This crate offers procedural macros designed to facilitate the swift implementation of Rust's built-in traits.
By default, every trait this crate supports will be enabled. You can disable all of them by turning off the default features and enable only the traits that you want to use by adding them to the features explicitly.
For example,
[dependencies.educe]
version = "*"
features = ["Debug", "Clone", "Copy", "Hash", "Default"]
default-features = falseBesides the traits, there is one more feature, full, which is off by default. It widens the set of expressions that can be parsed in an attribute, and is only needed for the advanced expressions of Default; see the Default section below.
When a trait is derived with Educe and no explicit bound is set, the where predicates of the generated impl are determined automatically. Every field type that the generated code touches (ignored fields and fields handled by a custom method are excluded) is processed with the following rules, in order:
- A type that is known to implement the trait unconditionally produces no predicate at all. This covers
PhantomData, raw pointers, and function pointers for every trait, shared references forCloneandCopy, plus the types in table A. - A type that does not use any generic type or const parameter produces no predicate, because such a predicate would be constant.
- A std type that implements the trait whenever its type arguments do (table B) produces the predicates of its type arguments instead, with these rules applied recursively: a field of type
Option<T>producesT: Trait, and one of typeVec<Box<T>>produces justT: CloneforClone. - A type that mentions the derived type itself, by name or through
Self(e.g.Box<List<T>>or(Self, T)insideList<T>) producesParam: Traitbounds for the type parameters it uses, because a self-referencing predicate would overflow the trait solver (E0275). - Any other type produces the precise predicate
FieldType: Trait, so the compiler verifies the real requirement: a field of typeWrapper<T>whereWrapperhas its own conditionalCloneimpl producesWrapper<T>: Clone, which works for exactly the type arguments thatWrappersupports.
Table A — types whose type arguments never need a bound:
| Trait | Types |
|---|---|
Clone, Copy |
Arc, Rc, Weak, NonNull, Cow, Discriminant |
Debug |
Weak, NonNull, AtomicPtr, Discriminant |
PartialEq, Eq, Hash |
NonNull, Discriminant |
PartialOrd, Ord |
NonNull |
Default |
Option, Vec, VecDeque, LinkedList, HashMap, HashSet, BTreeMap, BTreeSet, Weak |
A few entries hold because the type never implements the trait at all rather than because it always does: Arc, Rc, Weak, and Cow are never Copy, and raw pointers and function pointers never implement Default. Leaving out the predicate there makes the compiler reject the derive directly, instead of accepting an implementation whose bound could never be satisfied.
Table B — types that forward the trait to their type arguments:
| Trait | Types |
|---|---|
Clone |
Option, Result, Box, Vec, VecDeque, LinkedList, BTreeMap, BTreeSet, BinaryHeap, HashMap, HashSet, RefCell, Wrapping, Reverse, Saturating |
Copy |
Option, Result, Wrapping, Reverse, Saturating |
Debug |
Option, Result, Box, Vec, VecDeque, LinkedList, BTreeMap, BTreeSet, BinaryHeap, HashMap, HashSet, Arc, Rc, RefCell, Mutex, RwLock, Wrapping, Reverse, Saturating |
PartialEq, Eq, PartialOrd, Ord |
Option, Result, Box, Vec, VecDeque, LinkedList, BTreeMap, BTreeSet, Arc, Rc, RefCell, Wrapping, Reverse, Saturating |
Hash |
Option, Result, Box, Vec, VecDeque, LinkedList, BTreeMap, BTreeSet, Arc, Rc, Wrapping, Reverse, Saturating |
Default |
Box, Arc, Rc, Cell, RefCell, Mutex, RwLock, Wrapping, Reverse, Saturating |
HashMap and HashSet with an explicit hasher type use a whole-type predicate instead of either table. This preserves the hasher requirements of each trait: Default needs a default hasher, while Debug does not require the hasher to implement Debug.
HashMap and HashSet are not in the comparison rows of table B because their comparison impls additionally require K: Eq + Hash; such fields get the precise whole-type predicate from rule 5 instead.
Both tables match type names syntactically (by the last path segment), except for declared generic type parameters and paths that start with them, such as T::PhantomData. These parameters and associated types use their own trait requirements. Other user-defined types that share a name with a std type are still treated like that std type; if the resulting bounds do not fit, set them explicitly with bound(...).
An associated type such as T::Value or <T as Family>::Value is not treated as recursive just because the source type is also named Value.
It receives its own field-type predicate; any type arguments and qualified self type are still checked for actual recursion.
