darling is a crate for proc macro authors, which enables parsing attributes into structs. It is heavily inspired by serde both in its internals and in its API.
- Easy and declarative parsing of macro input - make your proc-macros highly controllable with minimal time investment.
- Great validation and errors, no work required. When users of your proc-macro make a mistake,
darlingmakes sure they get error markers at the right place in their source, and provides "did you mean" suggestions for misspelled fields.
darling provides a set of traits which can be derived or manually implemented.
FromMetais used to extract values from a meta-item in an attribute. Implementations are likely reusable for many libraries, much likeFromStrorserde::Deserialize. Trait implementations are provided for primitives, some std types, and somesyntypes.FromDeriveInputis implemented or derived by each proc-macro crate which depends ondarling. This is the root for input parsing; it gets access to the identity, generics, and visibility of the target type, and can specify which attribute names should be parsed or forwarded from the input AST.FromFieldis implemented or derived by each proc-macro crate which depends ondarling. Structs deriving this trait will get access to the identity (if it exists), type, and visibility of the field.FromVariantis implemented or derived by each proc-macro crate which depends ondarling. Structs deriving this trait will get access to the identity and contents of the variant, which can be transformed the same as any otherdarlinginput.FromAttributesis a lower-level version of the more-specificFromDeriveInput,FromField, andFromVarianttraits. Structs deriving this trait get a meta-item extractor and error collection which works for any syntax element, including traits, trait items, and functions. This is useful for non-derive proc macros.
darling::astprovides generic types for representing the AST.darling::usageprovides traits and functions for determining where type parameters and lifetimes are used in a struct or enum.darling::utilprovides helper types with specialFromMetaimplementations, such asPathList.
use darling::{FromDeriveInput, FromMeta};
#[derive(Default, FromMeta)]
#[darling(default)]
pub struct Lorem {
#[darling(rename = "sit")]
ipsum: bool,
dolor: Option<String>,
}
#[derive(FromDeriveInput)]
#[darling(attributes(my_crate), forward_attrs(allow, doc, cfg))]
pub struct MyTraitOpts {
ident: syn::Ident,
attrs: Vec<syn::Attribute>,
lorem: Lorem,
}The above code will then be able to parse this input:
/// A doc comment which will be available in `MyTraitOpts::attrs`.
#[derive(MyTrait)]
#[my_crate(lorem(dolor = "Hello", sit))]
pub struct ConsumingType;Non-derive attribute macros are supported.
To parse arguments for attribute macros, derive FromMeta on the argument receiver type, then use darling::ast::NestedMeta::parse_meta_list to convert the arguments TokenStream to a Vec<NestedMeta>, then pass that to the derived from_list method on your argument receiver type.
This will produce a normal darling::Result<T> that can be used the same as a result from parsing a DeriveInput.
use darling::{Error, FromMeta};
use darling::ast::NestedMeta;
use syn::ItemFn;
use proc_macro::TokenStream;
#[derive(Debug, FromMeta)]
struct MacroArgs {
#[darling(default)]
timeout_ms: Option<u16>,
path: String,
}
#[proc_macro_attribute]
pub fn your_attr(args: TokenStream, input: TokenStream) -> TokenStream {
let attr_args = match NestedMeta::parse_meta_list(args.into()) {
Ok(v) => v,
Err(e) => { return TokenStream::from(Error::from(e).write_errors()); }
};
let _input = syn::parse_macro_input!(input as ItemFn);
let _args = match MacroArgs::from_list(&attr_args) {
Ok(v) => v,
Err(e) => { return TokenStream::from(e.write_errors()); }
};
// do things with `args`
unimplemented!()
}use your_crate::your_attr;
#[your_attr(path = "hello", timeout_ms = 15)]
fn do_stuff() {
println!("Hello");
}Darling's features are built to work well for real-world projects.
- Defaults: Supports struct- and field-level defaults, using the same path syntax as
serde. Additionally,Option<T>anddarling::util::Flagfields are innately optional; you don't need to declare#[darling(default)]for those. - Field Renaming: Fields can have different names in usage vs. the backing code.
- Auto-populated fields: Structs deriving
FromDeriveInputandFromFieldcan declare properties namedident,vis,ty,attrs, andgenericsto automatically get copies of the matching values from the input AST.FromDeriveInputadditionally exposesdatato get access to the body of the deriving type, andFromVariantexposesfields.- Transformation of forwarded attributes: You can add
#[darling(with=path)]to theattrsfield to use a custom function to transform the forwarded attributes before they're provided to your struct. The function signature isfn(Vec<Attribute>) -> darling::Result<T>, whereTis the type you declared for theattrsfield. Returning an error from this function will propagate with all other parsing errors.
- Transformation of forwarded attributes: You can add
- Mapping function: Use
#[darling(map="path")]or#[darling(and_then="path")]to specify a function that runs on the result of parsing a meta-item field. This can change the return type, which enables you to parse to an intermediate form and convert that to the type you need in your struct. - Skip fields: Use
#[darling(skip)]to mark a field that shouldn't be read from attribute meta-items. - Multiple-occurrence fields: Use
#[darling(multiple)]on aVecfield to allow that field to appear multiple times in the meta-item. Each occurrence will be pushed into theVec. - Span access: Use
darling::util::SpannedValuein a struct to get access to that meta item's source code span. This can be used to emit warnings that point at a specific field from your proc macro. In addition, you can usedarling::Error::write_errorsto automatically get precise error location details in most cases. - "Did you mean" suggestions: Compile errors from derived darling trait impls include suggestions for misspelled fields.
- Struct flattening: Use
#[darling(flatten)]to remove one level of structure when presenting your meta item to users. Fields that are not known to the parent struct will be forwarded to theflattenfield.
Some proc-macros only work on structs, while others need enums whose variants are either unit or newtype variants.
Darling makes this sort of validation extremely simple.
On the receiver that derives FromDeriveInput, add #[darling(supports(...))] and then list the shapes that your macro should accept.
| Name | Description |
|---|---|
any |
Accept anything |
struct_any |
Accept any struct |
struct_named |
Accept structs with named fields, e.g. struct Example { field: String } |
struct_newtype |
Accept newtype structs, e.g. struct Example(String) |
struct_tuple |
Accept tuple structs, e.g. struct Example(String, String) |
struct_unit |
Accept unit structs, e.g. struct Example; |
enum_any |
Accept any enum |
enum_named |
Accept enum variants with named fields |
enum_newtype |
Accept newtype enum variants |
enum_tuple |
Accept tuple enum variants |
enum_unit |
Accept unit enum variants |
Each one is additive, so listing #[darling(supports(struct_any, enum_newtype))] would accept all structs and any enum where every variant is a newtype variant.
This can also be used when deriving FromVariant, without the enum_ prefix.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent