses the proc-macro2 crate to emulate the compiler's procedural macro API in a stable way that works all the way back to Rust 1.15.0. This shim makes it possible to write code without regard for whether the current compiler version supports the features we use.
In general all of your code should be written against proc-macro2 rather than proc-macro. The one exception is in the signatures of procedural macro entry points, which are required by the language to use proc_macro::TokenStream
.
The proc-macro2 crate will automatically detect and use the compiler's data structures on sufficiently new compilers.