bevy/crates/bevy_ecs/src/change_detection.rs

386 lines
13 KiB
Rust
Raw Normal View History

//! Types that detect when their internal data mutate.
untyped APIs for components and resources (#4447) # Objective Even if bevy itself does not provide any builtin scripting or modding APIs, it should have the foundations for building them yourself. For that it should be enough to have APIs that are not tied to the actual rust types with generics, but rather accept `ComponentId`s and `bevy_ptr` ptrs. ## Solution Add the following APIs to bevy ```rust fn EntityRef::get_by_id(ComponentId) -> Option<Ptr<'w>>; fn EntityMut::get_by_id(ComponentId) -> Option<Ptr<'_>>; fn EntityMut::get_mut_by_id(ComponentId) -> Option<MutUntyped<'_>>; fn World::get_resource_by_id(ComponentId) -> Option<Ptr<'_>>; fn World::get_resource_mut_by_id(ComponentId) -> Option<MutUntyped<'_>>; // Safety: `value` must point to a valid value of the component unsafe fn World::insert_resource_by_id(ComponentId, value: OwningPtr); fn ComponentDescriptor::new_with_layout(..) -> Self; fn World::init_component_with_descriptor(ComponentDescriptor) -> ComponentId; ``` ~~This PR would definitely benefit from #3001 (lifetime'd pointers) to make sure that the lifetimes of the pointers are valid and the my-move pointer in `insert_resource_by_id` could be an `OwningPtr`, but that can be adapter later if/when #3001 is merged.~~ ### Not in this PR - inserting components on entities (this is very tied to types with bundles and the `BundleInserter`) - an untyped version of a query (needs good API design, has a large implementation complexity, can be done in a third-party crate) Co-authored-by: Jakob Hellermann <hellermann@sipgate.de>
2022-05-30 15:32:47 +00:00
use crate::{component::ComponentTicks, ptr::PtrMut, system::Resource};
#[cfg(feature = "bevy_reflect")]
use bevy_reflect::Reflect;
use std::ops::{Deref, DerefMut};
Make change lifespan deterministic and update docs (#3956) ## Objective - ~~Make absurdly long-lived changes stay detectable for even longer (without leveling up to `u64`).~~ - Give all changes a consistent maximum lifespan. - Improve code clarity. ## Solution - ~~Increase the frequency of `check_tick` scans to increase the oldest reliably-detectable change.~~ (Deferred until we can benchmark the cost of a scan.) - Ignore changes older than the maximum reliably-detectable age. - General refactoring—name the constants, use them everywhere, and update the docs. - Update test cases to check for the specified behavior. ## Related This PR addresses (at least partially) the concerns raised in: - #3071 - #3082 (and associated PR #3084) ## Background - #1471 Given the minimum interval between `check_ticks` scans, `N`, the oldest reliably-detectable change is `u32::MAX - (2 * N - 1)` (or `MAX_CHANGE_AGE`). Reducing `N` from ~530 million (current value) to something like ~2 million would extend the lifetime of changes by a billion. | minimum `check_ticks` interval | oldest reliably-detectable change | usable % of `u32::MAX` | | --- | --- | --- | | `u32::MAX / 8` (536,870,911) | `(u32::MAX / 4) * 3` | 75.0% | | `2_000_000` | `u32::MAX - 3_999_999` | 99.9% | Similarly, changes are still allowed to be between `MAX_CHANGE_AGE`-old and `u32::MAX`-old in the interim between `check_tick` scans. While we prevent their age from overflowing, the test to detect changes still compares raw values. This makes failure ultimately unreliable, since when ancient changes stop being detected varies depending on when the next scan occurs. ## Open Question Currently, systems and system states are incorrectly initialized with their `last_change_tick` set to `0`, which doesn't handle wraparound correctly. For consistent behavior, they should either be initialized to the world's `last_change_tick` (and detect no changes) or to `MAX_CHANGE_AGE` behind the world's current `change_tick` (and detect everything as a change). I've currently gone with the latter since that was closer to the existing behavior. ## Follow-up Work (Edited: entire section) We haven't actually profiled how long a `check_ticks` scan takes on a "large" `World` , so we don't know if it's safe to increase their frequency. However, we are currently relying on play sessions not lasting long enough to trigger a scan and apps not having enough entities/archetypes for it to be "expensive" (our assumption). That isn't a real solution. (Either scanning never costs enough to impact frame times or we provide an option to use `u64` change ticks. Nobody will accept random hiccups.) To further extend the lifetime of changes, we actually only need to increment the world tick if a system has `Fetch: !ReadOnlySystemParamFetch`. The behavior will be identical because all writes are sequenced, but I'm not sure how to implement that in a way that the compiler can optimize the branch out. Also, since having no false positives depends on a `check_ticks` scan running at least every `2 * N - 1` ticks, a `last_check_tick` should also be stored in the `World` so that any lull in system execution (like a command flush) could trigger a scan if needed. To be completely robust, all the systems initialized on the world should be scanned, not just those in the current stage.
