bevy/crates/bevy_reflect/src/tuple_struct.rs

441 lines
12 KiB
Rust
Raw Normal View History

bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
use crate::utility::NonGenericTypeInfoCell;
use crate::{
DynamicInfo, Reflect, ReflectMut, ReflectOwned, ReflectRef, TypeInfo, Typed, UnnamedField,
};
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
use std::any::{Any, TypeId};
bevy_reflect: Improve debug formatting for reflected types (#4218) # Objective Debugging reflected types can be somewhat frustrating since all `dyn Reflect` trait objects return something like `Reflect(core::option::Option<alloc::string::String>)`. It would be much nicer to be able to see the actual value— or even use a custom `Debug` implementation. ## Solution Added `Reflect::debug` which allows users to customize the debug output. It sets defaults for all `ReflectRef` subtraits and falls back to `Reflect(type_name)` if no `Debug` implementation was registered. To register a custom `Debug` impl, users can add `#[reflect(Debug)]` like they can with other traits. ### Example Using the following structs: ```rust #[derive(Reflect)] pub struct Foo { a: usize, nested: Bar, #[reflect(ignore)] _ignored: NonReflectedValue, } #[derive(Reflect)] pub struct Bar { value: Vec2, tuple_value: (i32, String), list_value: Vec<usize>, // We can't determine debug formatting for Option<T> yet unknown_value: Option<String>, custom_debug: CustomDebug } #[derive(Reflect)] #[reflect(Debug)] struct CustomDebug; impl Debug for CustomDebug { fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result { write!(f, "This is a custom debug!") } } pub struct NonReflectedValue { _a: usize, } ``` We can do: ```rust let value = Foo { a: 1, _ignored: NonReflectedValue { _a: 10 }, nested: Bar { value: Vec2::new(1.23, 3.21), tuple_value: (123, String::from("Hello")), list_value: vec![1, 2, 3], unknown_value: Some(String::from("World")), custom_debug: CustomDebug }, }; let reflected_value: &dyn Reflect = &value; println!("{:#?}", reflected_value) ``` Which results in: ```rust Foo { a: 2, nested: Bar { value: Vec2( 1.23, 3.21, ), tuple_value: ( 123, "Hello", ), list_value: [ 1, 2, 3, ], unknown_value: Reflect(core::option::Option<alloc::string::String>), custom_debug: This is a custom debug!, }, } ``` Notice that neither `Foo` nor `Bar` implement `Debug`, yet we can still deduce it. This might be a concern if we're worried about leaking internal values. If it is, we might want to consider a way to exclude fields (possibly with a `#[reflect(hide)]` macro) or make it purely opt in (as opposed to the default implementation automatically handled by ReflectRef subtraits). Co-authored-by: Gino Valente <49806985+MrGVSV@users.noreply.github.com>
2022-05-30 16:41:31 +00:00
use std::fmt::{Debug, Formatter};
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
use std::slice::Iter;
2020-11-28 00:39:59 +00:00
/// A reflected Rust tuple struct.
///
/// Implementors of this trait allow their tuple fields to be addressed by
/// index.
///
/// This trait is automatically implemented for tuple struct types when using
/// `#[derive(Reflect)]`.
///
/// ```
/// use bevy_reflect::{Reflect, TupleStruct};
///
/// #[derive(Reflect)]
/// struct Foo(String);
///
/// # fn main() {
/// let foo = Foo("Hello, world!".to_string());
///
/// assert_eq!(foo.field_len(), 1);
///
/// let first = foo.field(0).unwrap();
/// assert_eq!(first.downcast_ref::<String>(), Some(&"Hello, world!".to_string()));
/// # }
/// ```
2020-11-28 00:39:59 +00:00
pub trait TupleStruct: Reflect {
/// Returns a reference to the value of the field with index `index` as a
/// `&dyn Reflect`.
2020-11-28 00:39:59 +00:00
fn field(&self, index: usize) -> Option<&dyn Reflect>;
/// Returns a mutable reference to the value of the field with index `index`
/// as a `&mut dyn Reflect`.
2020-11-28 00:39:59 +00:00
fn field_mut(&mut self, index: usize) -> Option<&mut dyn Reflect>;
/// Returns the number of fields in the tuple struct.
2020-11-28 00:39:59 +00:00
fn field_len(&self) -> usize;
/// Returns an iterator over the values of the tuple struct's fields.
2020-11-28 00:39:59 +00:00
fn iter_fields(&self) -> TupleStructFieldIter;
/// Clones the struct into a [`DynamicTupleStruct`].
2020-11-28 00:39:59 +00:00
fn clone_dynamic(&self) -> DynamicTupleStruct;
}
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
/// A container for compile-time tuple struct info.
#[derive(Clone, Debug)]
pub struct TupleStructInfo {
bevy_reflect: Improve serialization format even more (#5723) > Note: This is rebased off #4561 and can be viewed as a competitor to that PR. See `Comparison with #4561` section for details. # Objective The current serialization format used by `bevy_reflect` is both verbose and error-prone. Taking the following structs[^1] for example: ```rust // -- src/inventory.rs #[derive(Reflect)] struct Inventory { id: String, max_storage: usize, items: Vec<Item> } #[derive(Reflect)] struct Item { name: String } ``` Given an inventory of a single item, this would serialize to something like: ```rust // -- assets/inventory.ron { "type": "my_game::inventory::Inventory", "struct": { "id": { "type": "alloc::string::String", "value": "inv001", }, "max_storage": { "type": "usize", "value": 10 }, "items": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "my_game::inventory::Item", "struct": { "name": { "type": "alloc::string::String", "value": "Pickaxe" }, }, }, ], }, }, } ``` Aside from being really long and difficult to read, it also has a few "gotchas" that users need to be aware of if they want to edit the file manually. A major one is the requirement that you use the proper keys for a given type. For structs, you need `"struct"`. For lists, `"list"`. For tuple structs, `"tuple_struct"`. And so on. It also ***requires*** that the `"type"` entry come before the actual data. Despite being a map— which in programming is almost always orderless by default— the entries need to be in a particular order. Failure to follow the ordering convention results in a failure to deserialize the data. This makes it very prone to errors and annoyances. ## Solution Using #4042, we can remove a lot of the boilerplate and metadata needed by this older system. Since we now have static access to type information, we can simplify our serialized data to look like: ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( name: "Pickaxe" ), ], ), } ``` This is much more digestible and a lot less error-prone (no more key requirements and no more extra type names). Additionally, it is a lot more familiar to users as it follows conventional serde mechanics. For example, the struct is represented with `(...)` when serialized to RON. #### Custom Serialization Additionally, this PR adds the opt-in ability to specify a custom serde implementation to be used rather than the one created via reflection. For example[^1]: ```rust // -- src/inventory.rs #[derive(Reflect, Serialize)] #[reflect(Serialize)] struct Item { #[serde(alias = "id")] name: String } ``` ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( id: "Pickaxe" ), ], ), }, ``` By allowing users to define their own serialization methods, we do two things: 1. We give more control over how data is serialized/deserialized to the end user 2. We avoid having to re-define serde's attributes and forcing users to apply both (e.g. we don't need a `#[reflect(alias)]` attribute). ### Improved Formats One of the improvements this PR provides is the ability to represent data in ways that are more conventional and/or familiar to users. Many users are familiar with RON so here are some of the ways we can now represent data in RON: ###### Structs ```js { "my_crate::Foo": ( bar: 123 ) } // OR { "my_crate::Foo": Foo( bar: 123 ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Foo", "struct": { "bar": { "type": "usize", "value": 123 } } } ``` </details> ###### Tuples ```js { "(f32, f32)": (1.0, 2.0) } ``` <details> <summary>Old Format</summary> ```js { "type": "(f32, f32)", "tuple": [ { "type": "f32", "value": 1.0 }, { "type": "f32", "value": 2.0 } ] } ``` </details> ###### Tuple Structs ```js { "my_crate::Bar": ("Hello World!") } // OR { "my_crate::Bar": Bar("Hello World!") } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Bar", "tuple_struct": [ { "type": "alloc::string::String", "value": "Hello World!" } ] } ``` </details> ###### Arrays It may be a bit surprising to some, but arrays now also use the tuple format. This is because they essentially _are_ tuples (a sequence of values with a fixed size), but only allow for homogenous types. Additionally, this is how RON handles them and is probably a result of the 32-capacity limit imposed on them (both by [serde](https://docs.rs/serde/latest/serde/trait.Serialize.html#impl-Serialize-for-%5BT%3B%2032%5D) and by [bevy_reflect](https://docs.rs/bevy/latest/bevy/reflect/trait.GetTypeRegistration.html#impl-GetTypeRegistration-for-%5BT%3B%2032%5D)). ```js { "[i32; 3]": (1, 2, 3) } ``` <details> <summary>Old Format</summary> ```js { "type": "[i32; 3]", "array": [ { "type": "i32", "value": 1 }, { "type": "i32", "value": 2 }, { "type": "i32", "value": 3 } ] } ``` </details> ###### Enums To make things simple, I'll just put a struct variant here, but the style applies to all variant types: ```js { "my_crate::ItemType": Consumable( name: "Healing potion" ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::ItemType", "enum": { "variant": "Consumable", "struct": { "name": { "type": "alloc::string::String", "value": "Healing potion" } } } } ``` </details> ### Comparison with #4561 This PR is a rebased version of #4561. The reason for the split between the two is because this PR creates a _very_ different scene format. You may notice that the PR descriptions for either PR are pretty similar. This was done to better convey the changes depending on which (if any) gets merged first. If #4561 makes it in first, I will update this PR description accordingly. --- ## Changelog * Re-worked serialization/deserialization for reflected types * Added `TypedReflectDeserializer` for deserializing data with known `TypeInfo` * Renamed `ReflectDeserializer` to `UntypedReflectDeserializer` * ~~Replaced usages of `deserialize_any` with `deserialize_map` for non-self-describing formats~~ Reverted this change since there are still some issues that need to be sorted out (in a separate PR). By reverting this, crates like `bincode` can throw an error when attempting to deserialize non-self-describing formats (`bincode` results in `DeserializeAnyNotSupported`) * Structs, tuples, tuple structs, arrays, and enums are now all de/serialized using conventional serde methods ## Migration Guide * This PR reduces the verbosity of the scene format. Scenes will need to be updated accordingly: ```js // Old format { "type": "my_game::item::Item", "struct": { "id": { "type": "alloc::string::String", "value": "bevycraft:stone", }, "tags": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "alloc::string::String", "value": "material" }, ], }, } // New format { "my_game::item::Item": ( id: "bevycraft:stone", tags: ["material"] ) } ``` [^1]: Some derives omitted for brevity.
2022-09-20 19:38:18 +00:00
name: &'static str,
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
type_name: &'static str,
type_id: TypeId,
fields: Box<[UnnamedField]>,
#[cfg(feature = "documentation")]
docs: Option<&'static str>,
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
}
impl TupleStructInfo {
/// Create a new [`TupleStructInfo`].
///
/// # Arguments
///
bevy_reflect: Improve serialization format even more (#5723) > Note: This is rebased off #4561 and can be viewed as a competitor to that PR. See `Comparison with #4561` section for details. # Objective The current serialization format used by `bevy_reflect` is both verbose and error-prone. Taking the following structs[^1] for example: ```rust // -- src/inventory.rs #[derive(Reflect)] struct Inventory { id: String, max_storage: usize, items: Vec<Item> } #[derive(Reflect)] struct Item { name: String } ``` Given an inventory of a single item, this would serialize to something like: ```rust // -- assets/inventory.ron { "type": "my_game::inventory::Inventory", "struct": { "id": { "type": "alloc::string::String", "value": "inv001", }, "max_storage": { "type": "usize", "value": 10 }, "items": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "my_game::inventory::Item", "struct": { "name": { "type": "alloc::string::String", "value": "Pickaxe" }, }, }, ], }, }, } ``` Aside from being really long and difficult to read, it also has a few "gotchas" that users need to be aware of if they want to edit the file manually. A major one is the requirement that you use the proper keys for a given type. For structs, you need `"struct"`. For lists, `"list"`. For tuple structs, `"tuple_struct"`. And so on. It also ***requires*** that the `"type"` entry come before the actual data. Despite being a map— which in programming is almost always orderless by default— the entries need to be in a particular order. Failure to follow the ordering convention results in a failure to deserialize the data. This makes it very prone to errors and annoyances. ## Solution Using #4042, we can remove a lot of the boilerplate and metadata needed by this older system. Since we now have static access to type information, we can simplify our serialized data to look like: ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( name: "Pickaxe" ), ], ), } ``` This is much more digestible and a lot less error-prone (no more key requirements and no more extra type names). Additionally, it is a lot more familiar to users as it follows conventional serde mechanics. For example, the struct is represented with `(...)` when serialized to RON. #### Custom Serialization Additionally, this PR adds the opt-in ability to specify a custom serde implementation to be used rather than the one created via reflection. For example[^1]: ```rust // -- src/inventory.rs #[derive(Reflect, Serialize)] #[reflect(Serialize)] struct Item { #[serde(alias = "id")] name: String } ``` ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( id: "Pickaxe" ), ], ), }, ``` By allowing users to define their own serialization methods, we do two things: 1. We give more control over how data is serialized/deserialized to the end user 2. We avoid having to re-define serde's attributes and forcing users to apply both (e.g. we don't need a `#[reflect(alias)]` attribute). ### Improved Formats One of the improvements this PR provides is the ability to represent data in ways that are more conventional and/or familiar to users. Many users are familiar with RON so here are some of the ways we can now represent data in RON: ###### Structs ```js { "my_crate::Foo": ( bar: 123 ) } // OR { "my_crate::Foo": Foo( bar: 123 ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Foo", "struct": { "bar": { "type": "usize", "value": 123 } } } ``` </details> ###### Tuples ```js { "(f32, f32)": (1.0, 2.0) } ``` <details> <summary>Old Format</summary> ```js { "type": "(f32, f32)", "tuple": [ { "type": "f32", "value": 1.0 }, { "type": "f32", "value": 2.0 } ] } ``` </details> ###### Tuple Structs ```js { "my_crate::Bar": ("Hello World!") } // OR { "my_crate::Bar": Bar("Hello World!") } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Bar", "tuple_struct": [ { "type": "alloc::string::String", "value": "Hello World!" } ] } ``` </details> ###### Arrays It may be a bit surprising to some, but arrays now also use the tuple format. This is because they essentially _are_ tuples (a sequence of values with a fixed size), but only allow for homogenous types. Additionally, this is how RON handles them and is probably a result of the 32-capacity limit imposed on them (both by [serde](https://docs.rs/serde/latest/serde/trait.Serialize.html#impl-Serialize-for-%5BT%3B%2032%5D) and by [bevy_reflect](https://docs.rs/bevy/latest/bevy/reflect/trait.GetTypeRegistration.html#impl-GetTypeRegistration-for-%5BT%3B%2032%5D)). ```js { "[i32; 3]": (1, 2, 3) } ``` <details> <summary>Old Format</summary> ```js { "type": "[i32; 3]", "array": [ { "type": "i32", "value": 1 }, { "type": "i32", "value": 2 }, { "type": "i32", "value": 3 } ] } ``` </details> ###### Enums To make things simple, I'll just put a struct variant here, but the style applies to all variant types: ```js { "my_crate::ItemType": Consumable( name: "Healing potion" ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::ItemType", "enum": { "variant": "Consumable", "struct": { "name": { "type": "alloc::string::String", "value": "Healing potion" } } } } ``` </details> ### Comparison with #4561 This PR is a rebased version of #4561. The reason for the split between the two is because this PR creates a _very_ different scene format. You may notice that the PR descriptions for either PR are pretty similar. This was done to better convey the changes depending on which (if any) gets merged first. If #4561 makes it in first, I will update this PR description accordingly. --- ## Changelog * Re-worked serialization/deserialization for reflected types * Added `TypedReflectDeserializer` for deserializing data with known `TypeInfo` * Renamed `ReflectDeserializer` to `UntypedReflectDeserializer` * ~~Replaced usages of `deserialize_any` with `deserialize_map` for non-self-describing formats~~ Reverted this change since there are still some issues that need to be sorted out (in a separate PR). By reverting this, crates like `bincode` can throw an error when attempting to deserialize non-self-describing formats (`bincode` results in `DeserializeAnyNotSupported`) * Structs, tuples, tuple structs, arrays, and enums are now all de/serialized using conventional serde methods ## Migration Guide * This PR reduces the verbosity of the scene format. Scenes will need to be updated accordingly: ```js // Old format { "type": "my_game::item::Item", "struct": { "id": { "type": "alloc::string::String", "value": "bevycraft:stone", }, "tags": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "alloc::string::String", "value": "material" }, ], }, } // New format { "my_game::item::Item": ( id: "bevycraft:stone", tags: ["material"] ) } ``` [^1]: Some derives omitted for brevity.
