bevy/crates/bevy_asset/src/loader_builders.rs

306 lines
12 KiB
Rust
Raw Normal View History

Make LoadContext use the builder pattern for loading dependent assets (#13465) # Objective - Fixes #13445. ## Solution - Removes all `load_` methods from `LoadContext`. - Introduces `fn loader()` which returns a builder. ## Testing - I've tested with `cargo test --package=bevy_asset` and run the two relevant examples (`asset_processing` & `asset_decompression`). --- ## Changelog - Replaced all `load_` methods on `LoadContext` with the new `loader()` pattern. ## Migration Guide - Several LoadContext method calls will need to be updated: - `load_context.load_with_settings(path, settings)` => `load_context.loader().with_settings(settings).load(path)` - `load_context.load_untyped(path)` => `load_context.loader().untyped().load(path)` - `load_context.load_direct(path)` => `load_context.loader().direct().load(path)` - `load_context.load_direct_untyped(path)` => `load_context.loader().direct().untyped().load(path)` - `load_context.load_direct_with_settings(path, settings)` => `load_context.loader().with_settings(settings).direct().load(path)` - `load_context.load_direct_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).load(path)` - `load_context.load_direct_with_reader_and_settings(reader, path, settings)` => `load_context.loader().with_settings(settings).direct().with_reader(reader).load(path)` - `load_context.load_direct_untyped_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).untyped().load(path)` --- CC @alice-i-cecile / @bushrat011899 Examples: ```rust load_context.loader() .with_asset_type::<A>() .with_asset_type_id(TypeId::of::<A>()) .with_settings(|mut settings| { settings.key = value; }) // Then, for a Handle<A>: .load::<A>() // Or, for a Handle<LoadedUntypedAsset>: .untyped() .load() // Or, to load an `A` directly: .direct() .load::<A>() .await // Or, to load an `ErasedLoadedAsset` directly: .direct() .untyped() .load() .await ```
2024-05-22 23:35:41 +00:00
//! Implementations of the builder-pattern used for loading dependent assets via
//! [`LoadContext::loader`].
use crate::{
io::Reader,
meta::{meta_transform_settings, AssetMetaDyn, MetaTransform, Settings},
Asset, AssetLoadError, AssetPath, ErasedAssetLoader, ErasedLoadedAsset, Handle, LoadContext,
LoadDirectError, LoadedAsset, LoadedUntypedAsset,
};
use std::any::TypeId;
use std::sync::Arc;
// Utility type for handling the sources of reader references
Optimize common usages of `AssetReader` (#14082) # Objective The `AssetReader` trait allows customizing the behavior of fetching bytes for an `AssetPath`, and expects implementors to return `dyn AsyncRead + AsyncSeek`. This gives implementors of `AssetLoader` great flexibility to tightly integrate their asset loading behavior with the asynchronous task system. However, almost all implementors of `AssetLoader` don't use the async functionality at all, and just call `AsyncReadExt::read_to_end(&mut Vec<u8>)`. This is incredibly inefficient, as this method repeatedly calls `poll_read` on the trait object, filling the vector 32 bytes at a time. At my work we have assets that are hundreds of megabytes which makes this a meaningful overhead. ## Solution Turn the `Reader` type alias into an actual trait, with a provided method `read_to_end`. This provided method should be more efficient than the existing extension method, as the compiler will know the underlying type of `Reader` when generating this function, which removes the repeated dynamic dispatches and allows the compiler to make further optimizations after inlining. Individual implementors are able to override the provided implementation -- for simple asset readers that just copy bytes from one buffer to another, this allows removing a large amount of overhead from the provided implementation. Now that `Reader` is an actual trait, I also improved the ergonomics for implementing `AssetReader`. Currently, implementors are expected to box their reader and return it as a trait object, which adds unnecessary boilerplate to implementations. This PR changes that trait method to return a pseudo trait alias, which allows implementors to return `impl Reader` instead of `Box<dyn Reader>`. Now, the boilerplate for boxing occurs in `ErasedAssetReader`. ## Testing I made identical changes to my company's fork of bevy. Our app, which makes heavy use of `read_to_end` for asset loading, still worked properly after this. I am not aware if we have a more systematic way of testing asset loading for correctness. --- ## Migration Guide The trait method `bevy_asset::io::AssetReader::read` (and `read_meta`) now return an opaque type instead of a boxed trait object. Implementors of these methods should change the type signatures appropriately ```rust impl AssetReader for MyReader { // Before async fn read<'a>(&'a self, path: &'a Path) -> Result<Box<Reader<'a>>, AssetReaderError> { let reader = // construct a reader Box::new(reader) as Box<Reader<'a>> } // After async fn read<'a>(&'a self, path: &'a Path) -> Result<impl Reader + 'a, AssetReaderError> { // create a reader } } ``` `bevy::asset::io::Reader` is now a trait, rather than a type alias for a trait object. Implementors of `AssetLoader::load` will need to adjust the method signature accordingly ```rust impl AssetLoader for MyLoader { async fn load<'a>( &'a self, // Before: reader: &'a mut bevy::asset::io::Reader, // After: reader: &'a mut dyn bevy::asset::io::Reader, _: &'a Self::Settings, load_context: &'a mut LoadContext<'_>, ) -> Result<Self::Asset, Self::Error> { } ``` Additionally, implementors of `AssetReader` that return a type implementing `futures_io::AsyncRead` and `AsyncSeek` might need to explicitly implement `bevy::asset::io::Reader` for that type. ```rust impl bevy::asset::io::Reader for MyAsyncReadAndSeek {} ```