2022-05-09 14:00:16 +00:00
/// The (arbitrarily chosen) minimum number of world tick increments between `check_tick` scans.
///
/// Change ticks can only be scanned when systems aren't running. Thus, if the threshold is `N`,
/// the maximum is `2 * N - 1` (i.e. the world ticks `N - 1` times, then `N` times).
///
/// If no change is older than `u32::MAX - (2 * N - 1)` following a scan, none of their ages can
/// overflow and cause false positives.
// (518,400,000 = 1000 ticks per frame * 144 frames per second * 3600 seconds per hour)
pub const CHECK_TICK_THRESHOLD: u32 = 518_400_000;
/// The maximum change tick difference that won't overflow before the next `check_tick` scan.
///
/// Changes stop being detected once they become this old.
pub const MAX_CHANGE_AGE: u32 = u32::MAX - (2 * CHECK_TICK_THRESHOLD - 1);
/// Types that implement reliable change detection.
///
/// ## Example
/// Using types that implement [`DetectChanges`], such as [`ResMut`], provide
/// a way to query if a value has been mutated in another system.
/// Normally change detecting is triggered by either [`DerefMut`] or [`AsMut`], however
/// it can be manually triggered via [`DetectChanges::set_changed`].
///
/// ```
/// use bevy_ecs::prelude::*;
///
/// struct MyResource(u32);
///
/// fn my_system(mut resource: ResMut<MyResource>) {
/// if resource.is_changed() {
/// println!("My resource was mutated!");
/// }
///
/// resource.0 = 42; // triggers change detection via [`DerefMut`]
/// }
/// ```
///
pub trait DetectChanges {
Make change lifespan deterministic and update docs (#3956) ## Objective - ~~Make absurdly long-lived changes stay detectable for even longer (without leveling up to `u64`).~~ - Give all changes a consistent maximum lifespan. - Improve code clarity. ## Solution - ~~Increase the frequency of `check_tick` scans to increase the oldest reliably-detectable change.~~ (Deferred until we can benchmark the cost of a scan.) - Ignore changes older than the maximum reliably-detectable age. - General refactoring—name the constants, use them everywhere, and update the docs. - Update test cases to check for the specified behavior. ## Related This PR addresses (at least partially) the concerns raised in: - #3071 - #3082 (and associated PR #3084) ## Background - #1471 Given the minimum interval between `check_ticks` scans, `N`, the oldest reliably-detectable change is `u32::MAX - (2 * N - 1)` (or `MAX_CHANGE_AGE`). Reducing `N` from ~530 million (current value) to something like ~2 million would extend the lifetime of changes by a billion. | minimum `check_ticks` interval | oldest reliably-detectable change | usable % of `u32::MAX` | | --- | --- | --- | | `u32::MAX / 8` (536,870,911) | `(u32::MAX / 4) * 3` | 75.0% | | `2_000_000` | `u32::MAX - 3_999_999` | 99.9% | Similarly, changes are still allowed to be between `MAX_CHANGE_AGE`-old and `u32::MAX`-old in the interim between `check_tick` scans. While we prevent their age from overflowing, the test to detect changes still compares raw values. This makes failure ultimately unreliable, since when ancient changes stop being detected varies depending on when the next scan occurs. ## Open Question Currently, systems and system states are incorrectly initialized with their `last_change_tick` set to `0`, which doesn't handle wraparound correctly. For consistent behavior, they should either be initialized to the world's `last_change_tick` (and detect no changes) or to `MAX_CHANGE_AGE` behind the world's current `change_tick` (and detect everything as a change). I've currently gone with the latter since that was closer to the existing behavior. ## Follow-up Work (Edited: entire section) We haven't actually profiled how long a `check_ticks` scan takes on a "large" `World` , so we don't know if it's safe to increase their frequency. However, we are currently relying on play sessions not lasting long enough to trigger a scan and apps not having enough entities/archetypes for it to be "expensive" (our assumption). That isn't a real solution. (Either scanning never costs enough to impact frame times or we provide an option to use `u64` change ticks. Nobody will accept random hiccups.) To further extend the lifetime of changes, we actually only need to increment the world tick if a system has `Fetch: !ReadOnlySystemParamFetch`. The behavior will be identical because all writes are sequenced, but I'm not sure how to implement that in a way that the compiler can optimize the branch out. Also, since having no false positives depends on a `check_ticks` scan running at least every `2 * N - 1` ticks, a `last_check_tick` should also be stored in the `World` so that any lull in system execution (like a command flush) could trigger a scan if needed. To be completely robust, all the systems initialized on the world should be scanned, not just those in the current stage.