2022-09-20 19:38:18 +00:00
/// * `name`: The name of this struct (_without_ generics or lifetimes)
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
/// * `fields`: The fields of this struct in the order they are defined
///
bevy_reflect: Improve serialization format even more (#5723) > Note: This is rebased off #4561 and can be viewed as a competitor to that PR. See `Comparison with #4561` section for details. # Objective The current serialization format used by `bevy_reflect` is both verbose and error-prone. Taking the following structs[^1] for example: ```rust // -- src/inventory.rs #[derive(Reflect)] struct Inventory { id: String, max_storage: usize, items: Vec<Item> } #[derive(Reflect)] struct Item { name: String } ``` Given an inventory of a single item, this would serialize to something like: ```rust // -- assets/inventory.ron { "type": "my_game::inventory::Inventory", "struct": { "id": { "type": "alloc::string::String", "value": "inv001", }, "max_storage": { "type": "usize", "value": 10 }, "items": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "my_game::inventory::Item", "struct": { "name": { "type": "alloc::string::String", "value": "Pickaxe" }, }, }, ], }, }, } ``` Aside from being really long and difficult to read, it also has a few "gotchas" that users need to be aware of if they want to edit the file manually. A major one is the requirement that you use the proper keys for a given type. For structs, you need `"struct"`. For lists, `"list"`. For tuple structs, `"tuple_struct"`. And so on. It also ***requires*** that the `"type"` entry come before the actual data. Despite being a map— which in programming is almost always orderless by default— the entries need to be in a particular order. Failure to follow the ordering convention results in a failure to deserialize the data. This makes it very prone to errors and annoyances. ## Solution Using #4042, we can remove a lot of the boilerplate and metadata needed by this older system. Since we now have static access to type information, we can simplify our serialized data to look like: ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( name: "Pickaxe" ), ], ), } ``` This is much more digestible and a lot less error-prone (no more key requirements and no more extra type names). Additionally, it is a lot more familiar to users as it follows conventional serde mechanics. For example, the struct is represented with `(...)` when serialized to RON. #### Custom Serialization Additionally, this PR adds the opt-in ability to specify a custom serde implementation to be used rather than the one created via reflection. For example[^1]: ```rust // -- src/inventory.rs #[derive(Reflect, Serialize)] #[reflect(Serialize)] struct Item { #[serde(alias = "id")] name: String } ``` ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( id: "Pickaxe" ), ], ), }, ``` By allowing users to define their own serialization methods, we do two things: 1. We give more control over how data is serialized/deserialized to the end user 2. We avoid having to re-define serde's attributes and forcing users to apply both (e.g. we don't need a `#[reflect(alias)]` attribute). ### Improved Formats One of the improvements this PR provides is the ability to represent data in ways that are more conventional and/or familiar to users. Many users are familiar with RON so here are some of the ways we can now represent data in RON: ###### Structs ```js { "my_crate::Foo": ( bar: 123 ) } // OR { "my_crate::Foo": Foo( bar: 123 ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Foo", "struct": { "bar": { "type": "usize", "value": 123 } } } ``` </details> ###### Tuples ```js { "(f32, f32)": (1.0, 2.0) } ``` <details> <summary>Old Format</summary> ```js { "type": "(f32, f32)", "tuple": [ { "type": "f32", "value": 1.0 }, { "type": "f32", "value": 2.0 } ] } ``` </details> ###### Tuple Structs ```js { "my_crate::Bar": ("Hello World!") } // OR { "my_crate::Bar": Bar("Hello World!") } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Bar", "tuple_struct": [ { "type": "alloc::string::String", "value": "Hello World!" } ] } ``` </details> ###### Arrays It may be a bit surprising to some, but arrays now also use the tuple format. This is because they essentially _are_ tuples (a sequence of values with a fixed size), but only allow for homogenous types. Additionally, this is how RON handles them and is probably a result of the 32-capacity limit imposed on them (both by [serde](https://docs.rs/serde/latest/serde/trait.Serialize.html#impl-Serialize-for-%5BT%3B%2032%5D) and by [bevy_reflect](https://docs.rs/bevy/latest/bevy/reflect/trait.GetTypeRegistration.html#impl-GetTypeRegistration-for-%5BT%3B%2032%5D)). ```js { "[i32; 3]": (1, 2, 3) } ``` <details> <summary>Old Format</summary> ```js { "type": "[i32; 3]", "array": [ { "type": "i32", "value": 1 }, { "type": "i32", "value": 2 }, { "type": "i32", "value": 3 } ] } ``` </details> ###### Enums To make things simple, I'll just put a struct variant here, but the style applies to all variant types: ```js { "my_crate::ItemType": Consumable( name: "Healing potion" ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::ItemType", "enum": { "variant": "Consumable", "struct": { "name": { "type": "alloc::string::String", "value": "Healing potion" } } } } ``` </details> ### Comparison with #4561 This PR is a rebased version of #4561. The reason for the split between the two is because this PR creates a _very_ different scene format. You may notice that the PR descriptions for either PR are pretty similar. This was done to better convey the changes depending on which (if any) gets merged first. If #4561 makes it in first, I will update this PR description accordingly. --- ## Changelog * Re-worked serialization/deserialization for reflected types * Added `TypedReflectDeserializer` for deserializing data with known `TypeInfo` * Renamed `ReflectDeserializer` to `UntypedReflectDeserializer` * ~~Replaced usages of `deserialize_any` with `deserialize_map` for non-self-describing formats~~ Reverted this change since there are still some issues that need to be sorted out (in a separate PR). By reverting this, crates like `bincode` can throw an error when attempting to deserialize non-self-describing formats (`bincode` results in `DeserializeAnyNotSupported`) * Structs, tuples, tuple structs, arrays, and enums are now all de/serialized using conventional serde methods ## Migration Guide * This PR reduces the verbosity of the scene format. Scenes will need to be updated accordingly: ```js // Old format { "type": "my_game::item::Item", "struct": { "id": { "type": "alloc::string::String", "value": "bevycraft:stone", }, "tags": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "alloc::string::String", "value": "material" }, ], }, } // New format { "my_game::item::Item": ( id: "bevycraft:stone", tags: ["material"] ) } ``` [^1]: Some derives omitted for brevity.