2024-07-01 19:59:42 +00:00
enum ReaderRef<'a> {
Borrowed(&'a mut dyn Reader),
Boxed(Box<dyn Reader + 'a>),
Make LoadContext use the builder pattern for loading dependent assets (#13465) # Objective - Fixes #13445. ## Solution - Removes all `load_` methods from `LoadContext`. - Introduces `fn loader()` which returns a builder. ## Testing - I've tested with `cargo test --package=bevy_asset` and run the two relevant examples (`asset_processing` & `asset_decompression`). --- ## Changelog - Replaced all `load_` methods on `LoadContext` with the new `loader()` pattern. ## Migration Guide - Several LoadContext method calls will need to be updated: - `load_context.load_with_settings(path, settings)` => `load_context.loader().with_settings(settings).load(path)` - `load_context.load_untyped(path)` => `load_context.loader().untyped().load(path)` - `load_context.load_direct(path)` => `load_context.loader().direct().load(path)` - `load_context.load_direct_untyped(path)` => `load_context.loader().direct().untyped().load(path)` - `load_context.load_direct_with_settings(path, settings)` => `load_context.loader().with_settings(settings).direct().load(path)` - `load_context.load_direct_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).load(path)` - `load_context.load_direct_with_reader_and_settings(reader, path, settings)` => `load_context.loader().with_settings(settings).direct().with_reader(reader).load(path)` - `load_context.load_direct_untyped_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).untyped().load(path)` --- CC @alice-i-cecile / @bushrat011899 Examples: ```rust load_context.loader() .with_asset_type::<A>() .with_asset_type_id(TypeId::of::<A>()) .with_settings(|mut settings| { settings.key = value; }) // Then, for a Handle<A>: .load::<A>() // Or, for a Handle<LoadedUntypedAsset>: .untyped() .load() // Or, to load an `A` directly: .direct() .load::<A>() .await // Or, to load an `ErasedLoadedAsset` directly: .direct() .untyped() .load() .await ```
2024-05-22 23:35:41 +00:00
}
Optimize common usages of `AssetReader` (#14082) # Objective The `AssetReader` trait allows customizing the behavior of fetching bytes for an `AssetPath`, and expects implementors to return `dyn AsyncRead + AsyncSeek`. This gives implementors of `AssetLoader` great flexibility to tightly integrate their asset loading behavior with the asynchronous task system. However, almost all implementors of `AssetLoader` don't use the async functionality at all, and just call `AsyncReadExt::read_to_end(&mut Vec<u8>)`. This is incredibly inefficient, as this method repeatedly calls `poll_read` on the trait object, filling the vector 32 bytes at a time. At my work we have assets that are hundreds of megabytes which makes this a meaningful overhead. ## Solution Turn the `Reader` type alias into an actual trait, with a provided method `read_to_end`. This provided method should be more efficient than the existing extension method, as the compiler will know the underlying type of `Reader` when generating this function, which removes the repeated dynamic dispatches and allows the compiler to make further optimizations after inlining. Individual implementors are able to override the provided implementation -- for simple asset readers that just copy bytes from one buffer to another, this allows removing a large amount of overhead from the provided implementation. Now that `Reader` is an actual trait, I also improved the ergonomics for implementing `AssetReader`. Currently, implementors are expected to box their reader and return it as a trait object, which adds unnecessary boilerplate to implementations. This PR changes that trait method to return a pseudo trait alias, which allows implementors to return `impl Reader` instead of `Box<dyn Reader>`. Now, the boilerplate for boxing occurs in `ErasedAssetReader`. ## Testing I made identical changes to my company's fork of bevy. Our app, which makes heavy use of `read_to_end` for asset loading, still worked properly after this. I am not aware if we have a more systematic way of testing asset loading for correctness. --- ## Migration Guide The trait method `bevy_asset::io::AssetReader::read` (and `read_meta`) now return an opaque type instead of a boxed trait object. Implementors of these methods should change the type signatures appropriately ```rust