Struct `Attribute`

struct Attribute { ... }

An attribute, like #[repr(transparent)].

Syntax

Rust has six types of attributes.

Outer attributes like #[repr(transparent)]. These appear outside or in front of the item they describe.
Inner attributes like #![feature(proc_macro)]. These appear inside of the item they describe, usually a module.
Outer one-line doc comments like /// Example.
Inner one-line doc comments like //! Please file an issue.
Outer documentation blocks /** Example */.
Inner documentation blocks /*! Please file an issue */.

The style field of type AttrStyle distinguishes whether an attribute is outer or inner.

Every attribute has a path that indicates the intended interpretation of the rest of the attribute's contents. The path and the optional additional contents are represented together in the meta field of the attribute in three possible varieties:

Meta::Path — attributes whose information content conveys just a path, for example the #[test] attribute.
Meta::List — attributes that carry arbitrary tokens after the path, surrounded by a delimiter (parenthesis, bracket, or brace). For example #[derive(Copy)] or #[precondition(x < 5)].
Meta::NameValue — attributes with an = sign after the path, followed by a Rust expression. For example #[path = "sys/windows.rs"].

All doc comments are represented in the NameValue style with a path of "doc", as this is how they are processed by the compiler and by macro_rules! macros.

#[derive(Copy, Clone)]
  ~~~~~~Path
  ^^^^^^^^^^^^^^^^^^^Meta::List

#[path = "sys/windows.rs"]
  ~~~~Path
  ^^^^^^^^^^^^^^^^^^^^^^^Meta::NameValue

#[test]
  ^^^^Meta::Path

Parsing from tokens to Attribute

This type does not implement the Parse trait and thus cannot be parsed directly by ParseStream::parse. Instead use ParseStream::call with one of the two parser functions Attribute::parse_outer or Attribute::parse_inner depending on which you intend to parse.

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

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

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

Parsing from Attribute to structured arguments

The grammar of attributes in Rust is very flexible, which makes the syntax tree not that useful on its own. In particular, arguments of the Meta::List variety of attribute are held in an arbitrary tokens: TokenStream. Macros are expected to check the path of the attribute, decide whether they recognize it, and then parse the remaining tokens according to whatever grammar they wish to require for that kind of attribute. Use parse_args() to parse those tokens into the expected data structure.

Doc comments

The compiler transforms doc comments, such as /// comment and /*! comment */, into attributes before macros are expanded. Each comment is expanded into an attribute of the form #[doc = r"comment"].

As an example, the following mod items are expanded identically:

# use syn::{ItemMod, parse_quote};
let doc: ItemMod = parse_quote! {
    /// Single line doc comments
    /// We write so many!
    /**
     * Multi-line comments...
     * May span many lines
     */
    mod example {
        //! Of course, they can be inner too
        /*! And fit in a single line */
    }
};
let attr: ItemMod = parse_quote! {
    #[doc = r" Single line doc comments"]
    #[doc = r" We write so many!"]
    #[doc = r"
     * Multi-line comments...
     * May span many lines
     "]
    mod example {
        #![doc = r" Of course, they can be inner too"]
        #![doc = r" And fit in a single line "]
    }
};
assert_eq!(doc, attr);

Fields

pound_token: Pound
style: AttrStyle
bracket_token: Bracket
meta: Meta

Implementations

`impl Attribute`

fn path(self: &Self) -> &Path

Returns the path that identifies the interpretation of this attribute.

For example this would return the test in #[test], the derive in #[derive(Copy)], and the path in #[path = "sys/windows.rs"].

fn parse_args<T: Parse>(self: &Self) -> Result<T>

Parse the arguments to the attribute as a syntax tree.

This is similar to pulling out the TokenStream from Meta::List and doing syn::parse2::<T>(meta_list.tokens), except that using parse_args the error message has a more useful span when tokens is empty.

The surrounding delimiters are not included in the input to the parser.

#[my_attr(value < 5)]
          ^^^^^^^^^ what gets parsed

Example

use syn::{parse_quote, Attribute, Expr};

let attr: Attribute = parse_quote! {
    #[precondition(value < 5)]
};

if attr.path().is_ident("precondition") {
    let precondition: Expr = attr.parse_args()?;
    // ...
}
# anyhow::Ok(())

fn parse_args_with<F: Parser>(self: &Self, parser: F) -> Result<<F as >::Output>

Parse the arguments to the attribute using the given parser.

Example

use syn::{parse_quote, Attribute};

let attr: Attribute = parse_quote! {
    #[inception { #[brrrrrrraaaaawwwwrwrrrmrmrmmrmrmmmmm] }]
};

let bwom = attr.parse_args_with(Attribute::parse_outer)?;

// Attribute does not have a Parse impl, so we couldn't directly do:
// let bwom: Attribute = attr.parse_args()?;
# anyhow::Ok(())

fn parse_nested_meta<impl FnMut(ParseNestedMeta) -> Result<()>: FnMut(ParseNestedMeta<'_>) -> Result<()>>(self: &Self, logic: impl FnMut(ParseNestedMeta<'_>) -> Result<()>) -> Result<()>

Parse the arguments to the attribute, expecting it to follow the conventional structure used by most of Rust's built-in attributes.

The Meta Item Attribute Syntax section in the Rust reference explains the convention in more detail. Not all attributes follow this convention, so [parse_args()][Self::parse_args] is available if you need to parse arbitrarily goofy attribute syntax.