2022-09-20 19:38:18 +00:00
pub fn new<T: Reflect>(name: &'static str, fields: &[UnnamedField]) -> Self {
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
Self {
bevy_reflect: Improve serialization format even more (#5723) > Note: This is rebased off #4561 and can be viewed as a competitor to that PR. See `Comparison with #4561` section for details. # Objective The current serialization format used by `bevy_reflect` is both verbose and error-prone. Taking the following structs[^1] for example: ```rust // -- src/inventory.rs #[derive(Reflect)] struct Inventory { id: String, max_storage: usize, items: Vec<Item> } #[derive(Reflect)] struct Item { name: String } ``` Given an inventory of a single item, this would serialize to something like: ```rust // -- assets/inventory.ron { "type": "my_game::inventory::Inventory", "struct": { "id": { "type": "alloc::string::String", "value": "inv001", }, "max_storage": { "type": "usize", "value": 10 }, "items": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "my_game::inventory::Item", "struct": { "name": { "type": "alloc::string::String", "value": "Pickaxe" }, }, }, ], }, }, } ``` Aside from being really long and difficult to read, it also has a few "gotchas" that users need to be aware of if they want to edit the file manually. A major one is the requirement that you use the proper keys for a given type. For structs, you need `"struct"`. For lists, `"list"`. For tuple structs, `"tuple_struct"`. And so on. It also ***requires*** that the `"type"` entry come before the actual data. Despite being a map— which in programming is almost always orderless by default— the entries need to be in a particular order. Failure to follow the ordering convention results in a failure to deserialize the data. This makes it very prone to errors and annoyances. ## Solution Using #4042, we can remove a lot of the boilerplate and metadata needed by this older system. Since we now have static access to type information, we can simplify our serialized data to look like: ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( name: "Pickaxe" ), ], ), } ``` This is much more digestible and a lot less error-prone (no more key requirements and no more extra type names). Additionally, it is a lot more familiar to users as it follows conventional serde mechanics. For example, the struct is represented with `(...)` when serialized to RON. #### Custom Serialization Additionally, this PR adds the opt-in ability to specify a custom serde implementation to be used rather than the one created via reflection. For example[^1]: ```rust // -- src/inventory.rs #[derive(Reflect, Serialize)] #[reflect(Serialize)] struct Item { #[serde(alias = "id")] name: String } ``` ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( id: "Pickaxe" ), ], ), }, ``` By allowing users to define their own serialization methods, we do two things: 1. We give more control over how data is serialized/deserialized to the end user 2. We avoid having to re-define serde's attributes and forcing users to apply both (e.g. we don't need a `#[reflect(alias)]` attribute). ### Improved Formats One of the improvements this PR provides is the ability to represent data in ways that are more conventional and/or familiar to users. Many users are familiar with RON so here are some of the ways we can now represent data in RON: ###### Structs ```js { "my_crate::Foo": ( bar: 123 ) } // OR { "my_crate::Foo": Foo( bar: 123 ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Foo", "struct": { "bar": { "type": "usize", "value": 123 } } } ``` </details> ###### Tuples ```js { "(f32, f32)": (1.0, 2.0) } ``` <details> <summary>Old Format</summary> ```js { "type": "(f32, f32)", "tuple": [ { "type": "f32", "value": 1.0 }, { "type": "f32", "value": 2.0 } ] } ``` </details> ###### Tuple Structs ```js { "my_crate::Bar": ("Hello World!") } // OR { "my_crate::Bar": Bar("Hello World!") } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Bar", "tuple_struct": [ { "type": "alloc::string::String", "value": "Hello World!" } ] } ``` </details> ###### Arrays It may be a bit surprising to some, but arrays now also use the tuple format. This is because they essentially _are_ tuples (a sequence of values with a fixed size), but only allow for homogenous types. Additionally, this is how RON handles them and is probably a result of the 32-capacity limit imposed on them (both by [serde](https://docs.rs/serde/latest/serde/trait.Serialize.html#impl-Serialize-for-%5BT%3B%2032%5D) and by [bevy_reflect](https://docs.rs/bevy/latest/bevy/reflect/trait.GetTypeRegistration.html#impl-GetTypeRegistration-for-%5BT%3B%2032%5D)). ```js { "[i32; 3]": (1, 2, 3) } ``` <details> <summary>Old Format</summary> ```js { "type": "[i32; 3]", "array": [ { "type": "i32", "value": 1 }, { "type": "i32", "value": 2 }, { "type": "i32", "value": 3 } ] } ``` </details> ###### Enums To make things simple, I'll just put a struct variant here, but the style applies to all variant types: ```js { "my_crate::ItemType": Consumable( name: "Healing potion" ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::ItemType", "enum": { "variant": "Consumable", "struct": { "name": { "type": "alloc::string::String", "value": "Healing potion" } } } } ``` </details> ### Comparison with #4561 This PR is a rebased version of #4561. The reason for the split between the two is because this PR creates a _very_ different scene format. You may notice that the PR descriptions for either PR are pretty similar. This was done to better convey the changes depending on which (if any) gets merged first. If #4561 makes it in first, I will update this PR description accordingly. --- ## Changelog * Re-worked serialization/deserialization for reflected types * Added `TypedReflectDeserializer` for deserializing data with known `TypeInfo` * Renamed `ReflectDeserializer` to `UntypedReflectDeserializer` * ~~Replaced usages of `deserialize_any` with `deserialize_map` for non-self-describing formats~~ Reverted this change since there are still some issues that need to be sorted out (in a separate PR). By reverting this, crates like `bincode` can throw an error when attempting to deserialize non-self-describing formats (`bincode` results in `DeserializeAnyNotSupported`) * Structs, tuples, tuple structs, arrays, and enums are now all de/serialized using conventional serde methods ## Migration Guide * This PR reduces the verbosity of the scene format. Scenes will need to be updated accordingly: ```js // Old format { "type": "my_game::item::Item", "struct": { "id": { "type": "alloc::string::String", "value": "bevycraft:stone", }, "tags": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "alloc::string::String", "value": "material" }, ], }, } // New format { "my_game::item::Item": ( id: "bevycraft:stone", tags: ["material"] ) } ``` [^1]: Some derives omitted for brevity.
2022-09-20 19:38:18 +00:00
name,
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
type_name: std::any::type_name::<T>(),
type_id: TypeId::of::<T>(),
fields: fields.to_vec().into_boxed_slice(),
#[cfg(feature = "documentation")]
docs: None,
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
}
}
/// Sets the docstring for this struct.