impl AssetReader for MyReader { // Before async fn read<'a>(&'a self, path: &'a Path) -> Result<Box<Reader<'a>>, AssetReaderError> { let reader = // construct a reader Box::new(reader) as Box<Reader<'a>> } // After async fn read<'a>(&'a self, path: &'a Path) -> Result<impl Reader + 'a, AssetReaderError> { // create a reader } } ``` `bevy::asset::io::Reader` is now a trait, rather than a type alias for a trait object. Implementors of `AssetLoader::load` will need to adjust the method signature accordingly ```rust impl AssetLoader for MyLoader { async fn load<'a>( &'a self, // Before: reader: &'a mut bevy::asset::io::Reader, // After: reader: &'a mut dyn bevy::asset::io::Reader, _: &'a Self::Settings, load_context: &'a mut LoadContext<'_>, ) -> Result<Self::Asset, Self::Error> { } ``` Additionally, implementors of `AssetReader` that return a type implementing `futures_io::AsyncRead` and `AsyncSeek` might need to explicitly implement `bevy::asset::io::Reader` for that type. ```rust impl bevy::asset::io::Reader for MyAsyncReadAndSeek {} ```
2024-07-01 19:59:42 +00:00
impl ReaderRef<'_> {
pub fn as_mut(&mut self) -> &mut dyn Reader {
Make LoadContext use the builder pattern for loading dependent assets (#13465) # Objective - Fixes #13445. ## Solution - Removes all `load_` methods from `LoadContext`. - Introduces `fn loader()` which returns a builder. ## Testing - I've tested with `cargo test --package=bevy_asset` and run the two relevant examples (`asset_processing` & `asset_decompression`). --- ## Changelog - Replaced all `load_` methods on `LoadContext` with the new `loader()` pattern. ## Migration Guide - Several LoadContext method calls will need to be updated: - `load_context.load_with_settings(path, settings)` => `load_context.loader().with_settings(settings).load(path)` - `load_context.load_untyped(path)` => `load_context.loader().untyped().load(path)` - `load_context.load_direct(path)` => `load_context.loader().direct().load(path)` - `load_context.load_direct_untyped(path)` => `load_context.loader().direct().untyped().load(path)` - `load_context.load_direct_with_settings(path, settings)` => `load_context.loader().with_settings(settings).direct().load(path)` - `load_context.load_direct_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).load(path)` - `load_context.load_direct_with_reader_and_settings(reader, path, settings)` => `load_context.loader().with_settings(settings).direct().with_reader(reader).load(path)` - `load_context.load_direct_untyped_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).untyped().load(path)` --- CC @alice-i-cecile / @bushrat011899 Examples: ```rust load_context.loader() .with_asset_type::<A>() .with_asset_type_id(TypeId::of::<A>()) .with_settings(|mut settings| { settings.key = value; }) // Then, for a Handle<A>: .load::<A>() // Or, for a Handle<LoadedUntypedAsset>: .untyped() .load() // Or, to load an `A` directly: .direct() .load::<A>() .await // Or, to load an `ErasedLoadedAsset` directly: .direct() .untyped() .load() .await ```
2024-05-22 23:35:41 +00:00
match self {
Optimize common usages of `AssetReader` (#14082) # Objective The `AssetReader` trait allows customizing the behavior of fetching bytes for an `AssetPath`, and expects implementors to return `dyn AsyncRead + AsyncSeek`. This gives implementors of `AssetLoader` great flexibility to tightly integrate their asset loading behavior with the asynchronous task system. However, almost all implementors of `AssetLoader` don't use the async functionality at all, and just call `AsyncReadExt::read_to_end(&mut Vec<u8>)`. This is incredibly inefficient, as this method repeatedly calls `poll_read` on the trait object, filling the vector 32 bytes at a time. At my work we have assets that are hundreds of megabytes which makes this a meaningful overhead. ## Solution Turn the `Reader` type alias into an actual trait, with a provided method `read_to_end`. This provided method should be more efficient than the existing extension method, as the compiler will know the underlying type of `Reader` when generating this function, which removes the repeated dynamic dispatches and allows the compiler to make further optimizations after inlining. Individual implementors are able to override the provided implementation -- for simple asset readers that just copy bytes from one buffer to another, this allows removing a large amount of overhead from the provided implementation. Now that `Reader` is an actual trait, I also improved the ergonomics for implementing `AssetReader`. Currently, implementors are expected to box their reader and return it as a trait object, which adds unnecessary boilerplate to implementations. This PR changes that trait method to return a pseudo trait alias, which allows implementors to return `impl Reader` instead of `Box<dyn Reader>`. Now, the boilerplate for boxing occurs in `ErasedAssetReader`. ## Testing I made identical changes to my company's fork of bevy. Our app, which makes heavy use of `read_to_end` for asset loading, still worked properly after this. I am not aware if we