Example

We'll parse a struct, and then parse some of Rust's #[repr] attribute syntax.

use syn::{parenthesized, parse_quote, token, ItemStruct, LitInt};

let input: ItemStruct = parse_quote! {
    #[repr(C, align(4))]
    pub struct MyStruct(u16, u32);
};

let mut repr_c = false;
let mut repr_transparent = false;
let mut repr_align = None::<usize>;
let mut repr_packed = None::<usize>;
for attr in &input.attrs {
    if attr.path().is_ident("repr") {
        attr.parse_nested_meta(|meta| {
            // #[repr(C)]
            if meta.path.is_ident("C") {
                repr_c = true;
                return Ok(());
            }

            // #[repr(transparent)]
            if meta.path.is_ident("transparent") {
                repr_transparent = true;
                return Ok(());
            }

            // #[repr(align(N))]
            if meta.path.is_ident("align") {
                let content;
                parenthesized!(content in meta.input);
                let lit: LitInt = content.parse()?;
                let n: usize = lit.base10_parse()?;
                repr_align = Some(n);
                return Ok(());
            }

            // #[repr(packed)] or #[repr(packed(N))], omitted N means 1
            if meta.path.is_ident("packed") {
                if meta.input.peek(token::Paren) {
                    let content;
                    parenthesized!(content in meta.input);
                    let lit: LitInt = content.parse()?;
                    let n: usize = lit.base10_parse()?;
                    repr_packed = Some(n);
                } else {
                    repr_packed = Some(1);
                }
                return Ok(());
            }

            Err(meta.error("unrecognized repr"))
        })?;
    }
}
# anyhow::Ok(())

Alternatives

In some cases, for attributes which have nested layers of structured content, the following less flexible approach might be more convenient:

# use syn::{parse_quote, ItemStruct};
#
# let input: ItemStruct = parse_quote! {
#     #[repr(C, align(4))]
#     pub struct MyStruct(u16, u32);
# };
#
use syn::punctuated::Punctuated;
use syn::{parenthesized, token, Error, LitInt, Meta, Token};

let mut repr_c = false;
let mut repr_transparent = false;
let mut repr_align = None::<usize>;
let mut repr_packed = None::<usize>;
for attr in &input.attrs {
    if attr.path().is_ident("repr") {
        let nested = attr.parse_args_with(Punctuated::<Meta, Token![,]>::parse_terminated)?;
        for meta in nested {
            match meta {
                // #[repr(C)]
                Meta::Path(path) if path.is_ident("C") => {
                    repr_c = true;
                }

                // #[repr(align(N))]
                Meta::List(meta) if meta.path.is_ident("align") => {
                    let lit: LitInt = meta.parse_args()?;
                    let n: usize = lit.base10_parse()?;
                    repr_align = Some(n);
                }

                /* ... */

                _ => {
                    return Err(Error::new_spanned(meta, "unrecognized repr"));
                }
            }
        }
    }
}
# Ok(())

fn parse_outer(input: ParseStream<'_>) -> Result<Vec<Self>>

Parses zero or more outer attributes from the stream.

Example

See Parsing from tokens to Attribute.

fn parse_inner(input: ParseStream<'_>) -> Result<Vec<Self>>

Parses zero or more inner attributes from the stream.

Example

See Parsing from tokens to Attribute.

`impl Clone for Attribute`

fn clone(self: &Self) -> Self

`impl Debug for Attribute`

fn fmt(self: &Self, formatter: &mut Formatter<'_>) -> Result

`impl Eq for Attribute`

`impl Freeze for Attribute`

`impl Hash for Attribute`

fn hash<H>(self: &Self, state: &mut H) where H: Hasher

`impl PartialEq for Attribute`

fn eq(self: &Self, other: &Self) -> bool

`impl RefUnwindSafe for Attribute`

`impl Send for Attribute`

`impl Sync for Attribute`

`impl ToTokens for Attribute`

fn to_tokens(self: &Self, tokens: &mut TokenStream)

`impl Unpin for Attribute`

`impl UnsafeUnpin for Attribute`

`impl UnwindSafe for Attribute`

`impl<T> Any for Attribute`

fn type_id(self: &Self) -> TypeId

`impl<T> Borrow for Attribute`

fn borrow(self: &Self) -> &T

`impl<T> BorrowMut for Attribute`

fn borrow_mut(self: &mut Self) -> &mut T

`impl<T> CloneToUninit for Attribute`

unsafe fn clone_to_uninit(self: &Self, dest: *mut u8)

`impl<T> From for Attribute`

fn from(t: T) -> T: Returns the argument unchanged.

`impl<T> Spanned for Attribute`

fn span(self: &Self) -> Span

`impl<T> ToOwned for Attribute`

fn to_owned(self: &Self) -> T
fn clone_into(self: &Self, target: &mut T)

`impl<T, U> Into for Attribute`

fn into(self: Self) -> U

Calls U::from(self).

That is, this conversion is whatever the implementation of [From]<T> for U chooses to do.

`impl<T, U> TryFrom for Attribute`

fn try_from(value: U) -> Result<T, <T as TryFrom<U>>::Error>

`impl<T, U> TryInto for Attribute`

fn try_into(self: Self) -> Result<U, <U as TryFrom<T>>::Error>