When related traits are derived together with automatic bounds, a trait inherits the final predicates of its prerequisite traits: Eq and PartialOrd inherit from PartialEq, Ord inherits from Eq and PartialOrd, and Copy inherits from Clone. This way, a custom bound like #[educe(PartialEq(bound(T: MyTrait)), Eq)] automatically carries T: MyTrait into the Eq impl.
Educe cannot see the traits derived by other derive macros, including the built-in ones, so inheritance only applies between traits listed in the same #[educe(...)] attribute; a prerequisite trait implemented elsewhere contributes nothing.
When PartialEq and Eq are derived together, fields marked PartialEq(ignore) or PartialEq(method = ...) do not add automatic Eq bounds.
Explicit bounds are still used as written.
bound(where_predicates)orbound = "where_predicates"uses exactly the given predicates, without inheritance.bound(*)addsParam: Traitfor every generic type parameter, like the built-in derives.bound(false)adds no predicates at all.
An explicit bound is used verbatim; if a prerequisite impl carries predicates that the explicit bound does not imply, the compiler reports an unsatisfied supertrait and the missing predicates have to be added by hand.
Deref and DerefMut accept no bound parameter. They return a reference to a field instead of building a new value, so their impls add no predicates at all.
The Debug, PartialEq, Eq, and Hash implementations of a union read its whole storage as bytes instead of reading its fields, so their automatic bound is empty. bound(*) and custom predicates are still accepted and are used as written. Clone and Copy both add FieldType: Copy predicates, because a union can only be cloned by copying its whole storage.
A field of a #[repr(packed)] or #[repr(packed(N))] type cannot be borrowed where it lies, so the generated code copies each field it reads into a temporary first.
Every field that the generated code reads therefore has to implement Copy, and the automatic bound adds the matching FieldType: Copy predicates, just like the built-in derives do.
Any packing level is treated the same way, because a field's alignment cannot be worked out from the type syntax.
Deref and DerefMut return a reference to a field, so they can only be derived for a packed type whose target field does not need more alignment than the packing allows.
- Mutually recursive generic types (an
A<T>containingVec<B<T>>whileB<T>containsA<T>) cannot be detected from a single type definition, so automatic bounds make the trait solver overflow (E0275) on them; usebound(*)or a custom bound for such types. - The precise predicates appear in the public where clause of the impl, so private field types become visible in documentation and error messages, and changing a private field type can change the public bounds of the impl.
Custom methods accept full paths, including qualified paths such as <Type as Trait>::method::<T>.
The method = path, method(path), method = "path", and method("path") forms are supported.
In custom method paths, Self refers to the type being derived, including its generic arguments.
For Into, this also applies when the generated implementation is From for the target type; use an explicit target type path to call a target method.
Generated primitive types use ::core::primitive paths.
The Debug helper types and Hash method type parameters avoid identifiers in the input, including method paths written as strings.
Use #[derive(Educe)] and #[educe(Debug)] to implement the Debug trait for a struct, enum, or union. This allows you to modify the names of your types, variants, and fields. You can also choose to ignore specific fields or set a method to replace the Debug trait. Additionally, you have the option to format a struct as a tuple and vice versa.
use educe::Educe;
#[derive(Educe)]
#[educe(Debug)]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(Debug)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The name parameter can rename a type, a variant or a field. If you set it to false, the name can be ignored or forced to show otherwise.
use educe::Educe;
#[derive(Educe)]
#[educe(Debug(name(Struct2)))]
struct Struct {
#[educe(Debug(name(f)))]
f1: u8
}
#[derive(Educe)]
#[educe(Debug(name = true))]
enum Enum {
#[educe(Debug(name = false))]
V1,
#[educe(Debug(name(V)))]
V2 {
#[educe(Debug(name(f)))]
f1: u8,
},
#[educe(Debug(name = false))]
V3(u8),
}The ignore parameter can ignore a specific field.
use educe::Educe;
#[derive(Educe)]
#[educe(Debug)]
struct Struct {
#[educe(Debug(ignore))]
f1: u8
}
#[derive(Educe)]
#[educe(Debug)]
enum Enum {
V1,
V2 {
#[educe(Debug(ignore))]
f1: u8,
},
V3(
#[educe(Debug(ignore))]
u8
),
}With the named_field parameter, structs can be formatted as tuples and tuples can be formatted as structs.
use educe::Educe;
#[derive(Educe)]
#[educe(Debug(named_field = false))]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(Debug)]
enum Enum {
V1,
#[educe(Debug(named_field = false))]
V2 {
f1: u8,
},
#[educe(Debug(named_field = true))]
V3(
u8,
#[educe(Debug(name(value)))]
i32
),
}The method parameter can be utilized to replace the implementation of the Debug trait for a field, eliminating the need to implement the Debug trait for the type of that field.