#[cfg(feature = "documentation")]
pub fn with_docs(self, docs: Option<&'static str>) -> Self {
Self { docs, ..self }
}
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
/// Get the field at the given index.
pub fn field_at(&self, index: usize) -> Option<&UnnamedField> {
self.fields.get(index)
}
/// Iterate over the fields of this struct.
pub fn iter(&self) -> Iter<'_, UnnamedField> {
self.fields.iter()
}
/// The total number of fields in this struct.
pub fn field_len(&self) -> usize {
self.fields.len()
}
bevy_reflect: Improve serialization format even more (#5723) > Note: This is rebased off #4561 and can be viewed as a competitor to that PR. See `Comparison with #4561` section for details. # Objective The current serialization format used by `bevy_reflect` is both verbose and error-prone. Taking the following structs[^1] for example: ```rust // -- src/inventory.rs #[derive(Reflect)] struct Inventory { id: String, max_storage: usize, items: Vec<Item> } #[derive(Reflect)] struct Item { name: String } ``` Given an inventory of a single item, this would serialize to something like: ```rust // -- assets/inventory.ron { "type": "my_game::inventory::Inventory", "struct": { "id": { "type": "alloc::string::String", "value": "inv001", }, "max_storage": { "type": "usize", "value": 10 }, "items": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "my_game::inventory::Item", "struct": { "name": { "type": "alloc::string::String", "value": "Pickaxe" }, }, }, ], }, }, } ``` Aside from being really long and difficult to read, it also has a few "gotchas" that users need to be aware of if they want to edit the file manually. A major one is the requirement that you use the proper keys for a given type. For structs, you need `"struct"`. For lists, `"list"`. For tuple structs, `"tuple_struct"`. And so on. It also ***requires*** that the `"type"` entry come before the actual data. Despite being a map— which in programming is almost always orderless by default— the entries need to be in a particular order. Failure to follow the ordering convention results in a failure to deserialize the data. This makes it very prone to errors and annoyances. ## Solution Using #4042, we can remove a lot of the boilerplate and metadata needed by this older system. Since we now have static access to type information, we can simplify our serialized data to look like: ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( name: "Pickaxe" ), ], ), } ``` This is much more digestible and a lot less error-prone (no more key requirements and no more extra type names). Additionally, it is a lot more familiar to users as it follows conventional serde mechanics. For example, the struct is represented with `(...)` when serialized to RON. #### Custom Serialization Additionally, this PR adds the opt-in ability to specify a custom serde implementation to be used rather than the one created via reflection. For example[^1]: ```rust // -- src/inventory.rs #[derive(Reflect, Serialize)] #[reflect(Serialize)] struct Item { #[serde(alias = "id")] name: String } ``` ```rust // -- assets/inventory.ron { "my_game::inventory::Inventory": ( id: "inv001", max_storage: 10, items: [ ( id: "Pickaxe" ), ], ), }, ``` By allowing users to define their own serialization methods, we do two things: 1. We give more control over how data is serialized/deserialized to the end user 2. We avoid having to re-define serde's attributes and forcing users to apply both (e.g. we don't need a `#[reflect(alias)]` attribute). ### Improved Formats One of the improvements this PR provides is the ability to represent data in ways that are more conventional and/or familiar to users. Many users are familiar with RON so here are some of the ways we can now represent data in RON: ###### Structs ```js { "my_crate::Foo": ( bar: 123 ) } // OR { "my_crate::Foo": Foo( bar: 123 ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Foo", "struct": { "bar": { "type": "usize", "value": 123 } } } ``` </details> ###### Tuples ```js { "(f32, f32)": (1.0, 2.0) } ``` <details> <summary>Old Format</summary> ```js { "type": "(f32, f32)", "tuple": [ { "type": "f32", "value": 1.0 }, { "type": "f32", "value": 2.0 } ] } ``` </details> ###### Tuple Structs ```js { "my_crate::Bar": ("Hello World!") } // OR { "my_crate::Bar": Bar("Hello World!") } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::Bar", "tuple_struct": [ { "type": "alloc::string::String", "value": "Hello World!" } ] } ``` </details> ###### Arrays It may be a bit surprising to some, but arrays now also use the tuple format. This is because they essentially _are_ tuples (a sequence of values with a fixed size), but only allow for homogenous types. Additionally, this is how RON handles them and is probably a result of the 32-capacity limit imposed on them (both by [serde](https://docs.rs/serde/latest/serde/trait.Serialize.html#impl-Serialize-for-%5BT%3B%2032%5D) and by [bevy_reflect](https://docs.rs/bevy/latest/bevy/reflect/trait.GetTypeRegistration.html#impl-GetTypeRegistration-for-%5BT%3B%2032%5D)). ```js { "[i32; 3]": (1, 2, 3) } ``` <details> <summary>Old Format</summary> ```js { "type": "[i32; 3]", "array": [ { "type": "i32", "value": 1 }, { "type": "i32", "value": 2 }, { "type": "i32", "value": 3 } ] } ``` </details> ###### Enums To make things simple, I'll just put a struct variant here, but the style applies to all variant types: ```js { "my_crate::ItemType": Consumable( name: "Healing potion" ) } ``` <details> <summary>Old Format</summary> ```js { "type": "my_crate::ItemType", "enum": { "variant": "Consumable", "struct": { "name": { "type": "alloc::string::String", "value": "Healing potion" } } } } ``` </details> ### Comparison with #4561 This PR is a rebased version of #4561. The reason for the split between the two is because this PR creates a _very_ different scene format. You may notice that the PR descriptions for either PR are pretty similar. This was done to better convey the changes depending on which (if any) gets merged first. If #4561 makes it in first, I will update this PR description accordingly. --- ## Changelog * Re-worked serialization/deserialization for reflected types * Added `TypedReflectDeserializer` for deserializing data with known `TypeInfo` * Renamed `ReflectDeserializer` to `UntypedReflectDeserializer` * ~~Replaced usages of `deserialize_any` with `deserialize_map` for non-self-describing formats~~ Reverted this change since there are still some issues that need to be sorted out (in a separate PR). By reverting this, crates like `bincode` can throw an error when attempting to deserialize non-self-describing formats (`bincode` results in `DeserializeAnyNotSupported`) * Structs, tuples, tuple structs, arrays, and enums are now all de/serialized using conventional serde methods ## Migration Guide * This PR reduces the verbosity of the scene format. Scenes will need to be updated accordingly: ```js // Old format { "type": "my_game::item::Item", "struct": { "id": { "type": "alloc::string::String", "value": "bevycraft:stone", }, "tags": { "type": "alloc::vec::Vec<alloc::string::String>", "list": [ { "type": "alloc::string::String", "value": "material" }, ], }, } // New format { "my_game::item::Item": ( id: "bevycraft:stone", tags: ["material"] ) } ``` [^1]: Some derives omitted for brevity.
2022-09-20 19:38:18 +00:00
/// The name of the struct.
///
/// This does _not_ include any generics or lifetimes.
///
/// For example, `foo::bar::Baz<'a, T>` would simply be `Baz`.
pub fn name(&self) -> &'static str {
self.name
}
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
/// The [type name] of the tuple struct.
///
/// [type name]: std::any::type_name
pub fn type_name(&self) -> &'static str {
self.type_name
}
/// The [`TypeId`] of the tuple struct.
pub fn type_id(&self) -> TypeId {
self.type_id
}
/// Check if the given type matches the tuple struct type.
pub fn is<T: Any>(&self) -> bool {
TypeId::of::<T>() == self.type_id
}
/// The docstring of this struct, if any.