2022-05-09 14:00:16 +00:00
/// Returns `true` if this value was added after the system last ran.
fn is_added(&self) -> bool;
Make change lifespan deterministic and update docs (#3956) ## Objective - ~~Make absurdly long-lived changes stay detectable for even longer (without leveling up to `u64`).~~ - Give all changes a consistent maximum lifespan. - Improve code clarity. ## Solution - ~~Increase the frequency of `check_tick` scans to increase the oldest reliably-detectable change.~~ (Deferred until we can benchmark the cost of a scan.) - Ignore changes older than the maximum reliably-detectable age. - General refactoring—name the constants, use them everywhere, and update the docs. - Update test cases to check for the specified behavior. ## Related This PR addresses (at least partially) the concerns raised in: - #3071 - #3082 (and associated PR #3084) ## Background - #1471 Given the minimum interval between `check_ticks` scans, `N`, the oldest reliably-detectable change is `u32::MAX - (2 * N - 1)` (or `MAX_CHANGE_AGE`). Reducing `N` from ~530 million (current value) to something like ~2 million would extend the lifetime of changes by a billion. | minimum `check_ticks` interval | oldest reliably-detectable change | usable % of `u32::MAX` | | --- | --- | --- | | `u32::MAX / 8` (536,870,911) | `(u32::MAX / 4) * 3` | 75.0% | | `2_000_000` | `u32::MAX - 3_999_999` | 99.9% | Similarly, changes are still allowed to be between `MAX_CHANGE_AGE`-old and `u32::MAX`-old in the interim between `check_tick` scans. While we prevent their age from overflowing, the test to detect changes still compares raw values. This makes failure ultimately unreliable, since when ancient changes stop being detected varies depending on when the next scan occurs. ## Open Question Currently, systems and system states are incorrectly initialized with their `last_change_tick` set to `0`, which doesn't handle wraparound correctly. For consistent behavior, they should either be initialized to the world's `last_change_tick` (and detect no changes) or to `MAX_CHANGE_AGE` behind the world's current `change_tick` (and detect everything as a change). I've currently gone with the latter since that was closer to the existing behavior. ## Follow-up Work (Edited: entire section) We haven't actually profiled how long a `check_ticks` scan takes on a "large" `World` , so we don't know if it's safe to increase their frequency. However, we are currently relying on play sessions not lasting long enough to trigger a scan and apps not having enough entities/archetypes for it to be "expensive" (our assumption). That isn't a real solution. (Either scanning never costs enough to impact frame times or we provide an option to use `u64` change ticks. Nobody will accept random hiccups.) To further extend the lifetime of changes, we actually only need to increment the world tick if a system has `Fetch: !ReadOnlySystemParamFetch`. The behavior will be identical because all writes are sequenced, but I'm not sure how to implement that in a way that the compiler can optimize the branch out. Also, since having no false positives depends on a `check_ticks` scan running at least every `2 * N - 1` ticks, a `last_check_tick` should also be stored in the `World` so that any lull in system execution (like a command flush) could trigger a scan if needed. To be completely robust, all the systems initialized on the world should be scanned, not just those in the current stage.
2022-05-09 14:00:16 +00:00
/// Returns `true` if this value was added or mutably dereferenced after the system last ran.