use educe::Educe;
use std::fmt::{self, Formatter};
fn fmt<T>(_s: &T, f: &mut Formatter<'_>) -> fmt::Result {
f.write_str("Hi")
}
#[derive(Educe)]
#[educe(Debug)]
enum Enum<T> {
V1,
V2 {
#[educe(Debug(method(fmt)))]
f1: u8,
},
V3(
#[educe(Debug(method(std::fmt::UpperHex::fmt)))]
u8,
#[educe(Debug(method(fmt)))]
T
),
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules.
use educe::Educe;
#[derive(Educe)]
#[educe(Debug)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
use std::fmt::{self, Formatter};
fn fmt<D>(_s: &D, f: &mut Formatter<'_>) -> fmt::Result {
f.write_str("Hi")
}
#[derive(Educe)]
#[educe(Debug(bound(T: std::fmt::Debug)))]
enum Enum<T, K> {
V1,
V2 {
#[educe(Debug(method(fmt)))]
f1: K,
},
V3(
T
),
}In the above case, T is bound to the Debug trait, but K is not.
A union is formatted as a u8 slice because its active field is not tracked at runtime. Its fields cannot be ignored, renamed, or formatted with custom methods. Debug(unsafe) requires every byte of the union to be initialized and readable during each call, including padding and bytes outside the active field. The storage must not change during a call. This condition must hold after construction, writes, moves, and copies. Reading uninitialized bytes is undefined behavior, not just a risk of exposing memory. Initializing one field, or zeroing storage before moving the value, does not by itself guarantee this condition.
use educe::Educe;
#[derive(Educe)]
#[educe(Debug(unsafe))]
union Union {
f1: u8,
f2: i32,
}Use #[derive(Educe)] and #[educe(Clone)] to implement the Clone trait for a struct, an enum, or a union. You can set a method to replace the Clone trait.
When Clone and Copy are derived together, Educe can copy the whole value if no field uses a type or const parameter and no field has a custom clone method.
For a struct or enum with any generic parameters, a nonempty custom Copy(bound(...)) also disables this shortcut.
In that case, clone and clone_from use the existing field-wise implementation, while the custom Copy conditions are not added to Clone.
For example, Copy(bound('a: 'static)) on S<'a>(&'a i32) allows Copy only for 'static, but Clone still works with shorter lifetimes.
This rule also covers const parameters that do not appear in the fields.
Field clone methods can have visible side effects, and the performance difference has not been measured.
The automatic, bound(*), bound(false), and empty custom Copy modes keep the original shortcut rules; unions keep their existing clone behavior.
use educe::Educe;
#[derive(Educe)]
#[educe(Clone)]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(Clone)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The method parameter can be utilized to replace the implementation of the Clone trait for a field, eliminating the need to implement the Clone trait for the type of that field.
use educe::Educe;
fn clone(v: &u8) -> u8 {
v + 100
}
trait A {
fn add(&self, rhs: u8) -> Self;
}
fn clone2<T: A>(v: &T) -> T {
v.add(100)
}
#[derive(Educe)]
#[educe(Clone)]
enum Enum<T: A> {
V1,
V2 {
#[educe(Clone(method(clone)))]
f1: u8,
},
V3(
#[educe(Clone(method(clone2)))]
T
),
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules.
use educe::Educe;
#[derive(Educe)]
#[educe(Clone)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
trait A {
fn add(&self, rhs: u8) -> Self;
}
fn clone<T: A>(v: &T) -> T {
v.add(100)
}
#[derive(Educe)]
#[educe(Clone(bound(T: std::clone::Clone)))]
enum Enum<T, K: A> {
V1,
V2 {
#[educe(Clone(method(clone)))]
f1: K,
},
V3(
T
),
}In the above case, T is bound to the Clone trait, but K is not.
Refer to the introduction of the #[educe(Copy)] attribute.
Use #[derive(Educe)] and #[educe(Copy)] to implement the Copy trait for a struct, an enum, or a union.
use educe::Educe;
#[derive(Educe)]
#[educe(Copy, Clone)]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(Copy, Clone)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules. With automatic bounds, the Copy impl additionally inherits the predicates of the Clone impl generated by Educe, because Copy requires Clone.
use educe::Educe;
#[derive(Educe)]
#[educe(Copy, Clone)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
trait A {
fn add(&self, rhs: u8) -> Self;
}
fn clone<T: A>(v: &T) -> T {
v.add(100)
}
#[derive(Educe)]
#[educe(Copy, Clone(bound(T: Copy, K: A + Copy)))]
enum Enum<T, K> {
V1,
V2 {
#[educe(Clone(method(clone)))]
f1: K,
},
V3(
T
),
}Note that utilizing custom cloning methods for a type that implements the Copy and Clone traits may not be entirely appropriate.