have a more systematic way of testing asset loading for correctness. --- ## Migration Guide The trait method `bevy_asset::io::AssetReader::read` (and `read_meta`) now return an opaque type instead of a boxed trait object. Implementors of these methods should change the type signatures appropriately ```rust impl AssetReader for MyReader { // Before async fn read<'a>(&'a self, path: &'a Path) -> Result<Box<Reader<'a>>, AssetReaderError> { let reader = // construct a reader Box::new(reader) as Box<Reader<'a>> } // After async fn read<'a>(&'a self, path: &'a Path) -> Result<impl Reader + 'a, AssetReaderError> { // create a reader } } ``` `bevy::asset::io::Reader` is now a trait, rather than a type alias for a trait object. Implementors of `AssetLoader::load` will need to adjust the method signature accordingly ```rust impl AssetLoader for MyLoader { async fn load<'a>( &'a self, // Before: reader: &'a mut bevy::asset::io::Reader, // After: reader: &'a mut dyn bevy::asset::io::Reader, _: &'a Self::Settings, load_context: &'a mut LoadContext<'_>, ) -> Result<Self::Asset, Self::Error> { } ``` Additionally, implementors of `AssetReader` that return a type implementing `futures_io::AsyncRead` and `AsyncSeek` might need to explicitly implement `bevy::asset::io::Reader` for that type. ```rust impl bevy::asset::io::Reader for MyAsyncReadAndSeek {} ```
2024-07-01 19:59:42 +00:00
ReaderRef::Borrowed(r) => &mut **r,
ReaderRef::Boxed(b) => &mut **b,
Make LoadContext use the builder pattern for loading dependent assets (#13465) # Objective - Fixes #13445. ## Solution - Removes all `load_` methods from `LoadContext`. - Introduces `fn loader()` which returns a builder. ## Testing - I've tested with `cargo test --package=bevy_asset` and run the two relevant examples (`asset_processing` & `asset_decompression`). --- ## Changelog - Replaced all `load_` methods on `LoadContext` with the new `loader()` pattern. ## Migration Guide - Several LoadContext method calls will need to be updated: - `load_context.load_with_settings(path, settings)` => `load_context.loader().with_settings(settings).load(path)` - `load_context.load_untyped(path)` => `load_context.loader().untyped().load(path)` - `load_context.load_direct(path)` => `load_context.loader().direct().load(path)` - `load_context.load_direct_untyped(path)` => `load_context.loader().direct().untyped().load(path)` - `load_context.load_direct_with_settings(path, settings)` => `load_context.loader().with_settings(settings).direct().load(path)` - `load_context.load_direct_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).load(path)` - `load_context.load_direct_with_reader_and_settings(reader, path, settings)` => `load_context.loader().with_settings(settings).direct().with_reader(reader).load(path)` - `load_context.load_direct_untyped_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).untyped().load(path)` --- CC @alice-i-cecile / @bushrat011899 Examples: ```rust load_context.loader() .with_asset_type::<A>() .with_asset_type_id(TypeId::of::<A>()) .with_settings(|mut settings| { settings.key = value; }) // Then, for a Handle<A>: .load::<A>() // Or, for a Handle<LoadedUntypedAsset>: .untyped() .load() // Or, to load an `A` directly: .direct() .load::<A>() .await // Or, to load an `ErasedLoadedAsset` directly: .direct() .untyped() .load() .await ```
2024-05-22 23:35:41 +00:00
}
}
}
/// A builder for loading nested assets inside a `LoadContext`.
///
/// # Lifetimes
/// - `ctx`: the lifetime of the associated [`AssetServer`] reference
/// - `builder`: the lifetime of the temporary builder structs
pub struct NestedLoader<'ctx, 'builder> {
load_context: &'builder mut LoadContext<'ctx>,
meta_transform: Option<MetaTransform>,
asset_type_id: Option<TypeId>,
}
impl<'ctx, 'builder> NestedLoader<'ctx, 'builder> {
pub(crate) fn new(
load_context: &'builder mut LoadContext<'ctx>,
) -> NestedLoader<'ctx, 'builder> {
NestedLoader {
load_context,
meta_transform: None,
asset_type_id: None,
}
}
fn with_transform(
mut self,
transform: impl Fn(&mut dyn AssetMetaDyn) + Send + Sync + 'static,
) -> Self {
if let Some(prev_transform) = self.meta_transform {
self.meta_transform = Some(Box::new(move |meta| {
prev_transform(meta);
transform(meta);
}));
} else {
self.meta_transform = Some(Box::new(transform));
}
self
}
/// Configure the settings used to load the asset.
///
/// If the settings type `S` does not match the settings expected by `A`'s asset loader, an error will be printed to the log
/// and the asset load will fail.
#[must_use]
pub fn with_settings<S: Settings>(
self,
settings: impl Fn(&mut S) + Send + Sync + 'static,
) -> Self {
self.with_transform(move |meta| meta_transform_settings(meta, &settings))
}
/// Specify the output asset type.
#[must_use]
pub fn with_asset_type<A: Asset>(mut self) -> Self {
self.asset_type_id = Some(TypeId::of::<A>());
self
}
/// Specify the output asset type.
#[must_use]
pub fn with_asset_type_id(mut self, asset_type_id: TypeId) -> Self {
self.asset_type_id = Some(asset_type_id);
self
}
/// Load assets directly, rather than creating handles.