fn is_changed(&self) -> bool;
Make change lifespan deterministic and update docs (#3956) ## Objective - ~~Make absurdly long-lived changes stay detectable for even longer (without leveling up to `u64`).~~ - Give all changes a consistent maximum lifespan. - Improve code clarity. ## Solution - ~~Increase the frequency of `check_tick` scans to increase the oldest reliably-detectable change.~~ (Deferred until we can benchmark the cost of a scan.) - Ignore changes older than the maximum reliably-detectable age. - General refactoring—name the constants, use them everywhere, and update the docs. - Update test cases to check for the specified behavior. ## Related This PR addresses (at least partially) the concerns raised in: - #3071 - #3082 (and associated PR #3084) ## Background - #1471 Given the minimum interval between `check_ticks` scans, `N`, the oldest reliably-detectable change is `u32::MAX - (2 * N - 1)` (or `MAX_CHANGE_AGE`). Reducing `N` from ~530 million (current value) to something like ~2 million would extend the lifetime of changes by a billion. | minimum `check_ticks` interval | oldest reliably-detectable change | usable % of `u32::MAX` | | --- | --- | --- | | `u32::MAX / 8` (536,870,911) | `(u32::MAX / 4) * 3` | 75.0% | | `2_000_000` | `u32::MAX - 3_999_999` | 99.9% | Similarly, changes are still allowed to be between `MAX_CHANGE_AGE`-old and `u32::MAX`-old in the interim between `check_tick` scans. While we prevent their age from overflowing, the test to detect changes still compares raw values. This makes failure ultimately unreliable, since when ancient changes stop being detected varies depending on when the next scan occurs. ## Open Question Currently, systems and system states are incorrectly initialized with their `last_change_tick` set to `0`, which doesn't handle wraparound correctly. For consistent behavior, they should either be initialized to the world's `last_change_tick` (and detect no changes) or to `MAX_CHANGE_AGE` behind the world's current `change_tick` (and detect everything as a change). I've currently gone with the latter since that was closer to the existing behavior. ## Follow-up Work (Edited: entire section) We haven't actually profiled how long a `check_ticks` scan takes on a "large" `World` , so we don't know if it's safe to increase their frequency. However, we are currently relying on play sessions not lasting long enough to trigger a scan and apps not having enough entities/archetypes for it to be "expensive" (our assumption). That isn't a real solution. (Either scanning never costs enough to impact frame times or we provide an option to use `u64` change ticks. Nobody will accept random hiccups.) To further extend the lifetime of changes, we actually only need to increment the world tick if a system has `Fetch: !ReadOnlySystemParamFetch`. The behavior will be identical because all writes are sequenced, but I'm not sure how to implement that in a way that the compiler can optimize the branch out. Also, since having no false positives depends on a `check_ticks` scan running at least every `2 * N - 1` ticks, a `last_check_tick` should also be stored in the `World` so that any lull in system execution (like a command flush) could trigger a scan if needed. To be completely robust, all the systems initialized on the world should be scanned, not just those in the current stage.
2022-05-09 14:00:16 +00:00
/// Flags this value as having been changed.
///
Make change lifespan deterministic and update docs (#3956) ## Objective - ~~Make absurdly long-lived changes stay detectable for even longer (without leveling up to `u64`).~~ - Give all changes a consistent maximum lifespan. - Improve code clarity. ## Solution - ~~Increase the frequency of `check_tick` scans to increase the oldest reliably-detectable change.~~ (Deferred until we can benchmark the cost of a scan.) - Ignore changes older than the maximum reliably-detectable age. - General refactoring—name the constants, use them everywhere, and update the docs. - Update test cases to check for the specified behavior. ## Related This PR addresses (at least partially) the concerns raised in: - #3071 - #3082 (and associated PR #3084) ## Background - #1471 Given the minimum interval between `check_ticks` scans, `N`, the oldest reliably-detectable change is `u32::MAX - (2 * N - 1)` (or `MAX_CHANGE_AGE`). Reducing `N` from ~530 million (current value) to something like ~2 million would extend the lifetime of changes by a billion. | minimum `check_ticks` interval | oldest reliably-detectable change | usable % of `u32::MAX` | | --- | --- | --- | | `u32::MAX / 8` (536,870,911) | `(u32::MAX / 4) * 3` | 75.0% | | `2_000_000` | `u32::MAX - 3_999_999` | 99.9% | Similarly, changes are still allowed to be between `MAX_CHANGE_AGE`-old and `u32::MAX`-old in the interim between `check_tick` scans. While we prevent their age from overflowing, the test to detect changes still compares raw values. This makes failure ultimately unreliable, since when ancient changes stop being detected varies depending on when the next scan occurs. ## Open Question Currently, systems and system states are incorrectly initialized with their `last_change_tick` set to `0`, which doesn't handle wraparound correctly. For consistent behavior, they should either be initialized to the world's `last_change_tick` (and detect no changes) or to `MAX_CHANGE_AGE` behind the world's current `change_tick` (and detect everything as a change). I've currently gone with the latter since that was closer to the existing behavior. ## Follow-up Work (Edited: entire section) We haven't actually profiled how long a `check_ticks` scan takes on a "large" `World` , so we don't know if it's safe to increase their frequency. However, we are currently relying on play sessions not lasting long enough to trigger a scan and apps not having enough entities/archetypes for it to be "expensive" (our assumption). That isn't a real solution. (Either scanning never costs enough to impact frame times or we provide an option to use `u64` change ticks. Nobody will accept random hiccups.) To further extend the lifetime of changes, we actually only need to increment the world tick if a system has `Fetch: !ReadOnlySystemParamFetch`. The behavior will be identical because all writes are sequenced, but I'm not sure how to implement that in a way that the compiler can optimize the branch out. Also, since having no false positives depends on a `check_ticks` scan running at least every `2 * N - 1` ticks, a `last_check_tick` should also be stored in the `World` so that any lull in system execution (like a command flush) could trigger a scan if needed. To be completely robust, all the systems initialized on the world should be scanned, not just those in the current stage.