The #[educe(Copy, Clone)] attribute can be used for a union. The fields of a union cannot be cloned with other methods.
use educe::Educe;
#[derive(Educe)]
#[educe(Copy, Clone)]
union Union {
f1: u8,
}Use #[derive(Educe)] and #[educe(PartialEq)] to implement the PartialEq trait for a struct, enum, or union. You can also choose to ignore specific fields or set a method to replace the PartialEq trait.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq)]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(PartialEq)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The ignore parameter can ignore a specific field.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq)]
struct Struct {
#[educe(PartialEq(ignore))]
f1: u8
}
#[derive(Educe)]
#[educe(PartialEq)]
enum Enum {
V1,
V2 {
#[educe(PartialEq(ignore))]
f1: u8,
},
V3(
#[educe(PartialEq(ignore))]
u8
),
}The method parameter can be utilized to replace the implementation of the PartialEq trait for a field, eliminating the need to implement the PartialEq trait for the type of that field.
use educe::Educe;
fn eq(a: &u8, b: &u8) -> bool {
a + 1 == *b
}
trait A {
fn is_same(&self, other: &Self) -> bool;
}
fn eq2<T: A>(a: &T, b: &T) -> bool {
a.is_same(b)
}
#[derive(Educe)]
#[educe(PartialEq)]
enum Enum<T: A> {
V1,
V2 {
#[educe(PartialEq(method(eq)))]
f1: u8,
},
V3(
#[educe(PartialEq(method(eq2)))]
T
),
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
trait A {
fn is_same(&self, other: &Self) -> bool;
}
fn eq<T: A>(a: &T, b: &T) -> bool {
a.is_same(b)
}
#[derive(Educe)]
#[educe(PartialEq(bound(T: std::cmp::PartialEq, K: A)))]
enum Enum<T, K> {
V1,
V2 {
#[educe(PartialEq(method(eq)))]
f1: K,
},
V3(
T
),
}The #[educe(PartialEq(unsafe))] attribute compares the entire storage of two unions as bytes. Custom field methods are not supported. Every byte must be initialized and readable during each comparison, including padding and bytes outside the active field. The storage must not change during a call. This condition must hold after construction, writes, moves, and copies. Otherwise the comparison has undefined behavior. The unsafe attribute requires the user to uphold this contract; it does not make uninitialized reads valid.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq(unsafe))]
union Union {
f1: u8,
f2: i32
}Use #[derive(Educe)] and #[educe(Eq)] to implement the Eq trait for a struct, enum, or union. Eq is a marker trait, so it has no field attributes of its own; field-level equality settings such as ignore and method belong to the PartialEq attribute.
With automatic bounds, every ordinary struct or enum field must implement Eq, even when its type is concrete, such as f64.
This check uses the final impl bounds, adds no runtime calls, and does not put concrete field requirements in the public where clause.
When Educe also derives PartialEq, fields with PartialEq(ignore) or PartialEq(method = ...) are excluded from this check and from automatic Eq bounds.
The author must ensure that custom comparison methods form an equivalence relation: reflexive, symmetric, and transitive.
Educe cannot inspect an external PartialEq implementation, so all fields are treated as ordinary fields in that case.
If a manual implementation provides an equivalence relation without requiring every field to implement Eq, use Eq(bound(false)), Eq(bound(*)), or custom predicates.
These explicit modes and unions keep their existing behavior and do not perform the automatic field check; the author is responsible for the equality contract.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq, Eq)]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(PartialEq, Eq)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules. With automatic bounds, the Eq impl also inherits the predicates of the PartialEq impl generated by Educe.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq, Eq)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
trait A {
fn is_same(&self, other: &Self) -> bool;
}
fn eq<T: A>(a: &T, b: &T) -> bool {
a.is_same(b)
}
#[derive(Educe)]
#[educe(
PartialEq(bound(T: std::cmp::PartialEq, K: A)),
Eq(bound(T: std::cmp::Eq, K: A))
)]
enum Enum<T, K> {
V1,
V2 {
#[educe(PartialEq(method(eq)))]
f1: K,
},
V3(
T
),
}The #[educe(PartialEq(unsafe), Eq)] attribute compares the entire storage of two unions as bytes, without tracking their active fields. Custom field methods are not supported. Every byte must be initialized and readable during each comparison, including padding and bytes outside the active field. The storage must not change during a call. This condition must hold after construction, writes, moves, and copies; otherwise the comparison has undefined behavior.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq(unsafe), Eq)]
union Union {
f1: u8,
f2: i32
}Use #[derive(Educe)] and #[educe(PartialOrd)] to implement the PartialOrd trait for a struct or enum. You can also choose to ignore specific fields or set a method to replace the PartialOrd trait.