#[cfg(feature = "documentation")]
pub fn docs(&self) -> Option<&'static str> {
self.docs
}
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
}
/// An iterator over the field values of a tuple struct.
2020-11-28 00:39:59 +00:00
pub struct TupleStructFieldIter<'a> {
pub(crate) tuple_struct: &'a dyn TupleStruct,
pub(crate) index: usize,
}
impl<'a> TupleStructFieldIter<'a> {
pub fn new(value: &'a dyn TupleStruct) -> Self {
TupleStructFieldIter {
tuple_struct: value,
index: 0,
}
}
}
impl<'a> Iterator for TupleStructFieldIter<'a> {
type Item = &'a dyn Reflect;
fn next(&mut self) -> Option<Self::Item> {
let value = self.tuple_struct.field(self.index);
self.index += 1;
value
}
fn size_hint(&self) -> (usize, Option<usize>) {
let size = self.tuple_struct.field_len();
(size, Some(size))
}
2020-11-28 00:39:59 +00:00
}
impl<'a> ExactSizeIterator for TupleStructFieldIter<'a> {}
/// A convenience trait which combines fetching and downcasting of tuple
/// struct fields.
///
/// # Example
///
/// ```
/// use bevy_reflect::{GetTupleStructField, Reflect};
///
/// #[derive(Reflect)]
/// struct Foo(String);
///
/// # fn main() {
/// let mut foo = Foo("Hello, world!".to_string());
///
/// foo.get_field_mut::<String>(0).unwrap().truncate(5);
/// assert_eq!(foo.get_field::<String>(0), Some(&"Hello".to_string()));
/// # }
/// ```
2020-11-28 00:39:59 +00:00
pub trait GetTupleStructField {
/// Returns a reference to the value of the field with index `index`,
/// downcast to `T`.
2020-11-28 00:39:59 +00:00
fn get_field<T: Reflect>(&self, index: usize) -> Option<&T>;
/// Returns a mutable reference to the value of the field with index
/// `index`, downcast to `T`.
2020-11-28 00:39:59 +00:00
fn get_field_mut<T: Reflect>(&mut self, index: usize) -> Option<&mut T>;
}
impl<S: TupleStruct> GetTupleStructField for S {
fn get_field<T: Reflect>(&self, index: usize) -> Option<&T> {
self.field(index)
.and_then(|value| value.downcast_ref::<T>())
}
fn get_field_mut<T: Reflect>(&mut self, index: usize) -> Option<&mut T> {
self.field_mut(index)
.and_then(|value| value.downcast_mut::<T>())
}
}
impl GetTupleStructField for dyn TupleStruct {
fn get_field<T: Reflect>(&self, index: usize) -> Option<&T> {
self.field(index)
.and_then(|value| value.downcast_ref::<T>())
}
fn get_field_mut<T: Reflect>(&mut self, index: usize) -> Option<&mut T> {
self.field_mut(index)
.and_then(|value| value.downcast_mut::<T>())
}
}
/// A tuple struct which allows fields to be added at runtime.
2020-11-28 00:39:59 +00:00
#[derive(Default)]
pub struct DynamicTupleStruct {
name: String,
fields: Vec<Box<dyn Reflect>>,
}
impl DynamicTupleStruct {
/// Returns the type name of the tuple struct.
2020-11-28 00:39:59 +00:00
pub fn name(&self) -> &str {
&self.name
}
/// Sets the type name of the tuple struct.
2020-11-28 00:39:59 +00:00
pub fn set_name(&mut self, name: String) {
self.name = name;
}
/// Appends an element with value `value` to the tuple struct.
2020-11-28 00:39:59 +00:00
pub fn insert_boxed(&mut self, value: Box<dyn Reflect>) {
self.fields.push(value);
}
/// Appends a typed element with value `value` to the tuple struct.
2020-11-28 00:39:59 +00:00
pub fn insert<T: Reflect>(&mut self, value: T) {
self.insert_boxed(Box::new(value));
}
}
impl TupleStruct for DynamicTupleStruct {
#[inline]
fn field(&self, index: usize) -> Option<&dyn Reflect> {
self.fields.get(index).map(|field| &**field)
}
#[inline]
fn field_mut(&mut self, index: usize) -> Option<&mut dyn Reflect> {
self.fields.get_mut(index).map(|field| &mut **field)
}
#[inline]
fn field_len(&self) -> usize {
self.fields.len()
}
#[inline]
fn iter_fields(&self) -> TupleStructFieldIter {
TupleStructFieldIter {
tuple_struct: self,
index: 0,
}
}
fn clone_dynamic(&self) -> DynamicTupleStruct {
DynamicTupleStruct {
name: self.name.clone(),
fields: self
.fields
.iter()
.map(|value| value.clone_value())
.collect(),
}
}
}
impl Reflect for DynamicTupleStruct {
2020-11-28 00:39:59 +00:00
#[inline]
fn type_name(&self) -> &str {
self.name.as_str()
2020-11-28 00:39:59 +00:00
}
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
#[inline]
fn get_type_info(&self) -> &'static TypeInfo {
<Self as Typed>::type_info()
}
2020-11-28 00:39:59 +00:00
#[inline]
fn into_any(self: Box<Self>) -> Box<dyn Any> {
2020-11-28 00:39:59 +00:00
self
}
#[inline]
fn as_any(&self) -> &dyn Any {
self
}
#[inline]
fn as_any_mut(&mut self) -> &mut dyn Any {
2020-11-28 00:39:59 +00:00
self
}
bevy_reflect: Add `as_reflect` and `as_reflect_mut` (#4350) # Objective Trait objects that have `Reflect` as a supertrait cannot be upcast to a `dyn Reflect`. Attempting something like: ```rust trait MyTrait: Reflect { // ... } fn foo(value: &dyn MyTrait) { let reflected = value as &dyn Reflect; // Error! // ... } ``` Results in `error[E0658]: trait upcasting coercion is experimental`. The reason this is important is that a lot of `bevy_reflect` methods require a `&dyn Reflect`. This is trivial with concrete types, but if we don't know the concrete type (we only have the trait object), we can't use these methods. For example, we couldn't create a `ReflectSerializer` for the type since it expects a `&dyn Reflect` value— even though we should be able to. ## Solution Add `as_reflect` and `as_reflect_mut` to `Reflect` to allow upcasting to a `dyn Reflect`: ```rust trait MyTrait: Reflect { // ... } fn foo(value: &dyn MyTrait) { let reflected = value.as_reflect(); // ... } ``` ## Alternatives We could defer this type of logic to the crate/user. They can add these methods to their trait in the same exact way we do here. The main benefit of doing it ourselves is it makes things convenient for them (especially when using the derive macro). We could also create an `AsReflect` trait with a blanket impl over all reflected types, however, I could not get that to work for trait objects since they aren't sized. --- ## Changelog - Added trait method `Reflect::as_reflect(&self)` - Added trait method `Reflect::as_reflect_mut(&mut self)` ## Migration Guide - Manual implementors of `Reflect` will need to add implementations for the methods above (this should be pretty easy as most cases just need to return `self`)
2022-04-25 13:54:48 +00:00
#[inline]
fn as_reflect(&self) -> &dyn Reflect {
self
}
#[inline]
fn as_reflect_mut(&mut self) -> &mut dyn Reflect {
self
}
2020-11-28 00:39:59 +00:00
#[inline]
fn clone_value(&self) -> Box<dyn Reflect> {
Box::new(self.clone_dynamic())
}
#[inline]
fn reflect_ref(&self) -> ReflectRef {
ReflectRef::TupleStruct(self)
}
#[inline]
fn reflect_mut(&mut self) -> ReflectMut {
ReflectMut::TupleStruct(self)
}
#[inline]
fn reflect_owned(self: Box<Self>) -> ReflectOwned {
ReflectOwned::TupleStruct(self)
}
2020-11-28 00:39:59 +00:00
fn apply(&mut self, value: &dyn Reflect) {
if let ReflectRef::TupleStruct(tuple_struct) = value.reflect_ref() {
for (i, value) in tuple_struct.iter_fields().enumerate() {
