Crate derive_syn_parse

Derive macro for syn::parse::Parse

A common pattern when writing custom syn parsers is repeating <name>: input.parse()? for each field in the output. #[derive(Parse)] handles that for you, with some extra helpful customization.

Usage

Using this crate is as simple as adding it to your 'Cargo.toml' and importing the derive macro:

# Cargo.toml

[dependencies]
syn-derive-parse = "0.1"

// your_file.rs
use syn_derive_parse::Parse;

#[derive(Parse)]
struct CustomParseable {
    // ...
}

The derived implementation of Parse always parses in the order that the fields are given.

This crate is intended for users who are already making heavy use of syn.

Motivation

When writing rust code that makes heavy use of syn's parsing functionality, we often end up writing things like:

use syn::parse::{Parse, ParseStream};
use syn::{Ident, Token, Type};

// A simplified struct field
//
//     x: i32
struct MyField {
    ident: Ident,
    colon_token: Token![:],
    ty: Type,
}

impl Parse for MyField {
    fn parse(input: ParseStream) -> syn::Result<Self> {
        Ok(MyField {
            ident: input.parse()?,
            colon_token: input.parse()?,
            ty: input.parse()?,
        })
    }
}

This is really repetetive! Ideally, we'd like to just #[derive(Parse)] and have it work. And so we can! (for the most part) Adding #[derive(Parse)] to the previous struct produces an equivalent implementation of Parse:

use syn::{Ident, Token, Type};
use syn_derive_parse::Parse;

#[derive(Parse)]
struct MyField {
    ident: Ident,
    colon_token: Token![:],
    ty: Type,
}

Of course, there are more complicated cases. This is mainly covered immediately below in the 'Advanced usage' section.

Advanced usage

There's a moderate collection of helper attributes that can be applied to fields and generic parameters to customize the generated implementation of Parse. Each of these are demonstrated with the implementation that they produce. Please note that the produced implementation is typically not identical to what's shown here.

All of the examples are fairly contrived, I know. The reality of the matter is that - if you would find this useful - it's probably true that your use-case is much more complicated than would make sense for a short example. (If it isn't, let me know! It would be great to include it here!)

List of helper attributes

• #[paren]
• #[bracket]
• #[brace]
• #[inside]
• #[call]
• #[parse_terminated]
• #[no_parse_bound]

#[paren] / #[bracket] / #[brace]

Because the derive macro has no fool-proof method for determining by itself whether a field type is any of syn::token::{Paren, Bracket, Brace}, these three serve to provide that information instead.

These are typically used in conjunction with #[inside].

// A single-argument function call
//
//     so_long(and_thanks + for_all * the_fish)
#[derive(Parse)]
struct SingleArgFn {
    ident: Ident,
    #[paren]
    paren_token: Paren,
    #[inside(paren_token)]
    arg: Expr,
}

produces

impl Parse for SingleArgFn {
    fn parse(input: ParseStream) -> syn::Result<Self> {
        let paren;
        Ok(SingleArgFn {
            ident: input.parse()?,
            paren_token: syn::parenthesized!(paren in input),
            arg: paren.parse()?,
        })
    }
}

#[inside(..)]

This is a companion to #[paren]/#[bracket]/#[brace] - given a field name to use, this attribute indicates that the field should be parsed using a previous field as the source.

use syn::token::Bracket;
use syn::{Type, Token, Expr};

// An array type required to have a length
//
//     [i32; 4]
#[derive(Parse)]
struct KnownLengthArrayType {
    #[bracket]
    bracket_token: Bracket,

    // Note that `#[inside(..)]` must be applied to all of the fields that
    // are in the brackets!
    #[inside(bracket_token)]
    ty: Type,
    #[inside(bracket_token)]
    semi_token: Token![;],
    #[inside(bracket_token)]
    expr: Expr,
}

produces

impl Parse for KnownLengthArrayType {
    fn parse(input: ParseStream) -> syn::Result<Self> {
        let bracket;
        Ok(KnownLengthArrayType {
            bracket_token: syn::braced!(bracket in input),
            ty: bracket.parse()?,
            semi_token: bracket.parse()?,
            expr: bracket.parse()?,
        })
    }
}

#[call(..)]

Given a path to a function, this attribute specifies that the value of the field should be instead calculated by a call to input.parse(..) with a given path. The best example is taken straight from the syn documentation itself:

use syn::{Attribute, Ident, Token};

// Parses a unit struct with attributes.
//
//     #[path = "s.tmpl"]
//     struct S;
#[derive(Parse)]
struct UnitStruct {
    #[call(Attribute::parse_outer)]
    attrs: Vec<Attribute>,
    struct_token: Token![struct],
    name: Ident,
    semi_token: Token![;],
}

produces

impl Parse for UnitStruct {
    fn parse(input: ParseStream) -> syn::Result<Self> {
        Ok(UnitStruct {
            attrs: input.call(Attribute::parse_outer)?,
            struct_token: input.parse()?,
            name: input.parse()?,
            semi_token: input.parse()?,
        })
    }
}

#[parse_terminated(..)]

Just as we have #[call(..)] for ParseStream::call, we have #[parse_terminated] for ParseStream::parse_terminated. Here's the same example that the ParseStream method uses:

// Parse a simplified tuple struct syntax like:
//
//     struct S(A, B);
struct TupleStruct {
    struct_token: Token![struct],
    ident: Ident,
    #[paren]
    paren_token: token::Paren,
    #[inside(paren_token)]
    #[parse_terminated(Type::parse)]
    fields: Punctuated<Type, Token![,]>,
    semi_token: Token![;],
}

produces

impl Parse for TupleStruct {
    fn parse(input: ParseStream) -> Result<Self> {
        let content;
        Ok(TupleStruct {
            struct_token: input.parse()?,
            ident: input.parse()?,
            paren_token: parenthesized!(content in input),
            fields: content.parse_terminated(Type::parse)?,
            semi_token: input.parse()?,
        })
    }
}

#[no_parse_bound]

By default, all type parameters in the source struct are required to implement Parse. The #[no_parse_bound] attribute can be applied to them to lift that restriction. This is perhaps less applicable, but available for certain use-cases:

use std::marker::PhantomData;

// [pretend this has an implementation of `Parse` that does nothing]
struct ParseablePhantomData<T>(PhantomData<T>);

#[derive(Parse)]
struct Foo<#[no_parse_bound] T, S> {
    bar: S,
    _marker: ParseablePhantomData<T>,
}

produces

impl<T, S: Parse> Parse for Foo<T, S> {
    fn parse(input: ParseStream) -> syn::Result<Self> {
        Ok(Foo {
            bar: input.parse()?,
            _marker: input.parse()?,
        })
    }
}

Known limitations

The derive macro is only available for structs. While actually possible, it's currently considered outside of the scope of this crate to generate implementations of Parse for enums. This is because they will always require some kind of lookahead (either via ParseStream::peek or ParseStream::fork).

Derive Macros

Parse