use educe::Educe;
#[derive(PartialEq, Educe)]
#[educe(PartialOrd)]
struct Struct {
f1: u8
}
#[derive(PartialEq, Educe)]
#[educe(PartialOrd)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The ignore parameter can ignore a specific field.
use educe::Educe;
#[derive(PartialEq, Educe)]
#[educe(PartialOrd)]
struct Struct {
#[educe(PartialOrd(ignore))]
f1: u8
}
#[derive(PartialEq, Educe)]
#[educe(PartialOrd)]
enum Enum {
V1,
V2 {
#[educe(PartialOrd(ignore))]
f1: u8,
},
V3(
#[educe(PartialOrd(ignore))]
u8
),
}The method parameter can be utilized to replace the implementation of the PartialOrd trait for a field, eliminating the need to implement the PartialOrd trait for the type of that field.
When Ord is derived together, a field without its own PartialOrd attribute follows the ignore, rank, and method settings of its Ord attribute, so partial_cmp stays consistent with cmp; the result of an Ord comparison method is wrapped in Some automatically.
use educe::Educe;
use std::cmp::Ordering;
fn partial_cmp(a: &u8, b: &u8) -> Option<Ordering> {
if a > b {
Some(Ordering::Less)
} else if a < b {
Some(Ordering::Greater)
} else {
Some(Ordering::Equal)
}
}
trait A {
fn value(&self) -> u8;
}
fn partial_cmp2<T: A>(a: &T, b: &T) -> Option<Ordering> {
partial_cmp(&a.value(), &b.value())
}
#[derive(Educe)]
#[educe(PartialEq, PartialOrd)]
enum Enum<T: A> {
V1,
V2 {
#[educe(PartialOrd(method(partial_cmp)))]
f1: u8,
},
V3(
#[educe(PartialOrd(method(partial_cmp2)))]
T
),
}Each field can add a #[educe(PartialOrd(rank = priority_value))] attribute, where priority_value is an integer value indicating its comparison precedence (lower values indicate higher priority). The default priority_value for a field depends on its ordinal position (lower towards the front) and starts with isize::MIN.
use educe::Educe;
#[derive(PartialEq, Educe)]
#[educe(PartialOrd)]
struct Struct {
#[educe(PartialOrd(rank = 1))]
f1: u8,
#[educe(PartialOrd(rank = 0))]
f2: u8,
}For variants, the discriminant can be explicitly set for comparison. Constant expressions are evaluated by Rust, and integer representations retain their full signed or unsigned range. Alignment settings do not affect the comparison order.
use educe::Educe;
#[derive(PartialEq, Educe)]
#[educe(PartialOrd)]
#[repr(u8)]
enum Enum {
Three { f1: u8 } = 3,
Two(u8) = 2,
One = 1,
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules. With automatic bounds, the PartialOrd impl also inherits the predicates of the PartialEq impl generated by Educe.
use educe::Educe;
#[derive(PartialEq, Educe)]
#[educe(PartialOrd)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
use std::cmp::Ordering;
trait A {
fn value(&self) -> u8;
}
fn partial_cmp<T: A>(a: &T, b: &T) -> Option<Ordering> {
a.value().partial_cmp(&b.value())
}
#[derive(PartialEq, Educe)]
#[educe(PartialOrd(bound(T: std::cmp::PartialOrd, K: std::cmp::PartialOrd + A)))]
enum Enum<T, K> {
V1,
V2 {
#[educe(PartialOrd(method(partial_cmp)))]
f1: K,
},
V3(
T
),
}Use #[derive(Educe)] and #[educe(Ord)] to implement the Ord trait for a struct or enum. You can also choose to ignore specific fields or set a method to replace the Ord trait.
use educe::Educe;
#[derive(PartialEq, Eq, Educe)]
#[educe(PartialOrd, Ord)]
struct Struct {
f1: u8
}
#[derive(PartialEq, Eq, Educe)]
#[educe(PartialOrd, Ord)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The ignore parameter can ignore a specific field.
use educe::Educe;
#[derive(PartialEq, Eq, Educe)]
#[educe(PartialOrd, Ord)]
struct Struct {
#[educe(Ord(ignore))]
f1: u8
}
#[derive(PartialEq, Eq, Educe)]
#[educe(PartialOrd, Ord)]
enum Enum {
V1,
V2 {
#[educe(Ord(ignore))]
f1: u8,
},
V3(
#[educe(Ord(ignore))]
u8
),
}The method parameter can be utilized to replace the implementation of the Ord trait for a field, eliminating the need to implement the Ord trait for the type of that field.
When PartialOrd is derived together, a field without its own Ord attribute follows the ignore and rank settings of its PartialOrd attribute; a PartialOrd comparison method returns an Option<Ordering> and cannot be used by cmp, so such a field is compared with the built-in comparison.
use educe::Educe;
use std::cmp::Ordering;
fn cmp(a: &u8, b: &u8) -> Ordering {
if a > b {
Ordering::Less
} else if a < b {
Ordering::Greater
} else {
Ordering::Equal
}
}
trait A {
fn value(&self) -> u8;
}
fn cmp2<T: A>(a: &T, b: &T) -> Ordering {
cmp(&a.value(), &b.value())
}
#[derive(Educe)]
#[educe(PartialEq, Eq, PartialOrd, Ord)]
enum Enum<T: A> {
V1,
V2 {
#[educe(Ord(method(cmp)))]
f1: u8,
},
V3(
#[educe(Ord(method(cmp2)))]
T
),
}Each field can add a #[educe(Ord(rank = priority_value))] attribute, where priority_value is an integer value indicating its comparison precedence (lower values indicate higher priority). The default priority_value for a field depends on its ordinal position (lower towards the front) and starts with isize::MIN.
use educe::Educe;
#[derive(PartialEq, Eq, Educe)]
#[educe(PartialOrd, Ord)]
struct Struct {
#[educe(Ord(rank = 1))]
f1: u8,
#[educe(Ord(rank = 0))]
f2: u8,
}For variants, the discriminant can be explicitly set for comparison. Constant expressions are evaluated by Rust, and integer representations retain their full signed or unsigned range. Alignment settings do not affect the comparison order.
use educe::Educe;
#[derive(PartialEq, Eq, Educe)]
#[educe(PartialOrd, Ord)]
#[repr(u8)]
enum Enum {
Three { f1: u8 } = 3,
Two(u8) = 2,
One = 1,
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules. With automatic bounds, the Ord impl also inherits the predicates of the Eq and PartialOrd impls generated by Educe.
use educe::Educe;
#[derive(PartialEq, Eq, Educe)]
#[educe(PartialOrd, Ord)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
use std::cmp::Ordering;
trait A {
fn value(&self) -> u8;
}
fn cmp<T: A>(a: &T, b: &T) -> Ordering {
a.value().cmp(&b.value())
}
#[derive(PartialEq, Eq, Educe)]
#[educe(
PartialOrd(bound(T: std::cmp::PartialOrd, K: std::cmp::PartialEq + A)),
Ord(bound(T: std::cmp::Ord, K: std::cmp::Ord + A))
)]
enum Enum<T, K> {
V1,
V2 {
#[educe(Ord(method(cmp)))]
f1: K,
},
V3(
T
),
}Use #[derive(Educe)] and #[educe(Hash)] to implement the Hash trait for a struct, enum, or union. You can also choose to ignore specific fields or set a method to replace the Hash trait.
use educe::Educe;
#[derive(Educe)]
#[educe(Hash)]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(Hash)]
enum Enum {
V1,
V2 {
f1: u8,
},
V3(u8),
}The ignore parameter can ignore a specific field.
use educe::Educe;
#[derive(Educe)]
#[educe(Hash)]
struct Struct {
#[educe(Hash(ignore))]
f1: u8
}
#[derive(Educe)]
#[educe(Hash)]
enum Enum {
V1,
V2 {
#[educe(Hash(ignore))]
f1: u8,
},
V3(
#[educe(Hash(ignore))]
u8
),
}The method parameter can be utilized to replace the implementation of the Hash trait for a field, eliminating the need to implement the Hash trait for the type of that field.
use educe::Educe;
use std::hash::{Hash, Hasher};
fn hash<H: Hasher>(_s: &u8, state: &mut H) {
Hash::hash(&100, state)
}
fn hash2<H: Hasher, T>(_s: &T, state: &mut H) {
Hash::hash(&100, state)
}
#[derive(Educe)]
#[educe(Hash)]
enum Enum<T> {
V1,
V2 {
#[educe(Hash(method(hash)))]
f1: u8,
},
V3(
#[educe(Hash(method(hash2)))]
T
),
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules.
