-
Notifications
You must be signed in to change notification settings - Fork 13.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Sized Hierarchy #137944
base: master
Are you sure you want to change the base?
Sized Hierarchy #137944
Conversation
This comment has been minimized.
This comment has been minimized.
|
This comment has been minimized.
This comment has been minimized.
73f1b4f
to
7f509ab
Compare
This comment has been minimized.
This comment has been minimized.
2537bfb
to
b31fd85
Compare
cc @rust-lang/lang |
Does this perhaps fix #127336 by rejecting it? |
This comment was marked as resolved.
This comment was marked as resolved.
b31fd85
to
2beed43
Compare
It doesn't currently. |
2beed43
to
839b844
Compare
Undrafting now that CI passes |
These should never be shown to users at the moment.
One clippy test is `no_core` and needs to have `MetaSized` and `PointeeSized` added to it.
Existing lints that had special-casing for `Sized` predicates ought to have these same special cases applied to `MetaSized` predicates.
Unexpected Clippy lint triggering is fixed in previous commits but is necessary for `cfg(bootstrap)`.
It's unclear why this change in miri is necessary.
These error messages include lines of the standard library which have changed and so need updated.
Unstability is propagated when staged API is not used and `-Zforce-unstable-if-unmarked` is, but this only happened for const functions not const traits, which ends up being an issue for some of the minicore tests of codegen backends when introducing the sizedness traits.
As in many previous commits, adding the new traits and constness to the minicore.
☔ The latest upstream changes (presumably #138083) made this pull request unmergeable. Please resolve the merge conflicts. |
I've been meaning to run perf on this and see what the damage is: @bors try @rust-timer queue |
This comment has been minimized.
This comment has been minimized.
This comment was marked as resolved.
This comment was marked as resolved.
3e6ac54
to
78229bc
Compare
@bors try @rust-timer queue |
This comment has been minimized.
This comment has been minimized.
Sized Hierarchy This patch implements rust-lang/rfcs#3729. It introduces two new traits to the standard library, `MetaSized` and `PointeeSized`, and makes `MetaSized` and `Sized` into const traits (relying on unstable `feature(const_trait_impl)`). See the RFC for the rationale behind these traits and to discuss whether this change makes sense in the abstract. These traits are unstable (as is their constness), so users cannot refer to them without opting-in to `feature(sized_hierarchy)`. These traits are not behind `cfg`s as this would make implementation unfeasible, there would simply be too many `cfg`s required to add the necessary bounds everywhere. So, like `Sized`, these traits are automatically implemented by the compiler. RFC 3729 describes migrations which are necessary to preserve backwards compatibility given the introduction of these traits, which are implemented and as follows: - On the current edition, `Sized` is rewritten as `const Sized` - If the `sized_hierarchy` feature is enabled, then an edition migration lint to rewrite the bound to `const Sized` will be emitted. - On the next edition, non-const `Sized` will resume being the default bound. - On the current edition, `?Sized` is rewritten as `const MetaSized` - If the `sized_hierarchy` feature is enabled, then an edition migration lint to rewrite the bound to `const MetaSized` will be emitted. - On the next edition, writing `?Sized` will be prohibited. - On the current edition, `const MetaSized` is added as a default supertrait for all traits w/out an explicit sizedness supertrait already. - If the `sized_hierarchy` feature is enabled, then an edition migration lint to add an explicit `const MetaSized` supertrait will be emitted. - On the next edition, there is no default `const MetaSized` supertrait. Each of these migrations is not conditional on whether the item being migrated *needs* the migration to the stricter bound - this would be preferable but is not yet implemented (if it is possible to implement). All diagnostic output should remain the same (showing `?Sized` even if the compiler sees `const MetaSized`) unless the `sized_hierarchy` feature is enabled. Due to the use of unstable extern types in the standard library and rustc, some bounds in both projects have had to be relaxed already - this is unfortunate but unavoidable so that these extern types can continue to be used where they were before. Performing these relaxations in the standard library and rustc are desirable longer-term anyway, but some bounds are not as relaxed as they ideally would be due to the inability to relax `Deref::Target` (this will be investigated separately). It is hoped that this is implemented such that it could be merged and these traits could exist "under the hood" without that being observable to the user (other than in any performance impact this has on the compiler, etc). Only once `sized_hierarchy` is stabilised would edition migration lints start to be emitted and diagnostic output show the "real" sizedness traits behind-the-scenes, rather than `?Sized`. Some details might leak through due to the standard library relaxations, but this has not been observed in test output. **Notes:** - Any commits starting with "upstream:" can be ignored, as these correspond to other upstream PRs that this is based on which have yet to be merged. - This best reviewed commit-by-commit. I've attempted to make the implementation easy to follow and keep similar changes and test output updates together. - Each commit has a short description describing its purpose. - This patch is large but it's primarily in the test suite (library: +573/-184, compiler: +1268/-310, tests: +3720/-452). - It is expected that this will have performance regressions initially and I'll aim to resolve those prior to merging if possible. - I'd appreciate feedback on how best to go about this from those familiar with the type system. - On my local machine, this passes all of the test suites, a stage two build and a tidy check. - `PointeeSized` is a different name from the RFC just to make it more obvious that it is different from `std::ptr::Pointee` but all the names are yet to be bikeshed anyway. Fixes rust-lang#79409. r? `@ghost` (I'll discuss this with relevant teams to find a reviewer)
☀️ Try build successful - checks-actions |
This comment has been minimized.
