Attribute proc-macro to simplify deriving standard and other traits with custom generic type bounds.
The derive_where attribute can be used just like std's #[derive(...)] statements:
#[derive_where(Clone, Debug)] struct Example<T>(PhantomData<T>);
This will generate trait implementations for Example for any T, as opposed to std's derives, which would only implement these traits with T: Trait bound to the corresponding trait.
Multiple derive_where attributes can be added to an item, but only the first one must use any path qualifications.
#[derive_where::derive_where(Clone, Debug)] #[derive_where(Eq, PartialEq)] struct Example1<T>(PhantomData<T>);
If using a different package name, you must specify this:
#[derive_where(crate = derive_where_)] #[derive_where(Clone, Debug)] struct Example<T>(PhantomData<T>);
In addition, the following convenience options are available:
Separated from the list of traits with a semi-colon, types to bind to can be specified. This example will restrict the implementation for Example to T: Clone:
#[derive_where(Clone, Debug; T)] struct Example<T, U>(T, PhantomData<U>);
It is also possible to specify the bounds to be applied. This will bind implementation for Example to T: Super:
trait Super: Clone + Debug {} #[derive_where(Clone, Debug; T: Super)] struct Example<T>(PhantomData<T>);
But more complex trait bounds are possible as well. The example below will restrict the Clone implementation for Example to T::Type: Clone:
trait Trait { type Type; } struct Impl; impl Trait for Impl { type Type = i32; } #[derive_where(Clone, Debug; T::Type)] struct Example<T: Trait>(T::Type);
Any combination of options listed here can be used to satisfy a specific constrain. It is also possible to use multiple separate constrain specifications when required:
#[derive_where(Clone, Debug; T)] #[derive_where(Eq, PartialEq; U)] struct Example<T, U>(PhantomData<T>, PhantomData<U>);
Since Rust 1.62 deriving Default on an enum is possible with the #[default] attribute. Derive-where allows this with a #[derive_where(default)] attribute:
#[derive_where(Clone, Default)] enum Example<T> { #[derive_where(default)] A(PhantomData<T>), }
With a skip or skip_inner attribute fields can be skipped for traits that allow it, which are: Debug, Hash, Ord, PartialOrd, PartialEq, Zeroize and ZeroizeOnDrop.
#[derive_where(Debug, PartialEq; T)] struct Example<T>(#[derive_where(skip)] T); assert_eq!(format!("{:?}", Example(42)), "Example"); assert_eq!(Example(42), Example(0));
It is also possible to skip all fields in an item or variant if desired:
#[derive_where(Debug, PartialEq)] #[derive_where(skip_inner)] struct StructExample<T>(T); assert_eq!(format!("{:?}", StructExample(42)), "StructExample"); assert_eq!(StructExample(42), StructExample(0)); #[derive_where(Debug, PartialEq)] enum EnumExample<T> { #[derive_where(skip_inner)] A(T), } assert_eq!(format!("{:?}", EnumExample::A(42)), "A"); assert_eq!(EnumExample::A(42), EnumExample::A(0));
Selective skipping of fields for certain traits is also an option, both in skip and skip_inner. To prevent breaking invariants defined for these traits, some of them can only be skipped in groups. The following groups are available:
Clone: Uses Default instead of Clone.DebugEqHashOrd: Skips Eq, Hash, Ord, PartialOrd and PartialEq.HashZeroize: Skips Zeroize and ZeroizeOnDrop.#[derive_where(Debug, PartialEq)] #[derive_where(skip_inner(Debug))] struct Example<T>(i32, PhantomData<T>); assert_eq!(format!("{:?}", Example(42, PhantomData::<()>)), "Example"); assert_ne!( Example(42, PhantomData::<()>), Example(0, PhantomData::<()>) );
Similar to the skip attribute, incomparable can be used to skip variants or items in PartialEq and PartialOrd trait implementations, meaning they will always yield false for eq and None for partial_cmp. This results in all comparisons but !=, i.e. ==, <, <=, >= and >, with the marked variant or struct evaluating to false.
# use derive_where::derive_where; #[derive(Debug)] #[derive_where(PartialEq, PartialOrd)] enum EnumExample { #[derive_where(incomparable)] Incomparable, Comparable, } assert_eq!(EnumExample::Comparable, EnumExample::Comparable); assert_ne!(EnumExample::Incomparable, EnumExample::Incomparable); assert!(!(EnumExample::Comparable >= EnumExample::Incomparable)); assert!(!(EnumExample::Comparable <= EnumExample::Incomparable)); assert!(!(EnumExample::Incomparable >= EnumExample::Incomparable)); assert!(!(EnumExample::Incomparable <= EnumExample::Incomparable)); #[derive(Debug)] #[derive_where(PartialEq, PartialOrd)] #[derive_where(incomparable)] struct StructExample; assert_ne!(StructExample, StructExample); assert!(!(StructExample >= StructExample)); assert!(!(StructExample <= StructExample));
Note that it is not possible to use incomparable with Eq or Ord as that would break their invariants.
Deserialize and SerializeDeriving Deserialize and Serialize works as expected. While derive-where does not offer any attribute options, regular serde attributes can be used. Derive-where will respect #[serde(crate = "...")].
Zeroize optionsZeroize has two options:
crate: an item-level option which specifies a path to the zeroize crate in case of a re-export or rename.fqs: a field-level option which will use fully-qualified-syntax instead of calling the zeroize method on self directly. This is to avoid ambiguity between another method also called zeroize.#[derive_where(Zeroize(crate = zeroize_))] struct Example(#[derive_where(Zeroize(fqs))] i32); impl Example { // If we didn't specify the `fqs` option, this would lead to a compile // error because of method ambiguity. fn zeroize(&mut self) { self.0 = 1; } } let mut test = Example(42); // Will call the struct method. test.zeroize(); assert_eq!(test.0, 1); // WIll call the `Zeroize::zeroize` method. Zeroize::zeroize(&mut test); assert_eq!(test.0, 0);
ZeroizeOnDrop optionsIf the zeroize-on-drop feature is enabled, it implements ZeroizeOnDrop and can be implemented without Zeroize, otherwise it only implements Drop and requires Zeroize to be implemented.
ZeroizeOnDrop has two options:
crate: an item-level option which specifies a path to the zeroize crate in case of a re-export or rename.no_drop: an item-level option which will not implement Drop but instead only assert that every field implements ZeroizeOnDrop. Requires the zeroize-on-drop feature.#[derive_where(ZeroizeOnDrop(crate = zeroize_))] struct Example(i32); assert!(core::mem::needs_drop::<Example>());
The following traits can be derived with derive-where:
CloneCopyDebugDefaultDeserialize: Only available with the serde crate feature.EqHashOrdPartialEqPartialOrdSerialize: Only available with the serde crate feature.Zeroize: Only available with the zeroize crate feature.ZeroizeOnDrop: Only available with the zeroize crate feature. If the zeroize-on-drop feature is enabled, it implements ZeroizeOnDrop, otherwise it only implements Drop.Structs, tuple structs, unions and enums are supported. Derive-where tries it‘s best to discourage usage that could be covered by std’s derive. For example unit structs and enums only containing unit variants aren't supported.
Unions only support Clone and Copy.
no_std supportno_std support is provided by default.
nightly: Implements Ord and PartialOrd with the help of core::intrinsics::discriminant_value, which is what Rust does by default too. This requires a nightly version of the Rust compiler.safe: safe: Uses only safe ways to access the discriminant of the enum for Ord and PartialOrd. It also replaces all cases of core::hint::unreachable_unchecked in Ord, PartialEq and PartialOrd, which is what std uses, with unreachable.zeroize: Allows deriving Zeroize and zeroize on Drop.zeroize-on-drop: Allows deriving Zeroize and ZeroizeOnDrop and requires zeroize v1.5.The current MSRV is 1.57 and is being checked by the CI. A change will be accompanied by a minor version bump. If MSRV is important to you, use derive-where = "~1.x" to pin a specific minor version to your crate.
no_std and requires an extra #[derive(Derivative)] to use.See the CHANGELOG file for details.
Licensed under either of
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.