#[must_use]
pub fn direct<'c>(self) -> DirectNestedLoader<'ctx, 'builder, 'c> {
DirectNestedLoader {
base: self,
reader: None,
}
}
/// Load assets without static type information.
///
/// If you need to specify the type of asset, but cannot do it statically,
/// use `.with_asset_type_id()`.
#[must_use]
pub fn untyped(self) -> UntypedNestedLoader<'ctx, 'builder> {
UntypedNestedLoader { base: self }
}
/// Retrieves a handle for the asset at the given path and adds that path as a dependency of the asset.
/// If the current context is a normal [`AssetServer::load`], an actual asset load will be kicked off immediately, which ensures the load happens
/// as soon as possible.
/// "Normal loads" kicked from within a normal Bevy App will generally configure the context to kick off loads immediately.
/// If the current context is configured to not load dependencies automatically (ex: [`AssetProcessor`](crate::processor::AssetProcessor)),
/// a load will not be kicked off automatically. It is then the calling context's responsibility to begin a load if necessary.
pub fn load<'c, A: Asset>(self, path: impl Into<AssetPath<'c>>) -> Handle<A> {
let path = path.into().to_owned();
let handle = if self.load_context.should_load_dependencies {
self.load_context
.asset_server
.load_with_meta_transform(path, self.meta_transform, ())
Make LoadContext use the builder pattern for loading dependent assets (#13465) # Objective - Fixes #13445. ## Solution - Removes all `load_` methods from `LoadContext`. - Introduces `fn loader()` which returns a builder. ## Testing - I've tested with `cargo test --package=bevy_asset` and run the two relevant examples (`asset_processing` & `asset_decompression`). --- ## Changelog - Replaced all `load_` methods on `LoadContext` with the new `loader()` pattern. ## Migration Guide - Several LoadContext method calls will need to be updated: - `load_context.load_with_settings(path, settings)` => `load_context.loader().with_settings(settings).load(path)` - `load_context.load_untyped(path)` => `load_context.loader().untyped().load(path)` - `load_context.load_direct(path)` => `load_context.loader().direct().load(path)` - `load_context.load_direct_untyped(path)` => `load_context.loader().direct().untyped().load(path)` - `load_context.load_direct_with_settings(path, settings)` => `load_context.loader().with_settings(settings).direct().load(path)` - `load_context.load_direct_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).load(path)` - `load_context.load_direct_with_reader_and_settings(reader, path, settings)` => `load_context.loader().with_settings(settings).direct().with_reader(reader).load(path)` - `load_context.load_direct_untyped_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).untyped().load(path)` --- CC @alice-i-cecile / @bushrat011899 Examples: ```rust load_context.loader() .with_asset_type::<A>() .with_asset_type_id(TypeId::of::<A>()) .with_settings(|mut settings| { settings.key = value; }) // Then, for a Handle<A>: .load::<A>() // Or, for a Handle<LoadedUntypedAsset>: .untyped() .load() // Or, to load an `A` directly: .direct() .load::<A>() .await // Or, to load an `ErasedLoadedAsset` directly: .direct() .untyped() .load() .await ```
2024-05-22 23:35:41 +00:00
} else {
self.load_context
.asset_server
.get_or_create_path_handle(path, None)
};
self.load_context.dependencies.insert(handle.id().untyped());
handle
}
}
/// A builder for loading untyped nested assets inside a [`LoadContext`].
///
/// # Lifetimes
/// - `ctx`: the lifetime of the associated [`AssetServer`] reference
/// - `builder`: the lifetime of the temporary builder structs
pub struct UntypedNestedLoader<'ctx, 'builder> {
base: NestedLoader<'ctx, 'builder>,
}
impl<'ctx, 'builder> UntypedNestedLoader<'ctx, 'builder> {
/// Retrieves a handle for the asset at the given path and adds that path as a dependency of the asset without knowing its type.
pub fn load<'p>(self, path: impl Into<AssetPath<'p>>) -> Handle<LoadedUntypedAsset> {
let path = path.into().to_owned();
let handle = if self.base.load_context.should_load_dependencies {
self.base
.load_context
.asset_server
.load_untyped_with_meta_transform(path, self.base.meta_transform)
} else {
self.base
.load_context
.asset_server
.get_or_create_path_handle(path, self.base.meta_transform)
};
self.base
.load_context
.dependencies
.insert(handle.id().untyped());
handle
}
}
/// A builder for directly loading nested assets inside a `LoadContext`.
