Tracking Issue for const fn type_id
#77125
Comments
KodrAus
added
the
C-tracking-issue
KodrAus
added
T-lang
KodrAus
added
the
Libs-Tracked
|
I'm not sure I totally understand the steps needed to re-stabilize
Does this affect the soundness of
This seems impossible. If someone wants to Motivation: It would be extremely useful to make A perhaps related concern (may need a separate tracking issue) is that |
Can this be re-stabilized now?@KodrAus Going through the comment that was the basis of reverting stabilization (#75923 (comment)): Part 1
Sure, but this suggests that retrieving a Part 2
I think @benkay86 summed this up fairly well.
Part 3And finally, @eddyb remarks that we could potentially re-stabilize it if we are unlikely to change it.
It has been nearly 10 months since it was marked unstable and there have been no changes regarding this issue. At this point it seems like the implementation of TL;DRPlus even if we assume all of these points are true, these concerns only apply to If anything, the heavy reliance on
|
Nobody tried AFAIK. 10 months is pretty short on the scale of Rust evolution, especially when you take into account the year 2020. But also I believe it got stalled because of this potential breakage: #75923 (comment) In order to guarantee code that struct TypeId {
hash: u64,
// `str` can be replaced with a length-prefixed string to keep `TypeId`'s size down
v0_mangled_type: &'static str,
}
impl PartialEq for TypeId {
fn eq(&self, other: &TypeId) -> bool {
self.hash == other.hash // fast reject (!=)
&& (
self.v0_mangled_type as *const str == other.v0_mangled_type as *const str // fast accept (==)
|| self.v0_mangled_type == other.v0_mangled_type
)
}
}That way, it's always more than 8 bytes, no matter the target pointer size, and so a As for the linker concerns, if we take this specific approach, the only difference relying on their features would make is increasingly the situations in which the "fast accept ( I could try make a PR for this, but it would likely at least require #87194 (to avoid breaking when unstable
But we don't want that. You can't change |
Maybe it would be better to use this to shrink use core::{ptr, slice};
#[repr(C)]
struct StrWithLen<T: ?Sized> {
len: u16, // assume all TypeId strings are <64kB
text: T,
}
pub struct TypeId {
hash: u64,
text: &'static StrWithLen<()>, // actually StrWithLen<str>
}
impl PartialEq for TypeId {
fn eq(&self, rhs: &Self) -> bool {
if self.hash != rhs.hash {
false
} else if ptr::eq(self.text, rhs.text) {
true
} else {
unsafe {
let lhs_text = slice::from_raw_parts(
&self.text.text as *const _ as *const u8,
self.text.len as usize,
);
let rhs_text = slice::from_raw_parts(
&rhs.text.text as *const _ as *const u8,
rhs.text.len as usize,
);
lhs_text == rhs_text
}
}
}
} |
|
Yeah, that's what I meant by "length-prefixed". There's a "standard" way to do that using |
|
Update: finally opened #95845 with a (v0) mangling-based |
lcnr
added
the
S-blocked
|
blocking the stabilization of this on #97156 |
…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](#4042 (comment))). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e) and main (c309acd). <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 rust-lang/rust#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>
|
Why would anyone use I know that And why it is a problem anyway? I thought that transmuting arbitrary |
|
I'd say this is blocked on #99189 for now, if for no other reason than the "structural match" aspect. |
|
to work around #97156 and #99189, surely we can just add a new |
@eddyb What is the reasoning behind disallowing a structural match? While I can not name any specific reasons for its inclusion, I also can not think of any reasons to remove it either. Take for example this code. It does not perform any complex pattern matching or require that a trait ConstTypeId {
const ID: TypeId;
}
impl<T: 'static> ConstTypeId for T {
const ID: TypeId = TypeId::of::<T>();
}
match foo.type_id {
u16::ID | u32::ID | u64::ID => do_foo(&foo.data),
Foo::ID | Bar::ID | Baz::ID => do_bar(&foo.data),
_ => do_default(&foo.data),
}This would generally serve the use case of "I should probably be using an |
|
The problem is any future changes to (See all previous discussion across this issue and my two closed PRs on the matter of non-trivial Honestly, our structural match rules are very loose and prone to bite us eventually. It's not a clear independent opt-in, for ergonomics reasons, but that also makes it forwards-incompatable to "let it happen". EDIT: oh, another note on structural match: I believe it was never an intentionally designed feature, just |
…evyengine#4042) # Objective > Resolves bevyengine#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](bevyengine#4042 (comment))). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e) and main (c309acd). <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 rust-lang/rust#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>
…evyengine#4042) # Objective > Resolves bevyengine#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](bevyengine#4042 (comment))). I toggled `Reflect` and `FromReflect` derive macros using `cfg_attr` for both this PR (aa5178e) and main (c309acd). <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 rust-lang/rust#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>
I actually am just writing such a system, and that's how I found out about this issue. And given the prospect that structural matching of Not arguing for keeping it, just want to provide a data point. |
|
It seems |
|
I think a const-enabled @eddyb did we have any lingering objection to being able to do this kind of type inspection at all in CTFE, or was it just the issue of structural match on |
|
I guess that most of the other use-cases would be covered by allowing |
|
Looks like TypeId suddenly stopped implementing StructuralEq. Why? I was previously able to use this function without a problem: |
|
Removing structural match for |
|
Would it be possible to restore backward compatibility by having the compiler convert a |
|
Yes, but that will require stabilization of |
|
Most discussion about type ids in |
Previous versions of log used an alpha version of value-bag, which broke on the new nightly rustc: rust-lang/rust#77125
Previous versions of log used an alpha version of value-bag, which broke on the new nightly rustc: rust-lang/rust#77125
|
I think #10389 needs to be resolved first, becuase depending on the resolution, it may not be possible to calculate the TypeId at compile time; it might end up like C++ |
|
We can compute and equate type ids that are addresses just fine at compile time. You just can't inspect the address (so no sorting type ids, no hashing them, no printing them) |
This was recently marked fixed with #118247 Edit: started a discussion https://rust-lang.zulipchat.com/#narrow/stream/219381-t-libs/topic/Const.20.60TypeId.3A.3Aof.60.2C.20.60TypeId.3A.3Amatches.60.2C.20.60type_name.60.2C Edit2: From the discussion, this is still blocked by the possibility of collision #10389 |
oli-obk
removed
the
S-blocked
|
Any further updates on this ? |
|
Still blocked on #10389. |
This is a tracking issue for using
TypeId::ofin constant contexts.The feature gate for the issue is
#![feature(const_type_id)].About tracking issues
Tracking issues are used to record the overall progress of implementation.
They are also used as hubs connecting to other relevant issues, e.g., bugs or open design questions.
A tracking issue is however not meant for large scale discussion, questions, or bug reports about a feature.
Instead, open a dedicated issue for the specific matter and add the relevant feature gate label.
Steps
Unresolved Questions
Implementation history
type_idinto a constant intrinsic #47892The text was updated successfully, but these errors were encountered: