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: UsesDefaultinstead ofClone.DebugEqHashOrd: SkipsEq,Hash,Ord,PartialOrdandPartialEq.HashZeroize: SkipsZeroizeandZeroizeOnDrop.
#[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.
Deriving 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 has two options:
crate: an item-level option which specifies a path to thezeroizecrate in case of a re-export or rename.fqs: a field-level option which will use fully-qualified-syntax instead of calling thezeroizemethod onselfdirectly. This is to avoid ambiguity between another method also calledzeroize.
#[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);If 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 thezeroizecrate in case of a re-export or rename.no_drop: an item-level option which will not implementDropbut instead only assert that every field implementsZeroizeOnDrop. Requires thezeroize-on-dropfeature.
#[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 theserdecrate feature.EqHashOrdPartialEqPartialOrdSerialize: Only available with theserdecrate feature.Zeroize: Only available with thezeroizecrate feature.ZeroizeOnDrop: Only available with thezeroizecrate feature. If thezeroize-on-dropfeature is enabled, it implementsZeroizeOnDrop, otherwise it only implementsDrop.
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 support is provided by default.
nightly: ImplementsOrdandPartialOrdwith the help ofcore::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 forOrdandPartialOrd. It also replaces all cases ofcore::hint::unreachable_uncheckedinOrd,PartialEqandPartialOrd, which is what std uses, withunreachable.zeroize: Allows derivingZeroizeandzeroizeonDrop.zeroize-on-drop: Allows derivingZeroizeandZeroizeOnDropand requireszeroizev1.5.
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.
- derivative
is a great alternative with many options. Notably it doesn't support
no_stdand requires an extra#[derive(Derivative)]to use. - derive_bounded
is a new alternative still in development.
See the CHANGELOG file for details.
Licensed under either of
- Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
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.