///
/// # Lifetimes
/// - `ctx`: the lifetime of the associated [`AssetServer`] reference
/// - `builder`: the lifetime of the temporary builder structs
/// - `reader`: the lifetime of the [`Reader`] reference used to read the asset data
pub struct DirectNestedLoader<'ctx, 'builder, 'reader> {
base: NestedLoader<'ctx, 'builder>,
Optimize common usages of `AssetReader` (#14082) # Objective The `AssetReader` trait allows customizing the behavior of fetching bytes for an `AssetPath`, and expects implementors to return `dyn AsyncRead + AsyncSeek`. This gives implementors of `AssetLoader` great flexibility to tightly integrate their asset loading behavior with the asynchronous task system. However, almost all implementors of `AssetLoader` don't use the async functionality at all, and just call `AsyncReadExt::read_to_end(&mut Vec<u8>)`. This is incredibly inefficient, as this method repeatedly calls `poll_read` on the trait object, filling the vector 32 bytes at a time. At my work we have assets that are hundreds of megabytes which makes this a meaningful overhead. ## Solution Turn the `Reader` type alias into an actual trait, with a provided method `read_to_end`. This provided method should be more efficient than the existing extension method, as the compiler will know the underlying type of `Reader` when generating this function, which removes the repeated dynamic dispatches and allows the compiler to make further optimizations after inlining. Individual implementors are able to override the provided implementation -- for simple asset readers that just copy bytes from one buffer to another, this allows removing a large amount of overhead from the provided implementation. Now that `Reader` is an actual trait, I also improved the ergonomics for implementing `AssetReader`. Currently, implementors are expected to box their reader and return it as a trait object, which adds unnecessary boilerplate to implementations. This PR changes that trait method to return a pseudo trait alias, which allows implementors to return `impl Reader` instead of `Box<dyn Reader>`. Now, the boilerplate for boxing occurs in `ErasedAssetReader`. ## Testing I made identical changes to my company's fork of bevy. Our app, which makes heavy use of `read_to_end` for asset loading, still worked properly after this. I am not aware if we have a more systematic way of testing asset loading for correctness. --- ## Migration Guide The trait method `bevy_asset::io::AssetReader::read` (and `read_meta`) now return an opaque type instead of a boxed trait object. Implementors of these methods should change the type signatures appropriately ```rust impl AssetReader for MyReader { // Before async fn read<'a>(&'a self, path: &'a Path) -> Result<Box<Reader<'a>>, AssetReaderError> { let reader = // construct a reader Box::new(reader) as Box<Reader<'a>> } // After async fn read<'a>(&'a self, path: &'a Path) -> Result<impl Reader + 'a, AssetReaderError> { // create a reader } } ``` `bevy::asset::io::Reader` is now a trait, rather than a type alias for a trait object. Implementors of `AssetLoader::load` will need to adjust the method signature accordingly ```rust impl AssetLoader for MyLoader { async fn load<'a>( &'a self, // Before: reader: &'a mut bevy::asset::io::Reader, // After: reader: &'a mut dyn bevy::asset::io::Reader, _: &'a Self::Settings, load_context: &'a mut LoadContext<'_>, ) -> Result<Self::Asset, Self::Error> { } ``` Additionally, implementors of `AssetReader` that return a type implementing `futures_io::AsyncRead` and `AsyncSeek` might need to explicitly implement `bevy::asset::io::Reader` for that type. ```rust impl bevy::asset::io::Reader for MyAsyncReadAndSeek {} ```
2024-07-01 19:59:42 +00:00
reader: Option<&'builder mut (dyn Reader + 'reader)>,
Make LoadContext use the builder pattern for loading dependent assets (#13465) # Objective - Fixes #13445. ## Solution - Removes all `load_` methods from `LoadContext`. - Introduces `fn loader()` which returns a builder. ## Testing - I've tested with `cargo test --package=bevy_asset` and run the two relevant examples (`asset_processing` & `asset_decompression`). --- ## Changelog - Replaced all `load_` methods on `LoadContext` with the new `loader()` pattern. ## Migration Guide - Several LoadContext method calls will need to be updated: - `load_context.load_with_settings(path, settings)` => `load_context.loader().with_settings(settings).load(path)` - `load_context.load_untyped(path)` => `load_context.loader().untyped().load(path)` - `load_context.load_direct(path)` => `load_context.loader().direct().load(path)` - `load_context.load_direct_untyped(path)` => `load_context.loader().direct().untyped().load(path)` - `load_context.load_direct_with_settings(path, settings)` => `load_context.loader().with_settings(settings).direct().load(path)` - `load_context.load_direct_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).load(path)` - `load_context.load_direct_with_reader_and_settings(reader, path, settings)` => `load_context.loader().with_settings(settings).direct().with_reader(reader).load(path)` - `load_context.load_direct_untyped_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).untyped().load(path)` --- CC @alice-i-cecile / @bushrat011899 Examples: ```rust load_context.loader() .with_asset_type::<A>() .with_asset_type_id(TypeId::of::<A>()) .with_settings(|mut settings| { settings.key = value; }) // Then, for a Handle<A>: .load::<A>() // Or, for a Handle<LoadedUntypedAsset>: .untyped() .load() // Or, to load an `A` directly: .direct() .load::<A>() .await // Or, to load an `ErasedLoadedAsset` directly: .direct() .untyped() .load() .await ```
2024-05-22 23:35:41 +00:00
}
impl<'ctx: 'reader, 'builder, 'reader> DirectNestedLoader<'ctx, 'builder, 'reader> {
/// Specify the reader to use to read the asset data.