if let Some(v) = self.field_mut(i) {
v.apply(value);
2020-11-28 00:39:59 +00:00
}
}
} else {
panic!("Attempted to apply non-TupleStruct type to TupleStruct type.");
2020-11-28 00:39:59 +00:00
}
}
fn set(&mut self, value: Box<dyn Reflect>) -> Result<(), Box<dyn Reflect>> {
*self = value.take()?;
Ok(())
}
fn reflect_partial_eq(&self, value: &dyn Reflect) -> Option<bool> {
2020-11-28 00:39:59 +00:00
tuple_struct_partial_eq(self, value)
}
bevy_reflect: Improve debug formatting for reflected types (#4218) # Objective Debugging reflected types can be somewhat frustrating since all `dyn Reflect` trait objects return something like `Reflect(core::option::Option<alloc::string::String>)`. It would be much nicer to be able to see the actual value— or even use a custom `Debug` implementation. ## Solution Added `Reflect::debug` which allows users to customize the debug output. It sets defaults for all `ReflectRef` subtraits and falls back to `Reflect(type_name)` if no `Debug` implementation was registered. To register a custom `Debug` impl, users can add `#[reflect(Debug)]` like they can with other traits. ### Example Using the following structs: ```rust #[derive(Reflect)] pub struct Foo { a: usize, nested: Bar, #[reflect(ignore)] _ignored: NonReflectedValue, } #[derive(Reflect)] pub struct Bar { value: Vec2, tuple_value: (i32, String), list_value: Vec<usize>, // We can't determine debug formatting for Option<T> yet unknown_value: Option<String>, custom_debug: CustomDebug } #[derive(Reflect)] #[reflect(Debug)] struct CustomDebug; impl Debug for CustomDebug { fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result { write!(f, "This is a custom debug!") } } pub struct NonReflectedValue { _a: usize, } ``` We can do: ```rust let value = Foo { a: 1, _ignored: NonReflectedValue { _a: 10 }, nested: Bar { value: Vec2::new(1.23, 3.21), tuple_value: (123, String::from("Hello")), list_value: vec![1, 2, 3], unknown_value: Some(String::from("World")), custom_debug: CustomDebug }, }; let reflected_value: &dyn Reflect = &value; println!("{:#?}", reflected_value) ``` Which results in: ```rust Foo { a: 2, nested: Bar { value: Vec2( 1.23, 3.21, ), tuple_value: ( 123, "Hello", ), list_value: [ 1, 2, 3, ], unknown_value: Reflect(core::option::Option<alloc::string::String>), custom_debug: This is a custom debug!, }, } ``` Notice that neither `Foo` nor `Bar` implement `Debug`, yet we can still deduce it. This might be a concern if we're worried about leaking internal values. If it is, we might want to consider a way to exclude fields (possibly with a `#[reflect(hide)]` macro) or make it purely opt in (as opposed to the default implementation automatically handled by ReflectRef subtraits). Co-authored-by: Gino Valente <49806985+MrGVSV@users.noreply.github.com>
2022-05-30 16:41:31 +00:00
fn debug(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
write!(f, "DynamicTupleStruct(")?;
tuple_struct_debug(self, f)?;
write!(f, ")")
}
}
impl Debug for DynamicTupleStruct {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
self.debug(f)
}
2020-11-28 00:39:59 +00:00
}
bevy_reflect: Add statically available type info for reflected types (#4042) # Objective > Resolves #4504 It can be helpful to have access to type information without requiring an instance of that type. Especially for `Reflect`, a lot of the gathered type information is known at compile-time and should not necessarily require an instance. ## Solution Created a dedicated `TypeInfo` enum to store static type information. All types that derive `Reflect` now also implement the newly created `Typed` trait: ```rust pub trait Typed: Reflect { fn type_info() -> &'static TypeInfo; } ``` > Note: This trait was made separate from `Reflect` due to `Sized` restrictions. If you only have access to a `dyn Reflect`, just call `.get_type_info()` on it. This new trait method on `Reflect` should return the same value as if you had called it statically. If all you have is a `TypeId` or type name, you can get the `TypeInfo` directly from the registry using the `TypeRegistry::get_type_info` method (assuming it was registered). ### Usage Below is an example of working with `TypeInfo`. As you can see, we don't have to generate an instance of `MyTupleStruct` in order to get this information. ```rust #[derive(Reflect)] struct MyTupleStruct(usize, i32, MyStruct); let info = MyTupleStruct::type_info(); if let TypeInfo::TupleStruct(info) = info { assert!(info.is::<MyTupleStruct>()); assert_eq!(std::any::type_name::<MyTupleStruct>(), info.type_name()); assert!(info.field_at(1).unwrap().is::<i32>()); } else { panic!("Expected `TypeInfo::TupleStruct`"); } ``` ### Manual Implementations It's not recommended to manually implement `Typed` yourself, but if you must, you can use the `TypeInfoCell` to automatically create and manage the static `TypeInfo`s for you (which is very helpful for blanket/generic impls): ```rust use bevy_reflect::{Reflect, TupleStructInfo, TypeInfo, UnnamedField}; use bevy_reflect::utility::TypeInfoCell; struct Foo<T: Reflect>(T); impl<T: Reflect> Typed for Foo<T> { fn type_info() -> &'static TypeInfo { static CELL: TypeInfoCell = TypeInfoCell::generic(); CELL.get_or_insert::<Self, _>(|| { let fields = [UnnamedField::new::<T>()]; let info = TupleStructInfo::new::<Self>(&fields); TypeInfo::TupleStruct(info) }) } } ``` ## Benefits One major benefit is that this opens the door to other serialization methods. Since we can get all the type info at compile time, we can know how to properly deserialize something like: ```rust #[derive(Reflect)] struct MyType { foo: usize, bar: Vec<String> } // RON to be deserialized: ( type: "my_crate::MyType", // <- We now know how to deserialize the rest of this object value: { // "foo" is a value type matching "usize" "foo": 123, // "bar" is a list type matching "Vec<String>" with item type "String" "bar": ["a", "b", "c"] } ) ``` Not only is this more compact, but it has better compatibility (we can change the type of `"foo"` to `i32` without having to update our serialized data). Of course, serialization/deserialization strategies like this may need to be discussed and fully considered before possibly making a change. However, we will be better equipped to do that now that we can access type information right from the registry. ## Discussion Some items to discuss: 1. Duplication. There's a bit of overlap with the existing traits/structs since they require an instance of the type while the type info structs do not (for example, `Struct::field_at(&self, index: usize)` and `StructInfo::field_at(&self, index: usize)`, though only `StructInfo` is accessible without an instance object). Is this okay, or do we want to handle it in another way? 2. Should `TypeInfo::Dynamic` be removed? Since the dynamic types don't have type information available at runtime, we could consider them `TypeInfo::Value`s (or just even just `TypeInfo::Struct`). The intention with `TypeInfo::Dynamic` was to keep the distinction from these dynamic types and actual structs/values since users might incorrectly believe the methods of the dynamic type's info struct would map to some contained data (which isn't possible statically). 