This comment has been minimized.
Finished benchmarking commit (e8cd94d): comparison URL. Overall result: ❌ regressions - please read the text belowBenchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf. Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @bors rollup=never Instruction countThis is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.
Max RSS (memory usage)Results (primary 2.2%, secondary 3.2%)This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
CyclesResults (primary 4.8%, secondary 7.2%)This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
Binary sizeResults (primary 0.3%, secondary 0.3%)This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
Bootstrap: 779.458s -> 790.01s (1.35%) |
Okay, so there's some work to do on perf, as expected. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Second pass. Will spend more time on it today after I get to my PC and not from my laptop :)
|
||
/// Types that may or may not have a size. | ||
#[unstable(feature = "sized_hierarchy", issue = "none")] | ||
#[cfg_attr(not(bootstrap), lang = "pointeesized")] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
lang = "pointee_sized"
/// [`structural_traits::instantiate_constituent_tys_for_sized_trait`], | ||
/// [`structural_traits::instantiate_constituent_tys_for_metasized_trait`] and | ||
/// [`structural_traits::instantiate_constituent_tys_for_pointeesized_trait`]. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe better to just refer to instantiate_constituent_tys_for_sizedness_trait
only
match sizedness { | ||
SizedTraitKind::Sized => { | ||
structural_traits::instantiate_constituent_tys_for_sized_trait | ||
} | ||
SizedTraitKind::MetaSized => { | ||
structural_traits::instantiate_constituent_tys_for_metasized_trait | ||
} | ||
SizedTraitKind::PointeeSized => { | ||
structural_traits::instantiate_constituent_tys_for_pointeesized_trait | ||
} | ||
}, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think we this extra layer of indirection as there is already instantiate_constituent_tys_for_sizedness_trait
which takes any SizedTraitKind
// impl {Meta,}Sized for Adt<Args...> where {meta,}sized_constraint(Adt)<Args...>: {Meta,}Sized | ||
// `{meta,}sized_constraint(Adt)` is the deepest struct trail that can be determined | ||
// by the definition of `Adt`, independent of the generic args. | ||
// impl Sized for Adt<Args...> if sized_constraint(Adt) == None | ||
// As a performance optimization, `sized_constraint(Adt)` can return `None` | ||
// if the ADTs definition implies that it is sized by for all possible args. | ||
// impl {Meta,}Sized for Adt<Args...> if {meta,}sized_constraint(Adt) == None | ||
// As a performance optimization, `{meta,}sized_constraint(Adt)` can return `None` | ||
// if the ADTs definition implies that it is {meta,}sized by for all possible args. | ||
// In this case, the builtin impl will have no nested subgoals. This is a | ||
// "best effort" optimization and `sized_constraint` may return `Some`, even | ||
// if the ADT is sized for all possible args. | ||
// "best effort" optimization and `{meta,}sized_constraint` may return `Some`, even | ||
// if the ADT is {meta,}sized for all possible args. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
these should all contain pointee also
if ptr.trait_ref.path.res == Res::Def(DefKind::Trait, pointeesized_did) { | ||
match ptr.modifiers.polarity { | ||
hir::BoundPolarity::Maybe(_) => pointeesized.maybe = true, | ||
hir::BoundPolarity::Negative(_) => pointeesized.negative = true, | ||
hir::BoundPolarity::Positive => pointeesized.positive = true, | ||
} | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This part can be deduplicated through doing something like
let collected = if ... { &mut sized } else if ... { &mut meta_sized } else if { &mut pointee_sized } else { continue };
let mut idx = None; | ||
for (cur_idx, (clause, _)) in bounds.into_iter().enumerate() { | ||
if let Some(clause) = clause.as_trait_clause() | ||
&& clause.skip_binder().def_id() == did | ||
{ | ||
idx = Some(cur_idx); | ||
} | ||
} | ||
|
||
if let Some(idx) = idx { | ||
bounds.remove(idx); | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Any chance for duplication? In any case we should use
bounds.retain(|(clause, _)| clause.as_trait_clause().map_or(false, |clause| clause.skip_binder().def_id() != did))
instead.