#[must_use]
Optimize common usages of `AssetReader` (#14082) # Objective The `AssetReader` trait allows customizing the behavior of fetching bytes for an `AssetPath`, and expects implementors to return `dyn AsyncRead + AsyncSeek`. This gives implementors of `AssetLoader` great flexibility to tightly integrate their asset loading behavior with the asynchronous task system. However, almost all implementors of `AssetLoader` don't use the async functionality at all, and just call `AsyncReadExt::read_to_end(&mut Vec<u8>)`. This is incredibly inefficient, as this method repeatedly calls `poll_read` on the trait object, filling the vector 32 bytes at a time. At my work we have assets that are hundreds of megabytes which makes this a meaningful overhead. ## Solution Turn the `Reader` type alias into an actual trait, with a provided method `read_to_end`. This provided method should be more efficient than the existing extension method, as the compiler will know the underlying type of `Reader` when generating this function, which removes the repeated dynamic dispatches and allows the compiler to make further optimizations after inlining. Individual implementors are able to override the provided implementation -- for simple asset readers that just copy bytes from one buffer to another, this allows removing a large amount of overhead from the provided implementation. Now that `Reader` is an actual trait, I also improved the ergonomics for implementing `AssetReader`. Currently, implementors are expected to box their reader and return it as a trait object, which adds unnecessary boilerplate to implementations. This PR changes that trait method to return a pseudo trait alias, which allows implementors to return `impl Reader` instead of `Box<dyn Reader>`. Now, the boilerplate for boxing occurs in `ErasedAssetReader`. ## Testing I made identical changes to my company's fork of bevy. Our app, which makes heavy use of `read_to_end` for asset loading, still worked properly after this. I am not aware if we have a more systematic way of testing asset loading for correctness. --- ## Migration Guide The trait method `bevy_asset::io::AssetReader::read` (and `read_meta`) now return an opaque type instead of a boxed trait object. Implementors of these methods should change the type signatures appropriately ```rust impl AssetReader for MyReader { // Before async fn read<'a>(&'a self, path: &'a Path) -> Result<Box<Reader<'a>>, AssetReaderError> { let reader = // construct a reader Box::new(reader) as Box<Reader<'a>> } // After async fn read<'a>(&'a self, path: &'a Path) -> Result<impl Reader + 'a, AssetReaderError> { // create a reader } } ``` `bevy::asset::io::Reader` is now a trait, rather than a type alias for a trait object. Implementors of `AssetLoader::load` will need to adjust the method signature accordingly ```rust impl AssetLoader for MyLoader { async fn load<'a>( &'a self, // Before: reader: &'a mut bevy::asset::io::Reader, // After: reader: &'a mut dyn bevy::asset::io::Reader, _: &'a Self::Settings, load_context: &'a mut LoadContext<'_>, ) -> Result<Self::Asset, Self::Error> { } ``` Additionally, implementors of `AssetReader` that return a type implementing `futures_io::AsyncRead` and `AsyncSeek` might need to explicitly implement `bevy::asset::io::Reader` for that type. ```rust impl bevy::asset::io::Reader for MyAsyncReadAndSeek {} ```
2024-07-01 19:59:42 +00:00
pub fn with_reader(mut self, reader: &'builder mut (dyn Reader + 'reader)) -> Self {
Make LoadContext use the builder pattern for loading dependent assets (#13465) # Objective - Fixes #13445. ## Solution - Removes all `load_` methods from `LoadContext`. - Introduces `fn loader()` which returns a builder. ## Testing - I've tested with `cargo test --package=bevy_asset` and run the two relevant examples (`asset_processing` & `asset_decompression`). --- ## Changelog - Replaced all `load_` methods on `LoadContext` with the new `loader()` pattern. ## Migration Guide - Several LoadContext method calls will need to be updated: - `load_context.load_with_settings(path, settings)` => `load_context.loader().with_settings(settings).load(path)` - `load_context.load_untyped(path)` => `load_context.loader().untyped().load(path)` - `load_context.load_direct(path)` => `load_context.loader().direct().load(path)` - `load_context.load_direct_untyped(path)` => `load_context.loader().direct().untyped().load(path)` - `load_context.load_direct_with_settings(path, settings)` => `load_context.loader().with_settings(settings).direct().load(path)` - `load_context.load_direct_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).load(path)` - `load_context.load_direct_with_reader_and_settings(reader, path, settings)` => `load_context.loader().with_settings(settings).direct().with_reader(reader).load(path)` - `load_context.load_direct_untyped_with_reader(reader, path)` => `load_context.loader().direct().with_reader(reader).untyped().load(path)` --- CC @alice-i-cecile / @bushrat011899 Examples: ```rust load_context.loader() .with_asset_type::<A>() .with_asset_type_id(TypeId::of::<A>()) .with_settings(|mut settings| { settings.key = value; }) // Then, for a Handle<A>: .load::<A>() // Or, for a Handle<LoadedUntypedAsset>: .untyped() .load() // Or, to load an `A` directly: .direct() .load::<A>() .await // Or, to load an `ErasedLoadedAsset` directly: .direct() .untyped() .load() .await ```
2024-05-22 23:35:41 +00:00
self.reader = Some(reader);
self
}
/// Load the asset without providing static type information.