2022-05-09 14:00:16 +00:00
/// Mutably accessing this smart pointer will automatically flag this value as having been changed.
/// However, mutation through interior mutability requires manual reporting.
///
/// **Note**: This operation cannot be undone.
fn set_changed(&mut self);
/// Returns the change tick recording the previous time this component (or resource) was changed.
///
/// Note that components and resources are also marked as changed upon insertion.
///
/// For comparison, the previous change tick of a system can be read using the
/// [`SystemChangeTick`](crate::system::SystemChangeTick)
/// [`SystemParam`](crate::system::SystemParam).
fn last_changed(&self) -> u32;
}
macro_rules! change_detection_impl {
($name:ident < $( $generics:tt ),+ >, $target:ty, $($traits:ident)?) => {
impl<$($generics),* $(: $traits)?> DetectChanges for $name<$($generics),*> {
#[inline]
fn is_added(&self) -> bool {
self.ticks
.component_ticks
.is_added(self.ticks.last_change_tick, self.ticks.change_tick)
}
#[inline]
fn is_changed(&self) -> bool {
self.ticks
.component_ticks
.is_changed(self.ticks.last_change_tick, self.ticks.change_tick)
}
#[inline]
fn set_changed(&mut self) {
self.ticks
.component_ticks
.set_changed(self.ticks.change_tick);
}
#[inline]
fn last_changed(&self) -> u32 {
self.ticks.last_change_tick
}
}
impl<$($generics),* $(: $traits)?> Deref for $name<$($generics),*> {
type Target = $target;
#[inline]
fn deref(&self) -> &Self::Target {
self.value
}
}
impl<$($generics),* $(: $traits)?> DerefMut for $name<$($generics),*> {
#[inline]
fn deref_mut(&mut self) -> &mut Self::Target {
self.set_changed();
self.value
}
}
impl<$($generics),* $(: $traits)?> AsRef<$target> for $name<$($generics),*> {
#[inline]
fn as_ref(&self) -> &$target {
self.deref()
}
}
impl<$($generics),* $(: $traits)?> AsMut<$target> for $name<$($generics),*> {
#[inline]
fn as_mut(&mut self) -> &mut $target {
self.deref_mut()
}
}
};
}
macro_rules! impl_into_inner {
($name:ident < $( $generics:tt ),+ >, $target:ty, $($traits:ident)?) => {
impl<$($generics),* $(: $traits)?> $name<$($generics),*> {
/// Consume `self` and return a mutable reference to the
/// contained value while marking `self` as "changed".
#[inline]
pub fn into_inner(mut self) -> &'a mut $target {
self.set_changed();
self.value
}
}
};
}
macro_rules! impl_debug {
($name:ident < $( $generics:tt ),+ >, $($traits:ident)?) => {
impl<$($generics),* $(: $traits)?> std::fmt::Debug for $name<$($generics),*>
where T: std::fmt::Debug
{
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_tuple(stringify!($name))
.field(self.value)
.finish()
}
}
};
}
pub(crate) struct Ticks<'a> {
pub(crate) component_ticks: &'a mut ComponentTicks,
pub(crate) last_change_tick: u32,
pub(crate) change_tick: u32,
}
/// Unique mutable borrow of a resource.
///
/// See the [`World`](crate::world::World) documentation to see the usage of a resource.
///
/// If you need a shared borrow, use [`Res`](crate::system::Res) instead.
///
/// # Panics
///
/// Panics when used as a [`SystemParam`](crate::system::SystemParam) if the resource does not exist.
///
/// Use `Option<ResMut<T>>` instead if the resource might not always exist.
pub struct ResMut<'a, T: Resource> {
pub(crate) value: &'a mut T,
pub(crate) ticks: Ticks<'a>,
}
change_detection_impl!(ResMut<'a, T>, T, Resource);
impl_into_inner!(ResMut<'a, T>, T, Resource);
impl_debug!(ResMut<'a, T>, Resource);
/// Unique borrow of a non-[`Send`] resource.
///
/// Only [`Send`] resources may be accessed with the [`ResMut`] [`SystemParam`](crate::system::SystemParam). In case that the
/// resource does not implement `Send`, this `SystemParam` wrapper can be used. This will instruct
/// the scheduler to instead run the system on the main thread so that it doesn't send the resource
/// over to another thread.
///
/// # Panics
///
/// Panics when used as a `SystemParameter` if the resource does not exist.
