Dev 0.7.0

.github/workflows/ci-version.yml

@@ -41,7 +41,7 @@ jobs:
     name: Test ${{ matrix.toolchain }} on ${{ matrix.os }} (${{ matrix.features }})
     runs-on: ${{ matrix.os }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           toolchain: ${{ matrix.toolchain }}
@@ -57,7 +57,7 @@ jobs:
           - macos-latest
           - windows-latest
         toolchain:
-          - "1.60"
+          - "1.89"
         features:
           -
           - --no-default-features --features Debug
@@ -79,7 +79,7 @@ jobs:
     name: Test ${{ matrix.toolchain }} on ${{ matrix.os }} (${{ matrix.features }})
     runs-on: ${{ matrix.os }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           toolchain: ${{ matrix.toolchain }}

.github/workflows/ci.yml

@@ -9,7 +9,7 @@ jobs:
   rustfmt:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           toolchain: nightly
@@ -19,7 +19,7 @@ jobs:
   clippy:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           components: clippy
@@ -57,7 +57,7 @@ jobs:
     name: Test ${{ matrix.toolchain }} on ${{ matrix.os }} (${{ matrix.features }})
     runs-on: ${{ matrix.os }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           toolchain: ${{ matrix.toolchain }}
@@ -73,7 +73,7 @@ jobs:
           - macos-latest
           - windows-latest
         toolchain:
-          - "1.60"
+          - "1.89"
         features:
           -
           - --no-default-features --features Debug
@@ -95,7 +95,7 @@ jobs:
     name: Test ${{ matrix.toolchain }} on ${{ matrix.os }} (${{ matrix.features }})
     runs-on: ${{ matrix.os }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           toolchain: ${{ matrix.toolchain }}