trait Sized_: Sized { } | ||
|
||
trait NegSized: ?Sized { } | ||
//~^ ERROR `?Trait` is not permitted in supertraits | ||
|
||
trait MetaSized_: MetaSized { } | ||
|
||
trait NegMetaSized: ?MetaSized { } | ||
//~^ ERROR `?Trait` is not permitted in supertraits | ||
|
||
|
||
trait PointeeSized_: PointeeSized { } | ||
|
||
trait NegPointeeSized: ?PointeeSized { } | ||
//~^ ERROR `?Trait` is not permitted in supertraits | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should add a trait Bare {}
here and test
//@ check-pass | ||
#![feature(sized_hierarchy)] | ||
|
||
use std::marker::PhantomData; | ||
|
||
pub trait Bar<'a> { | ||
type Foo; | ||
} | ||
|
||
pub struct Foo<'a, T: Bar<'a>> { | ||
phantom: PhantomData<&'a T>, | ||
} | ||
|
||
impl<'a, 'b, T> PartialEq<Foo<'b, T>> for Foo<'a, T> | ||
where | ||
T: for<'c> Bar<'c>, | ||
<T as Bar<'a>>::Foo: PartialEq<<T as Bar<'b>>::Foo>, | ||
{ | ||
fn eq(&self, _: &Foo<'b, T>) -> bool { | ||
loop {} | ||
} | ||
} | ||
|
||
fn main() { } |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmmm. Where does this test come from?
let pointeesized_predicate = { | ||
let pointeesized_did = tcx.require_lang_item(LangItem::PointeeSized, None); | ||
ty::TraitRef::new(tcx, pointeesized_did, [unsized_self_ty]).upcast(tcx) | ||
}; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Does this need to be proven? It's not referenced in the commit message, and also U: MetaSized
already implies it.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
How would the pretty printer work if we're working with a dependency crate that uses the sized_hierarchy
feature but the caller does not enable it? Add a test please
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have some general issues with the commits starting from "tests: PointeeSized bounds with extern types" to "tests: bless remaining tests".
In those tests' cases, were enabling the sized_hierarchy
required for some tests to work? For example, before the feature was enabled, do functions like fn foo<T: ?Sized>()
still accept opaque types? If they don't and need the feature to be enabled to work again, then that's a potential breakage.
impl<V: ?Sized> Callable for () { | ||
impl<V: PointeeSized> Callable for () { | ||
//~^ ERROR the type parameter `V` is not constrained by the impl trait, self type, or predicates | ||
fn call() {} | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This error still fires for impl<V: ?Sized>
right? They probably don't need to use feature(sized_hierarchy)
?
|
||
#![feature(sized_hierarchy)] | ||
#![feature(non_lifetime_binders)] | ||
//~^ WARN the feature `non_lifetime_binders` is incomplete | ||
|
||
trait Trait<T: ?Sized> {} | ||
use std::marker::PointeeSized; | ||
|
||
impl<T: ?Sized> Trait<T> for i32 {} | ||
trait Trait<T: PointeeSized> {} | ||
|
||
impl<T: PointeeSized> Trait<T> for i32 {} | ||
|
||
fn produce() -> impl for<T> Trait<T> { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
After the sized hierarchy is introduced, does this need sized_hierarchy
enabled to work? If yes, that is a potential concern. If not, these tests should remain without using the feature, while you can add the tests to under the experimental feature.