///
/// Use `Option<NonSendMut<T>>` instead if the resource might not always exist.
pub struct NonSendMut<'a, T: 'static> {
pub(crate) value: &'a mut T,
pub(crate) ticks: Ticks<'a>,
}
change_detection_impl!(NonSendMut<'a, T>, T,);
impl_into_inner!(NonSendMut<'a, T>, T,);
impl_debug!(NonSendMut<'a, T>,);
/// Unique mutable borrow of an entity's component
pub struct Mut<'a, T> {
pub(crate) value: &'a mut T,
pub(crate) ticks: Ticks<'a>,
}
change_detection_impl!(Mut<'a, T>, T,);
impl_into_inner!(Mut<'a, T>, T,);
impl_debug!(Mut<'a, T>,);
/// Unique mutable borrow of a reflected component or resource
#[cfg(feature = "bevy_reflect")]
pub struct ReflectMut<'a> {
pub(crate) value: &'a mut dyn Reflect,
pub(crate) ticks: Ticks<'a>,
}
#[cfg(feature = "bevy_reflect")]
change_detection_impl!(ReflectMut<'a>, dyn Reflect,);
#[cfg(feature = "bevy_reflect")]
impl_into_inner!(ReflectMut<'a>, dyn Reflect,);
Make change lifespan deterministic and update docs (#3956) ## Objective - ~~Make absurdly long-lived changes stay detectable for even longer (without leveling up to `u64`).~~ - Give all changes a consistent maximum lifespan. - Improve code clarity. ## Solution - ~~Increase the frequency of `check_tick` scans to increase the oldest reliably-detectable change.~~ (Deferred until we can benchmark the cost of a scan.) - Ignore changes older than the maximum reliably-detectable age. - General refactoring—name the constants, use them everywhere, and update the docs. - Update test cases to check for the specified behavior. ## Related This PR addresses (at least partially) the concerns raised in: - #3071 - #3082 (and associated PR #3084) ## Background - #1471 Given the minimum interval between `check_ticks` scans, `N`, the oldest reliably-detectable change is `u32::MAX - (2 * N - 1)` (or `MAX_CHANGE_AGE`). Reducing `N` from ~530 million (current value) to something like ~2 million would extend the lifetime of changes by a billion. | minimum `check_ticks` interval | oldest reliably-detectable change | usable % of `u32::MAX` | | --- | --- | --- | | `u32::MAX / 8` (536,870,911) | `(u32::MAX / 4) * 3` | 75.0% | | `2_000_000` | `u32::MAX - 3_999_999` | 99.9% | Similarly, changes are still allowed to be between `MAX_CHANGE_AGE`-old and `u32::MAX`-old in the interim between `check_tick` scans. While we prevent their age from overflowing, the test to detect changes still compares raw values. This makes failure ultimately unreliable, since when ancient changes stop being detected varies depending on when the next scan occurs. ## Open Question Currently, systems and system states are incorrectly initialized with their `last_change_tick` set to `0`, which doesn't handle wraparound correctly. For consistent behavior, they should either be initialized to the world's `last_change_tick` (and detect no changes) or to `MAX_CHANGE_AGE` behind the world's current `change_tick` (and detect everything as a change). I've currently gone with the latter since that was closer to the existing behavior. ## Follow-up Work (Edited: entire section) We haven't actually profiled how long a `check_ticks` scan takes on a "large" `World` , so we don't know if it's safe to increase their frequency. However, we are currently relying on play sessions not lasting long enough to trigger a scan and apps not having enough entities/archetypes for it to be "expensive" (our assumption). That isn't a real solution. (Either scanning never costs enough to impact frame times or we provide an option to use `u64` change ticks. Nobody will accept random hiccups.) To further extend the lifetime of changes, we actually only need to increment the world tick if a system has `Fetch: !ReadOnlySystemParamFetch`. The behavior will be identical because all writes are sequenced, but I'm not sure how to implement that in a way that the compiler can optimize the branch out. Also, since having no false positives depends on a `check_ticks` scan running at least every `2 * N - 1` ticks, a `last_check_tick` should also be stored in the `World` so that any lull in system execution (like a command flush) could trigger a scan if needed. To be completely robust, all the systems initialized on the world should be scanned, not just those in the current stage.
2022-05-09 14:00:16 +00:00
untyped APIs for components and resources (#4447) # Objective Even if bevy itself does not provide any builtin scripting or modding APIs, it should have the foundations for building them yourself. For that it should be enough to have APIs that are not tied to the actual rust types with generics, but rather accept `ComponentId`s and `bevy_ptr` ptrs. ## Solution Add the following APIs to bevy ```rust fn EntityRef::get_by_id(ComponentId) -> Option<Ptr<'w>>; fn EntityMut::get_by_id(ComponentId) -> Option<Ptr<'_>>; fn EntityMut::get_mut_by_id(ComponentId) -> Option<MutUntyped<'_>>; fn World::get_resource_by_id(ComponentId) -> Option<Ptr<'_>>; fn World::get_resource_mut_by_id(ComponentId) -> Option<MutUntyped<'_>>; // Safety: `value` must point to a valid value of the component unsafe fn World::insert_resource_by_id(ComponentId, value: OwningPtr); fn ComponentDescriptor::new_with_layout(..) -> Self; fn World::init_component_with_descriptor(ComponentDescriptor) -> ComponentId; ``` ~~This PR would definitely benefit from #3001 (lifetime'd pointers) to make sure that the lifetimes of the pointers are valid and the my-move pointer in `insert_resource_by_id` could be an `OwningPtr`, but that can be adapter later if/when #3001 is merged.~~ ### Not in this PR - inserting components on entities (this is very tied to types with bundles and the `BundleInserter`) - an untyped version of a query (needs good API design, has a large implementation complexity, can be done in a third-party crate) Co-authored-by: Jakob Hellermann <hellermann@sipgate.de>
2022-05-30 15:32:47 +00:00
/// Unique mutable borrow of resources or an entity's component.
///
/// Similar to [`Mut`], but not generic over the component type, instead
/// exposing the raw pointer as a `*mut ()`.
///
/// Usually you don't need to use this and can instead use the APIs returning a
/// [`Mut`], but in situations where the types are not known at compile time
/// or are defined outside of rust this can be used.
pub struct MutUntyped<'a> {
pub(crate) value: PtrMut<'a>,
pub(crate) ticks: Ticks<'a>,
}
impl<'a> MutUntyped<'a> {
/// Returns the pointer to the value, without marking it as changed.
///
/// In order to mark the value as changed, you need to call [`set_changed`](DetectChanges::set_changed) manually.
pub fn into_inner(self) -> PtrMut<'a> {
self.value
}
}
impl DetectChanges for MutUntyped<'_> {
fn is_added(&self) -> bool {
self.ticks
.component_ticks
.is_added(self.ticks.last_change_tick, self.ticks.change_tick)
}
fn is_changed(&self) -> bool {
self.ticks
.component_ticks
.is_changed(self.ticks.last_change_tick, self.ticks.change_tick)
}
fn set_changed(&mut self) {
self.ticks
.component_ticks
.set_changed(self.ticks.change_tick);
}
fn last_changed(&self) -> u32 {
self.ticks.last_change_tick
}
}
impl std::fmt::Debug for MutUntyped<'_> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_tuple("MutUntyped")
.field(&self.value.as_ptr())
.finish()
}
}
Make change lifespan deterministic and update docs (#3956) ## Objective - ~~Make absurdly long-lived changes stay detectable for even longer (without leveling up to `u64`).~~ - Give all changes a consistent maximum lifespan. - Improve code clarity. ## Solution - ~~Increase the frequency of `check_tick` scans to increase the oldest reliably-detectable change.~~ (Deferred until we can benchmark the cost of a scan.) - Ignore changes older than the maximum reliably-detectable age. - General refactoring—name the constants, use them everywhere, and update the docs. - Update test cases to check for the specified behavior. ## Related This PR addresses (at least partially) the concerns raised in: - #3071 - #3082 (and associated PR #3084) ## Background - #1471 Given the minimum interval between `check_ticks` scans, `N`, the oldest reliably-detectable change is `u32::MAX - (2 * N - 1)` (or `MAX_CHANGE_AGE`). Reducing `N` from ~530 million (current value) to something like ~2 million would extend the lifetime of changes by a billion. | minimum `check_ticks` interval | oldest reliably-detectable change | usable % of `u32::MAX` | | --- | --- | --- | | `u32::MAX / 8` (536,870,911) | `(u32::MAX / 4) * 3` | 75.0% | | `2_000_000` | `u32::MAX - 3_999_999` | 99.9% | Similarly, changes are still allowed to be between `MAX_CHANGE_AGE`-old and `u32::MAX`-old in the interim between `check_tick` scans. While we prevent their age from overflowing, the test to detect changes still compares raw values. This makes failure ultimately unreliable, since when ancient changes stop being detected varies depending on when the next scan occurs. ## Open Question Currently, systems and system states are incorrectly initialized with their `last_change_tick` set to `0`, which doesn't handle wraparound correctly. For consistent behavior, they should either be initialized to the world's `last_change_tick` (and detect no changes) or to `MAX_CHANGE_AGE` behind the world's current `change_tick` (and detect everything as a change). I've currently gone with the latter since that was closer to the existing behavior. ## Follow-up Work (Edited: entire section) We haven't actually profiled how long a `check_ticks` scan takes on a "large" `World` , so we don't know if it's safe to increase their frequency. However, we are currently relying on play sessions not lasting long enough to trigger a scan and apps not having enough entities/archetypes for it to be "expensive" (our assumption). That isn't a real solution. (Either scanning never costs enough to impact frame times or we provide an option to use `u64` change ticks. Nobody will accept random hiccups.) To further extend the lifetime of changes, we actually only need to increment the world tick if a system has `Fetch: !ReadOnlySystemParamFetch`. The behavior will be identical because all writes are sequenced, but I'm not sure how to implement that in a way that the compiler can optimize the branch out. Also, since having no false positives depends on a `check_ticks` scan running at least every `2 * N - 1` ticks, a `last_check_tick` should also be stored in the `World` so that any lull in system execution (like a command flush) could trigger a scan if needed. To be completely robust, all the systems initialized on the world should be scanned, not just those in the current stage.