use educe::Educe;
#[derive(Educe)]
#[educe(Hash)]
enum Enum<T, K> {
V1,
V2 {
f1: K,
},
V3(
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
use std::hash::{Hash, Hasher};
trait A {
fn value(&self) -> u8;
}
fn hash<H: Hasher, T: A>(s: &T, state: &mut H) {
Hash::hash(&s.value(), state)
}
#[derive(Educe)]
#[educe(Hash(bound(T: std::hash::Hash, K: A)))]
enum Enum<T, K> {
V1,
V2 {
#[educe(Hash(method(hash)))]
f1: K,
},
V3(
T
),
}The #[educe(PartialEq(unsafe), Eq, Hash(unsafe))] attribute compares and hashes the entire storage of a union as bytes, without tracking its active field. Custom field methods are not supported. Every byte must be initialized and readable during each call, including padding and bytes outside the active field. The storage must not change during a call. This condition must hold after construction, writes, moves, and copies; otherwise these operations have undefined behavior.
use educe::Educe;
#[derive(Educe)]
#[educe(PartialEq(unsafe), Eq, Hash(unsafe))]
union Union {
f1: u8,
f2: i32
}Use #[derive(Educe)] and #[educe(Default)] to implement the Default trait for a struct, enum, or union.
You can set a default expression for the entire type or for individual fields.
For an enum, mark the default variant with #[educe(Default)]; for a union, mark the field to initialize.
Fields without a custom value use their type's Default implementation.
For enums and unions, it is necessary to designate a default variant (for enums) and a default field (for unions) unless the enum has only one variant or the union has only one field.
use educe::Educe;
#[derive(Educe)]
#[educe(Default)]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(Default)]
enum Enum {
V1,
#[educe(Default)]
V2 {
f1: u8,
},
V3(u8),
}
#[derive(Educe)]
#[educe(Default)]
union Union {
f1: u8,
#[educe(Default)]
f2: f64,
}use educe::Educe;
#[derive(Educe)]
#[educe(Default(expression = Struct { f1: 1 }))]
struct Struct {
f1: u8
}
#[derive(Educe)]
#[educe(Default(expression = Enum::Struct { f1: 1 }))]
enum Enum {
Unit,
Struct {
f1: u8
},
Tuple(u8),
}
#[derive(Educe)]
#[educe(Default(expression = Union { f1: 1 }))]
union Union {
f1: u8,
f2: f64,
}You may need to activate the full feature to enable support for advanced expressions.
Note that the expression is pasted into the generated default method as it is written, so for a generic type it has to be valid for every possible instantiation; an expression producing a concrete type does not work for a generic field. A bare literal is the one exception: it is wrapped in Into::into, so a literal of another type works as long as the conversion exists.
use educe::Educe;
#[derive(Educe)]
#[educe(Default)]
struct Struct {
#[educe(Default = 1)]
f1: u8,
#[educe(Default = 11111111111111111111111111111)]
f2: i128,
#[educe(Default = 1.1)]
f3: f64,
#[educe(Default = true)]
f4: bool,
#[educe(Default = "Hi")]
f5: &'static str,
#[educe(Default = "Hello")]
f6: String,
#[educe(Default = 'M')]
f7: char,
}
#[derive(Educe)]
#[educe(Default)]
enum Enum {
Unit,
#[educe(Default)]
Tuple(
#[educe(Default(expression = 0 + 1))]
u8,
#[educe(Default(expression = -11111111111111111111111111111 * -1))]
i128,
#[educe(Default(expression = 1.0 + 0.1))]
f64,
#[educe(Default(expression = !false))]
bool,
#[educe(Default(expression = "Hi"))]
&'static str,
#[educe(Default(expression = String::from("Hello")))]
String,
#[educe(Default(expression = 'M'))]
char,
),
}
#[derive(Educe)]
#[educe(Default)]
union Union {
f1: u8,
f2: i128,
f3: f64,
f4: bool,
#[educe(Default = "Hi")]
f5: &'static str,
f6: char,
}The where predicates of the generated impl are determined from the field types automatically; see the "Trait Bounds" section above for the exact rules.
use educe::Educe;
#[derive(Educe)]
#[educe(Default)]
enum Enum<T> {
Unit,
#[educe(Default)]
Struct {
f1: T
},
Tuple(T),
}Or you can set the where predicates by yourself.
use educe::Educe;
#[derive(Educe)]
#[educe(Default(bound(T: std::default::Default)))]
enum Enum<T> {
Unit,
#[educe(Default)]
Struct {
f1: T
},
Tuple(T),
}With the #[educe(Default(new))] attribute, your type will include an additional associated function called new. This function can be utilized to invoke the default method of the Default trait.
use educe::Educe;
#[derive(Educe)]
#[educe(Default(new))]
struct Struct {
f1: u8
}Use #[derive(Educe)] and #[educe(Deref)] to implement the Deref trait for a struct or enum.