///
/// If you need to specify the type of asset, but cannot do it statically,
/// use `.with_asset_type_id()`.
#[must_use]
pub fn untyped(self) -> UntypedDirectNestedLoader<'ctx, 'builder, 'reader> {
UntypedDirectNestedLoader { base: self }
}
async fn load_internal(
self,
path: &AssetPath<'static>,
) -> Result<(Arc<dyn ErasedAssetLoader>, ErasedLoadedAsset), LoadDirectError> {
let (mut meta, loader, mut reader) = if let Some(reader) = self.reader {
let loader = if let Some(asset_type_id) = self.base.asset_type_id {
self.base
.load_context
.asset_server
.get_asset_loader_with_asset_type_id(asset_type_id)
.await
.map_err(|error| LoadDirectError {
dependency: path.clone(),
error: error.into(),
})?
} else {
self.base
.load_context
.asset_server
.get_path_asset_loader(path)
.await
.map_err(|error| LoadDirectError {
dependency: path.clone(),
error: error.into(),
})?
};
let meta = loader.default_meta();
(meta, loader, ReaderRef::Borrowed(reader))
} else {
let (meta, loader, reader) = self
.base
.load_context
.asset_server
.get_meta_loader_and_reader(path, self.base.asset_type_id)
.await
.map_err(|error| LoadDirectError {
dependency: path.clone(),
error,
})?;
(meta, loader, ReaderRef::Boxed(reader))
};
if let Some(meta_transform) = self.base.meta_transform {
meta_transform(&mut *meta);
}
let asset = self
.base
.load_context
.load_direct_internal(path.clone(), meta, &*loader, reader.as_mut())
.await?;
Ok((loader, asset))
}
/// Loads the asset at the given `path` directly. This is an async function that will wait until the asset is fully loaded before
/// returning. Use this if you need the _value_ of another asset in order to load the current asset. For example, if you are
/// deriving a new asset from the referenced asset, or you are building a collection of assets. This will add the `path` as a
/// "load dependency".
///
/// If the current loader is used in a [`Process`] "asset preprocessor", such as a [`LoadTransformAndSave`] preprocessor,
/// changing a "load dependency" will result in re-processing of the asset.
///
/// [`Process`]: crate::processor::Process
/// [`LoadTransformAndSave`]: crate::processor::LoadTransformAndSave
pub async fn load<'p, A: Asset>(
mut self,
path: impl Into<AssetPath<'p>>,
) -> Result<LoadedAsset<A>, LoadDirectError> {
self.base.asset_type_id = Some(TypeId::of::<A>());
let path = path.into().into_owned();
self.load_internal(&path)
.await
.and_then(move |(loader, untyped_asset)| {
untyped_asset.downcast::<A>().map_err(|_| LoadDirectError {
dependency: path.clone(),
error: AssetLoadError::RequestedHandleTypeMismatch {
path,
requested: TypeId::of::<A>(),
actual_asset_name: loader.asset_type_name(),
loader_name: loader.type_name(),
},
})
})
}
}
/// A builder for directly loading untyped nested assets inside a `LoadContext`.
///
/// # Lifetimes
/// - `ctx`: the lifetime of the associated [`AssetServer`] reference
/// - `builder`: the lifetime of the temporary builder structs
/// - `reader`: the lifetime of the [`Reader`] reference used to read the asset data
pub struct UntypedDirectNestedLoader<'ctx, 'builder, 'reader> {
base: DirectNestedLoader<'ctx, 'builder, 'reader>,
}
impl<'ctx: 'reader, 'builder, 'reader> UntypedDirectNestedLoader<'ctx, 'builder, 'reader> {
/// Loads the asset at the given `path` directly. This is an async function that will wait until the asset is fully loaded before
/// returning. Use this if you need the _value_ of another asset in order to load the current asset. For example, if you are
/// deriving a new asset from the referenced asset, or you are building a collection of assets. This will add the `path` as a
/// "load dependency".
///
/// If the current loader is used in a [`Process`] "asset preprocessor", such as a [`LoadTransformAndSave`] preprocessor,
/// changing a "load dependency" will result in re-processing of the asset.
///
/// [`Process`]: crate::processor::Process
/// [`LoadTransformAndSave`]: crate::processor::LoadTransformAndSave
pub async fn load<'p>(
self,
path: impl Into<AssetPath<'p>>,
) -> Result<ErasedLoadedAsset, LoadDirectError> {
let path = path.into().into_owned();
self.base.load_internal(&path).await.map(|(_, asset)| asset)
}
}