CHANGELOG-0.5.11-to-0.7.0.md

@@ -0,0 +1,58 @@
+# Changelog: 0.5.11 to 0.7.0
+
+This document compares `v0.5.11` with the current `0.7.0` worktree. It focuses on changes that can affect users of the crate. Internal refactoring details are not listed.
+
+## Breaking and Upgrade Notes
+
+- The minimum supported Rust version was raised from Rust 1.60 to Rust 1.89.
+- The crate now uses the Rust 2024 edition.
+- Automatic trait bounds changed a lot. The generated `where` clauses are now based on the field types that the generated code really uses. This can remove old unnecessary bounds, but it can also change the public bounds of generated impls.
+- Precise generated bounds can show private field types in docs, compiler errors, or public `where` clauses. If your type is public, check that this is acceptable for your API.
+- Use `bound(*)` when you want Educe to add `Param: Trait` for every generic type parameter, similar to the built-in derives. This option was added after 0.5.11.
+- `Eq` is now treated as a marker trait. Field-level equality settings such as `ignore` and `method` should be written on `PartialEq`, not on `Eq`.
+- `Into` derives now generate `impl From<YourType> for Target` by default. Calling `value.into()` still works through Rust's blanket impl, and `Target::from(value)` also works. Use `#[educe(Into(Target, into))]` if you need a direct one-way `Into` impl instead.
+- If you already wrote your own `From<YourType> for Target` impl, check for conflicts with the new default `Into` behavior.
+
+## Trait Bounds
+
+- Automatic bounds are more precise and more clearly documented.
+- Ignored fields and fields handled by a custom `method` do not add automatic bounds for that trait.
+- Field types without generic parameters do not add constant bounds.
+- Common standard library wrapper types are handled more carefully. For example, `Option<T>` can add `T: Trait`, and `Vec<Box<T>>` can add only the needed inner bound.
+- Types that always implement a trait, such as `PhantomData`, raw pointers, function pointers, and selected standard library types, can avoid unnecessary bounds.
+- Other field types use precise bounds such as `Wrapper<T>: Clone` instead of always falling back to `T: Clone`.
+- Self-recursive generic types are handled better. For example, a type like `List<T>` containing `Box<List<T>>` can avoid trait solver overflow by adding bounds on the type parameters.
+- Mutually recursive generic types are still a known limitation. Use `bound(*)` or a custom `bound(...)` for those cases.
+- Related traits can inherit generated predicates when they are derived together by Educe: `Copy` from `Clone`, `Eq` and `PartialOrd` from `PartialEq`, and `Ord` from `Eq` and `PartialOrd`.
+- Explicit `bound(...)` values are used as written. If an explicit bound does not satisfy a required supertrait, the compiler will ask for the missing bound.
+- Older cases where bounds could leak from one generated impl into another were fixed.
+
+## Trait Behavior
+
+- `Debug` with explicit field methods was fixed for generic types and now avoids more name clashes with user code.
+- `Debug` output handles field names as strings. Tuple field names and custom names are printed more correctly, including tuple indexes such as `0`.
+- `Copy`, `Clone`, `Eq`, `PartialOrd`, and `Ord` derive behavior is more consistent with their real Rust trait relationships.
+- `Copy` can work with an Educe-generated or built-in `Clone` impl, while still respecting the bounds that `Copy` really needs.
+- `PartialOrd` and `Ord` work together more consistently. When both are derived, `PartialOrd` can follow `Ord` field settings such as `ignore`, `rank`, and `method` so that `partial_cmp` and `cmp` stay consistent.
+- When both comparison traits are derived, `Ord` can follow `PartialOrd` field settings for `ignore` and `rank`. A `PartialOrd` method cannot be reused for `Ord` because it returns `Option<Ordering>`.
+- `Ord` can be built when the `PartialOrd` feature is not enabled.
+- `Into` supports the new default `From` generation for concrete target types and for generic target types whose type parameters are covered by the target type.
+
+## Generated Code and Lints
+
+- Generated impls are marked with `#[automatically_derived]`.
+- Lint-level attributes from the input type, such as `allow`, `expect`, `warn`, and `deny`, are copied to generated impls. This makes generated code behave more like built-in derives under strict lint settings.
+- Diagnostics for invalid attributes and helper parsing were improved.
+
+## Documentation and Safety Notes
+
+- The README now has a dedicated Trait Bounds section that explains the automatic bound rules, bound inheritance, `bound(*)`, and known limitations.
+- Documentation for unsafe union derives is clearer. It now states that `Debug`, `PartialEq`, and `Hash` union impls may read, compare, or hash the whole union memory, including padding bytes.
+- The trait documentation was updated so each trait points to the shared bound rules instead of repeating older simplified rules.
+
+## Dependency and Package Changes
+
+- `quote` now has a minimum version of `1.0.44`.
+- `proc-macro2` now has a minimum version of `1.0.80`.
+- `enum-ordinalize` was updated from `4.2` to `4.4`.
+- The `rustversion` dev-dependency was removed.

CHANGELOG-0.6.0-to-0.7.0.md

@@ -0,0 +1,54 @@
+# Changelog: 0.6.0 to 0.7.0
+
+This document compares `v0.6.0` with the current `0.7.0` worktree. It focuses on changes that can affect users of the crate. Internal refactoring details are not listed.
+
+## Breaking and Upgrade Notes
+
+- The minimum supported Rust version was raised from Rust 1.60 to Rust 1.89.
+- The crate now uses the Rust 2024 edition.
+- Automatic trait bounds were redesigned again. The generated `where` clauses may differ from 0.6.0, especially for wrapper types, recursive types, and traits with prerequisite traits.
+- Precise generated bounds can show private field types in docs, compiler errors, or public `where` clauses. Check public types whose generated impl bounds are part of your API.
+- `Eq` is now treated as a marker trait. Field-level equality settings such as `ignore` and `method` should be written on `PartialEq`, not on `Eq`.
+- `Into` derives now generate `impl From<YourType> for Target` by default. Calling `value.into()` still works through Rust's blanket impl, and `Target::from(value)` also works. Use `#[educe(Into(Target, into))]` if you need a direct one-way `Into` impl instead.
+- If you already wrote your own `From<YourType> for Target` impl, check for conflicts with the new default `Into` behavior.
+
+## Trait Bounds
+
+- Automatic bounds are now documented with one shared rule set.
+- Ignored fields and fields handled by a custom `method` do not add automatic bounds for that trait.
+- Field types without generic parameters do not add constant bounds.
+- Common standard library wrapper types are handled more carefully. For example, `Option<T>` can add `T: Trait`, and `Vec<Box<T>>` can add only the needed inner bound.
+- Types that always implement a trait, such as `PhantomData`, raw pointers, function pointers, and selected standard library types, can avoid unnecessary bounds.
+- Other field types use precise bounds such as `Wrapper<T>: Clone` instead of always falling back to `T: Clone`.
+- Self-recursive generic types are handled better. For example, a type like `List<T>` containing `Box<List<T>>` can avoid trait solver overflow by adding bounds on the type parameters.
+- Mutually recursive generic types are still a known limitation. Use `bound(*)` or a custom `bound(...)` for those cases.
+- Related traits can inherit generated predicates when they are derived together by Educe: `Copy` from `Clone`, `Eq` and `PartialOrd` from `PartialEq`, and `Ord` from `Eq` and `PartialOrd`.
+- Explicit `bound(...)` values are used as written. If an explicit bound does not satisfy a required supertrait, the compiler will ask for the missing bound.
+
+## Trait Behavior
+
+- `Debug` output handles field names as strings. Tuple field names and custom names are printed more correctly, including tuple indexes such as `0`.
+- `Copy`, `Clone`, `Eq`, `PartialOrd`, and `Ord` derive behavior is more consistent with their real Rust trait relationships.
+- `Copy` can work with an Educe-generated or built-in `Clone` impl, while still respecting the bounds that `Copy` really needs.
+- `PartialOrd` and `Ord` work together more consistently. When both are derived, `PartialOrd` can follow `Ord` field settings such as `ignore`, `rank`, and `method` so that `partial_cmp` and `cmp` stay consistent.
+- When both comparison traits are derived, `Ord` can follow `PartialOrd` field settings for `ignore` and `rank`. A `PartialOrd` method cannot be reused for `Ord` because it returns `Option<Ordering>`.
+- `Into` supports the new default `From` generation for concrete target types and for generic target types whose type parameters are covered by the target type.
+
+## Generated Code and Lints
+
+- Generated impls are marked with `#[automatically_derived]`.
+- Lint-level attributes from the input type, such as `allow`, `expect`, `warn`, and `deny`, are copied to generated impls. This makes generated code behave more like built-in derives under strict lint settings.
+- Diagnostics for invalid attributes and helper parsing were improved.
+
+## Documentation and Safety Notes
+
+- The README now has a dedicated Trait Bounds section that explains the automatic bound rules, bound inheritance, `bound(*)`, and known limitations.
+- Documentation for unsafe union derives is clearer. It now states that `Debug`, `PartialEq`, and `Hash` union impls may read, compare, or hash the whole union memory, including padding bytes.
+- The trait documentation was updated so each trait points to the shared bound rules instead of repeating older simplified rules.
+
+## Dependency and Package Changes
+
+- `quote` now has a minimum version of `1.0.44`.
+- `proc-macro2` now has a minimum version of `1.0.80`.
+- `enum-ordinalize` was updated from `4.2` to `4.4`.
+- The `rustversion` dev-dependency was removed.

Cargo.toml

@@ -1,9 +1,9 @@
 [package]
 name = "educe"
-version = "0.6.0"
+version = "0.7.0"
 authors = ["Magic Len <len@magiclen.org>"]
-edition = "2021"
-rust-version = "1.60"
+edition = "2024"
+rust-version = "1.89"
 repository = "https://github.com/magiclen/educe"
 homepage = "https://magiclen.org/educe"
 keywords = ["derive", "macro", "trait", "field", "procedural"]
@@ -17,15 +17,14 @@ proc-macro = true
 
 [dependencies]
 syn = "2"
-quote = "1"
-proc-macro2 = "1"
+quote = "1.0.44"
+proc-macro2 = "1.0.80"
 
-enum-ordinalize = { version = "4.2", default-features = false, features = ["derive"] }
+enum-ordinalize = { version = "4.4", default-features = false, features = ["derive"] }
 
 [dev-dependencies]
 syn = { version = "2", features = ["full"] }
 assert-eq-float = "0.1"
-rustversion = "1"
 
 [features]
 default = ["Debug", "Clone", "Copy", "PartialEq", "Eq", "PartialOrd", "Ord", "Hash", "Default", "Deref", "DerefMut", "Into"]