You must designate a field as the default for obtaining an immutable reference unless the number of fields is exactly one.
use educe::Educe;
#[derive(Educe)]
#[educe(Deref)]
struct Struct {
f1: u8,
#[educe(Deref)]
f2: u8,
}
#[derive(Educe)]
#[educe(Deref)]
enum Enum {
Struct {
f1: u8
},
Struct2 {
f1: u8,
#[educe(Deref)]
f2: u8,
},
Tuple(u8),
Tuple2(
u8,
#[educe(Deref)]
u8
),
}Use #[derive(Educe)] and #[educe(DerefMut)] to implement the DerefMut trait for a struct or enum.
You must designate a field as the default for obtaining a mutable reference unless the number of fields is exactly one.
use educe::Educe;
#[derive(Educe)]
#[educe(Deref, DerefMut)]
struct Struct {
f1: u8,
#[educe(Deref, DerefMut)]
f2: u8,
}
#[derive(Educe)]
#[educe(Deref, DerefMut)]
enum Enum {
Struct {
f1: u8
},
Struct2 {
f1: u8,
#[educe(Deref, DerefMut)]
f2: u8,
},
Tuple(u8),
Tuple2(
#[educe(DerefMut)]
u8,
#[educe(Deref)]
u8
),
}The mutable dereferencing fields do not need to be the same as the immutable dereferencing fields, but their types must be consistent.
Use #[derive(Educe)] and #[educe(Into(type))] to make a struct or enum convertible into another type.
Educe generates an impl From<YourType> for type, which automatically provides the corresponding Into through the standard library's blanket implementation. Use the bare into flag — #[educe(Into(type, into))] — to generate a direct impl Into<type> instead.
You need to designate a field as the default for Into<type> conversion unless the number of fields is exactly one. If you don't, Educe will automatically try to find a proper one.
Reference targets preserve explicit lifetimes, mut, and each reference layer, so Into(&'a mut T) and Into(&'a &'b T) keep those types in the generated interface.
An omitted lifetime on a target reference defaults to 'static.
Automatic field matching ignores reference lifetimes, preserves reference depth, and permits an outer mutable reference to become a shared reference when Rust accepts the coercion.
Original generic declarations and where clauses keep Self tied to the source type in a generated From impl.
Custom bound(...) predicates are used verbatim: their Self means the target type in From, and the source type in a direct Into impl.
Custom method paths always use the source type for Self.
use educe::Educe;
#[derive(Educe)]
#[educe(Into(u8), Into(u16))]
struct Struct {
f1: u8,
f2: u16,
}
#[derive(Educe)]
#[educe(Into(u8))]
enum Enum {
V1 {
f1: u8,
#[educe(Into(u8))]
f2: u8,
},
V2 (
u8
),
}A From impl also lets callers write Target::from(value), whereas a direct Into impl only supports value.into(). Use the into flag when you deliberately want the conversion to be one-directional, exposed only as value.into().
The method parameter can be utilized to replace the implementation of the Into trait for a field, eliminating the need to implement the Into trait for the type of that field.
use educe::Educe;
fn into(v: u16) -> u8 {
v as u8
}
#[derive(Educe)]
#[educe(Into(u8))]
enum Enum {
V1 {
#[educe(Into(u8, method(into)))]
f1: u16,
},
V2 (
u8
),
}A generic type parameter is automatically bound to Into<type> only when it is itself the type of a field, because a nested parameter cannot meaningfully receive an Into bound.
use educe::Educe;
#[derive(Educe)]
#[educe(Into(u8))]
enum Enum<T, K> {
V1 {
f1: K,
},
V2 (
T
),
}Or you can set the where predicates by yourself.
use educe::Educe;
fn into<T>(_v: T) -> u8 {
0
}
#[derive(Educe)]
#[educe(Into(u8, bound(K: Into<u8>)))]
enum Enum<T, K> {
V1 {
f1: K,
},
V2 (
#[educe(Into(u8, method(into)))]
T
),
}