2022-05-09 14:00:16 +00:00
#[cfg(test)]
mod tests {
use crate::{
self as bevy_ecs,
change_detection::{CHECK_TICK_THRESHOLD, MAX_CHANGE_AGE},
component::Component,
query::ChangeTrackers,
system::{IntoSystem, Query, System},
world::World,
};
#[derive(Component)]
struct C;
#[test]
fn change_expiration() {
fn change_detected(query: Query<ChangeTrackers<C>>) -> bool {
query.single().is_changed()
}
fn change_expired(query: Query<ChangeTrackers<C>>) -> bool {
query.single().is_changed()
}
let mut world = World::new();
// component added: 1, changed: 1
world.spawn().insert(C);
let mut change_detected_system = IntoSystem::into_system(change_detected);
let mut change_expired_system = IntoSystem::into_system(change_expired);
change_detected_system.initialize(&mut world);
change_expired_system.initialize(&mut world);
// world: 1, system last ran: 0, component changed: 1
// The spawn will be detected since it happened after the system "last ran".
assert!(change_detected_system.run((), &mut world));
// world: 1 + MAX_CHANGE_AGE
let change_tick = world.change_tick.get_mut();
*change_tick = change_tick.wrapping_add(MAX_CHANGE_AGE);
// Both the system and component appeared `MAX_CHANGE_AGE` ticks ago.
// Since we clamp things to `MAX_CHANGE_AGE` for determinism,
// `ComponentTicks::is_changed` will now see `MAX_CHANGE_AGE > MAX_CHANGE_AGE`
// and return `false`.
assert!(!change_expired_system.run((), &mut world));
}
#[test]
fn change_tick_wraparound() {
fn change_detected(query: Query<ChangeTrackers<C>>) -> bool {
query.single().is_changed()
}
let mut world = World::new();
world.last_change_tick = u32::MAX;
*world.change_tick.get_mut() = 0;
// component added: 0, changed: 0
world.spawn().insert(C);
// system last ran: u32::MAX
let mut change_detected_system = IntoSystem::into_system(change_detected);
change_detected_system.initialize(&mut world);
// Since the world is always ahead, as long as changes can't get older than `u32::MAX` (which we ensure),
// the wrapping difference will always be positive, so wraparound doesn't matter.
assert!(change_detected_system.run((), &mut world));
}
#[test]
fn change_tick_scan() {
let mut world = World::new();
// component added: 1, changed: 1
world.spawn().insert(C);
// a bunch of stuff happens, the component is now older than `MAX_CHANGE_AGE`
*world.change_tick.get_mut() += MAX_CHANGE_AGE + CHECK_TICK_THRESHOLD;
let change_tick = world.change_tick();
let mut query = world.query::<ChangeTrackers<C>>();
for tracker in query.iter(&world) {
let ticks_since_insert = change_tick.wrapping_sub(tracker.component_ticks.added);
let ticks_since_change = change_tick.wrapping_sub(tracker.component_ticks.changed);
assert!(ticks_since_insert > MAX_CHANGE_AGE);
assert!(ticks_since_change > MAX_CHANGE_AGE);
}
// scan change ticks and clamp those at risk of overflow
world.check_change_ticks();
for tracker in query.iter(&world) {
let ticks_since_insert = change_tick.wrapping_sub(tracker.component_ticks.added);
let ticks_since_change = change_tick.wrapping_sub(tracker.component_ticks.changed);
assert!(ticks_since_insert == MAX_CHANGE_AGE);
assert!(ticks_since_change == MAX_CHANGE_AGE);
}
}
}