4. General usefulness of this change, including missing/unnecessary parts. 5. Possible changes to the scene format? (One possible issue with changing it like in the example above might be that we'd have to be careful when handling generic or trait object types.) ## Compile Tests I ran a few tests to compare compile times (as suggested [here](https://github.com/bevyengine/bevy/pull/4042#discussion_r876408143)). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e7736a6f8252e10e543e52722107649d3f) and main (c309acd4322b1c3b2089e247a2d28b938eb7b56d). <details> <summary>See More</summary> The test project included 250 of the following structs (as well as a few other structs): ```rust #[derive(Default)] #[cfg_attr(feature = "reflect", derive(Reflect))] #[cfg_attr(feature = "from_reflect", derive(FromReflect))] pub struct Big001 { inventory: Inventory, foo: usize, bar: String, baz: ItemDescriptor, items: [Item; 20], hello: Option<String>, world: HashMap<i32, String>, okay: (isize, usize, /* wesize */), nope: ((String, String), (f32, f32)), blah: Cow<'static, str>, } ``` > I don't know if the compiler can optimize all these duplicate structs away, but I think it's fine either way. We're comparing times, not finding the absolute worst-case time. I only ran each build 3 times using `cargo build --timings` (thank you @devil-ira), each of which were preceeded by a `cargo clean --package bevy_reflect_compile_test`. Here are the times I got: | Test | Test 1 | Test 2 | Test 3 | Average | | -------------------------------- | ------ | ------ | ------ | ------- | | Main | 1.7s | 3.1s | 1.9s | 2.33s | | Main + `Reflect` | 8.3s | 8.6s | 8.1s | 8.33s | | Main + `Reflect` + `FromReflect` | 11.6s | 11.8s | 13.8s | 12.4s | | PR | 3.5s | 1.8s | 1.9s | 2.4s | | PR + `Reflect` | 9.2s | 8.8s | 9.3s | 9.1s | | PR + `Reflect` + `FromReflect` | 12.9s | 12.3s | 12.5s | 12.56s | </details> --- ## Future Work Even though everything could probably be made `const`, we unfortunately can't. This is because `TypeId::of::<T>()` is not yet `const` (see https://github.com/rust-lang/rust/issues/77125). When it does get stabilized, it would probably be worth coming back and making things `const`. Co-authored-by: MrGVSV <49806985+MrGVSV@users.noreply.github.com>
2022-06-09 21:18:15 +00:00
impl Typed for DynamicTupleStruct {
fn type_info() -> &'static TypeInfo {
static CELL: NonGenericTypeInfoCell = NonGenericTypeInfoCell::new();
CELL.get_or_set(|| TypeInfo::Dynamic(DynamicInfo::new::<Self>()))
}
}
/// Compares a [`TupleStruct`] with a [`Reflect`] value.
///
/// Returns true if and only if all of the following are true:
/// - `b` is a tuple struct;
/// - `b` has the same number of fields as `a`;
/// - [`Reflect::reflect_partial_eq`] returns `Some(true)` for pairwise fields of `a` and `b`.
///
/// Returns [`None`] if the comparison couldn't even be performed.
2020-11-28 00:39:59 +00:00
#[inline]
pub fn tuple_struct_partial_eq<S: TupleStruct>(a: &S, b: &dyn Reflect) -> Option<bool> {
let ReflectRef::TupleStruct(tuple_struct) = b.reflect_ref() else {
2020-11-28 00:39:59 +00:00
return Some(false);
};
if a.field_len() != tuple_struct.field_len() {
return Some(false);
}
for (i, value) in tuple_struct.iter_fields().enumerate() {
if let Some(field_value) = a.field(i) {
let eq_result = field_value.reflect_partial_eq(value);
if let failed @ (Some(false) | None) = eq_result {
return failed;
2020-11-28 00:39:59 +00:00
}
} else {
return Some(false);
}
}
Some(true)
}
bevy_reflect: Improve debug formatting for reflected types (#4218) # Objective Debugging reflected types can be somewhat frustrating since all `dyn Reflect` trait objects return something like `Reflect(core::option::Option<alloc::string::String>)`. It would be much nicer to be able to see the actual value— or even use a custom `Debug` implementation. ## Solution Added `Reflect::debug` which allows users to customize the debug output. It sets defaults for all `ReflectRef` subtraits and falls back to `Reflect(type_name)` if no `Debug` implementation was registered. To register a custom `Debug` impl, users can add `#[reflect(Debug)]` like they can with other traits. ### Example Using the following structs: ```rust #[derive(Reflect)] pub struct Foo { a: usize, nested: Bar, #[reflect(ignore)] _ignored: NonReflectedValue, } #[derive(Reflect)] pub struct Bar { value: Vec2, tuple_value: (i32, String), list_value: Vec<usize>, // We can't determine debug formatting for Option<T> yet unknown_value: Option<String>, custom_debug: CustomDebug } #[derive(Reflect)] #[reflect(Debug)] struct CustomDebug; impl Debug for CustomDebug { fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result { write!(f, "This is a custom debug!") } } pub struct NonReflectedValue { _a: usize, } ``` We can do: ```rust let value = Foo { a: 1, _ignored: NonReflectedValue { _a: 10 }, nested: Bar { value: Vec2::new(1.23, 3.21), tuple_value: (123, String::from("Hello")), list_value: vec![1, 2, 3], unknown_value: Some(String::from("World")), custom_debug: CustomDebug }, }; let reflected_value: &dyn Reflect = &value; println!("{:#?}", reflected_value) ``` Which results in: ```rust Foo { a: 2, nested: Bar { value: Vec2( 1.23, 3.21, ), tuple_value: ( 123, "Hello", ), list_value: [ 1, 2, 3, ], unknown_value: Reflect(core::option::Option<alloc::string::String>), custom_debug: This is a custom debug!, }, } ``` Notice that neither `Foo` nor `Bar` implement `Debug`, yet we can still deduce it. This might be a concern if we're worried about leaking internal values. If it is, we might want to consider a way to exclude fields (possibly with a `#[reflect(hide)]` macro) or make it purely opt in (as opposed to the default implementation automatically handled by ReflectRef subtraits). Co-authored-by: Gino Valente <49806985+MrGVSV@users.noreply.github.com>
2022-05-30 16:41:31 +00:00
/// The default debug formatter for [`TupleStruct`] types.
///
/// # Example
/// ```
/// use bevy_reflect::Reflect;
/// #[derive(Reflect)]
/// struct MyTupleStruct(usize);
///
/// let my_tuple_struct: &dyn Reflect = &MyTupleStruct(123);
/// println!("{:#?}", my_tuple_struct);
///
/// // Output:
///
/// // MyTupleStruct (
/// // 123,
/// // )
/// ```
#[inline]
pub fn tuple_struct_debug(
dyn_tuple_struct: &dyn TupleStruct,
f: &mut std::fmt::Formatter<'_>,
) -> std::fmt::Result {
let mut debug = f.debug_tuple(dyn_tuple_struct.type_name());
for field in dyn_tuple_struct.iter_fields() {
debug.field(&field as &dyn Debug);
}
debug.finish()
}