@@ -688,6 +688,10 @@ passes_rustc_lint_opt_ty = | |||
`#[rustc_lint_opt_ty]` should be applied to a struct | |||
.label = not a struct | |||
|
|||
passes_rustc_non_const_sized = | |||
attribute should be applied to a struct or enum | |||
.label = not an struct or enum |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
not a struct or enum
if !self_ty.has_non_const_sizedness() && matches!(sizedness, SizedTraitKind::Sized) { | ||
let metasized_def_id = cx.require_lang_item(TraitSolverLangItem::MetaSized); | ||
let metasized_trait_ref = ty::TraitRef::new(cx, metasized_def_id, [self_ty]); | ||
Ok(vec![metasized_trait_ref]) | ||
} else if !self_ty.has_non_const_sizedness() { | ||
// `MetaSized` has no conditionally const supertrait | ||
Ok(vec![]) | ||
} else { | ||
Err(NoSolution) | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
rewrite this into first if self_ty.has_non_const_sizedness() { return Err(NoSolution) }
, then explicitly match the sizedness
, handle the PointeeSized
variant too.
if !self_ty.has_non_const_sizedness() && matches!(sizedness, SizedTraitKind::Sized) { | ||
let metasized_def_id = tcx.require_lang_item(LangItem::MetaSized, None); | ||
let metasized_trait_ref = ty::TraitRef::new(tcx, metasized_def_id, [self_ty]); | ||
let metasized_obligation = obligation.with( | ||
tcx, | ||
ty::Binder::dummy(metasized_trait_ref) | ||
.to_host_effect_clause(tcx, obligation.predicate.constness), | ||
); | ||
Ok(thin_vec![metasized_obligation]) | ||
} else if !self_ty.has_non_const_sizedness() { | ||
// `MetaSized` has no conditionally const supertrait | ||
Ok(thin_vec![]) | ||
} else { | ||
Err(EvaluationFailure::NoSolution) | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
same as above
//@ check-pass | ||
//@ compile-flags: --crate-type=lib | ||
#![feature(const_trait_impl)] | ||
|
||
// This test can fail because of the implicit const bounds/supertraits. | ||
|
||
pub struct Bar; | ||
|
||
#[const_trait] | ||
pub trait Foo {} | ||
|
||
impl const Foo for Bar {} | ||
|
||
pub const fn bar<T: ~const Foo>() {} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
where does this test come from?
This patch implements rust-lang/rfcs#3729. It introduces two new traits to the standard library,
MetaSized
andPointeeSized
, and makesMetaSized
andSized
into const traits (relying on unstablefeature(const_trait_impl)
). See the RFC for the rationale behind these traits and to discuss whether this change makes sense in the abstract.These traits are unstable (as is their constness), so users cannot refer to them without opting-in to
feature(sized_hierarchy)
. These traits are not behindcfg
s as this would make implementation unfeasible, there would simply be too manycfg
s required to add the necessary bounds everywhere. So, likeSized
, these traits are automatically implemented by the compiler.RFC 3729 describes migrations which are necessary to preserve backwards compatibility given the introduction of these traits, which are implemented and as follows:
Sized
is rewritten asconst Sized
sized_hierarchy
feature is enabled, then an edition migration lint to rewrite the bound toconst Sized
will be emitted.Sized
will resume being the default bound.?Sized
is rewritten asconst MetaSized
sized_hierarchy
feature is enabled, then an edition migration lint to rewrite the bound toconst MetaSized
will be emitted.?Sized
will be prohibited.const MetaSized
is added as a default supertrait for all traits w/out an explicit sizedness supertrait already.sized_hierarchy
feature is enabled, then an edition migration lint to add an explicitconst MetaSized
supertrait will be emitted.const MetaSized
supertrait.Each of these migrations is not conditional on whether the item being migrated needs the migration to the stricter bound - this would be preferable but is not yet implemented (if it is possible to implement). All diagnostic output should remain the same (showing
?Sized
even if the compiler seesconst MetaSized
) unless thesized_hierarchy
feature is enabled.Due to the use of unstable extern types in the standard library and rustc, some bounds in both projects have had to be relaxed already - this is unfortunate but unavoidable so that these extern types can continue to be used where they were before. Performing these relaxations in the standard library and rustc are desirable longer-term anyway, but some bounds are not as relaxed as they ideally would be due to the inability to relax
Deref::Target
(this will be investigated separately).It is hoped that this is implemented such that it could be merged and these traits could exist "under the hood" without that being observable to the user (other than in any performance impact this has on the compiler, etc). Only once
sized_hierarchy
is stabilised would edition migration lints start to be emitted and diagnostic output show the "real" sizedness traits behind-the-scenes, rather than?Sized
. Some details might leak through due to the standard library relaxations, but this has not been observed in test output.Notes:
PointeeSized
is a different name from the RFC just to make it more obvious that it is different fromstd::ptr::Pointee
but all the names are yet to be bikeshed anyway.Fixes #79409.
r? @ghost (I'll discuss this with relevant teams to find a reviewer)