Add `rustc_fluent_macro` to decouple fluent from `rustc_macros`
Fluent, with all the icu4x it brings in, takes quite some time to compile. `fluent_messages!` is only needed in further downstream rustc crates, but is blocking more upstream crates like `rustc_index`. By splitting it out, we allow `rustc_macros` to be compiled earlier, which speeds up `x check compiler` by about 5 seconds (and even more after the needless dependency on `serde_json` is removed from `rustc_data_structures`).
Fluent, with all the icu4x it brings in, takes quite some time to
compile. `fluent_messages!` is only needed in further downstream rustc
crates, but is blocking more upstream crates like `rustc_index`. By
splitting it out, we allow `rustc_macros` to be compiled earlier, which
speeds up `x check compiler` by about 5 seconds (and even more after the
needless dependency on `serde_json` is removed from
`rustc_data_structures`).
Update `ty::VariantDef` to use `IndexVec<FieldIdx, FieldDef>`
And while doing the updates for that, also uses `FieldIdx` in `ProjectionKind::Field` and `TypeckResults::field_indices`.
There's more places that could use it (like `rustc_const_eval` and `LayoutS`), but I tried to keep this PR from exploding to *even more* places.
Part 2/? of https://github.com/rust-lang/compiler-team/issues/606
And while doing the updates for that, also uses `FieldIdx` in `ProjectionKind::Field` and `TypeckResults::field_indices`.
There's more places that could use it (like `rustc_const_eval` and `LayoutS`), but I tried to keep this PR from exploding to *even more* places.
Part 2/? of https://github.com/rust-lang/compiler-team/issues/606
Give return-position impl traits in trait a (synthetic) name to avoid name collisions with new lowering strategy
The only needed commit from this PR is the last one.
r? `@compiler-errors`
Needs #109455.
Since structs are always `VariantIdx(0)`, there's a bunch of files where the only reason they had `VariantIdx` or `vec::Idx` imported at all was to get the first variant.
So this uses a constant for that, and adds some doc-comments to `VariantIdx` while I'm there, since it doesn't have any today.
a general type system cleanup
removes the helper functions `traits::fully_solve_X` as they add more complexity then they are worth. It's confusing which of these helpers should be used in which context.
changes the way we deal with overflow to always add depth in `evaluate_predicates_recursively`. It may make sense to actually fully transition to not have `recursion_depth` on obligations but that's probably a bit too much for this PR.
also removes some other small - and imo unnecessary - helpers.
r? types
Walk un-shifted nested `impl Trait` in trait when setting up default trait method assumptions
Fixes a double subtraction in some binder math in return-position `impl Trait` in trait handling code.
Fixes#109239
Fix generics_of for impl's RPITIT synthesized associated type
The only useful commit is the last one.
This makes `generics_of` for the impl side RPITIT copy from the trait's associated type and avoid the fn on the impl side which was previously wrongly used.
This solution is better but we still need to fix resolution of the generated generics.
r? ``@compiler-errors``
Fall back to old metadata computation when type references errors
Projection is a bit too aggressive normalizing `<dyn Trait<[type error]> as Pointee>::Metadata` to `[type error]`, rather than to `DynMetadata<..>`. Side-step that by just falling back to the old structural metadata computation.
Fixes#109078
Simplify message paths
This makes it easier to open the messages file. Right now I have to first click on the `locales` dir to open it, and then on the `en-US.ftl` file. `Cargo.toml` and `build.rs` files are also in the top level, and I think there should not be more than one file, so a directory isn't really needed. The [chosen strategy for pontoon adoption](https://rust-lang.zulipchat.com/#narrow/stream/336883-i18n/topic/pontoon.20and.20next.20steps) is out of tree. Even if this descision is changed in the future, the `messages.ftl` approach is also compatible with non-english translations living in-tree, as long as the non-english translations don't live in the `compiler/rustc_foo/` directories but in different ones. That would also be helpful for grepability purposes.
The commit was the result of automated changes:
```
for p in compiler/rustc_*; do mv $p/locales/en-US.ftl $p/messages.ftl; rmdir $p/locales; done
for p in compiler/rustc_*; do sed -i "s#\.\./locales/en-US.ftl#../messages.ftl#" $p/src/lib.rs; done
```
r? `@davidtwco`
This makes it easier to open the messages file while developing on features.
The commit was the result of automatted changes:
for p in compiler/rustc_*; do mv $p/locales/en-US.ftl $p/messages.ftl; rmdir $p/locales; done
for p in compiler/rustc_*; do sed -i "s#\.\./locales/en-US.ftl#../messages.ftl#" $p/src/lib.rs; done
Do not implement HashStable for HashSet (MCP 533)
This PR removes all occurrences of `HashSet` in query results, replacing it either with `FxIndexSet` or with `UnordSet`, and then removes the `HashStable` implementation of `HashSet`. This is part of implementing [MCP 533](https://github.com/rust-lang/compiler-team/issues/533), that is, removing the `HashStable` implementations of all collection types with unstable iteration order.
The changes are mostly mechanical. The only place where additional sorting is happening is in Miri's override implementation of the `exported_symbols` query.
Account for binders correctly when adding default RPITIT method assumption
As of #108203, we install extra projection predicates into the param-env of a default trait method when it has return-position `impl Trait` (or is async).
The implementation didn't account for the fact that it's walking into and out of binders, so we just need to shift all the debruijn indices accordingly when constructing the projection predicates.
Fixes#108579
r? types
rustc_middle: Remove trait `DefIdTree`
This trait was a way to generalize over both `TyCtxt` and `Resolver`, but now `Resolver` has access to `TyCtxt`, so this trait is no longer necessary.
(This is a large commit. The changes to
`compiler/rustc_middle/src/ty/context.rs` are the most important ones.)
The current naming scheme is a mess, with a mix of `_intern_`, `intern_`
and `mk_` prefixes, with little consistency. In particular, in many
cases it's easy to use an iterator interner when a (preferable) slice
interner is available.
The guiding principles of the new naming system:
- No `_intern_` prefixes.
- The `intern_` prefix is for internal operations.
- The `mk_` prefix is for external operations.
- For cases where there is a slice interner and an iterator interner,
the former is `mk_foo` and the latter is `mk_foo_from_iter`.
Also, `slice_interners!` and `direct_interners!` can now be `pub` or
non-`pub`, which helps enforce the internal/external operations
division.
It's not perfect, but I think it's a clear improvement.
The following lists show everything that was renamed.
slice_interners
- const_list
- mk_const_list -> mk_const_list_from_iter
- intern_const_list -> mk_const_list
- substs
- mk_substs -> mk_substs_from_iter
- intern_substs -> mk_substs
- check_substs -> check_and_mk_substs (this is a weird one)
- canonical_var_infos
- intern_canonical_var_infos -> mk_canonical_var_infos
- poly_existential_predicates
- mk_poly_existential_predicates -> mk_poly_existential_predicates_from_iter
- intern_poly_existential_predicates -> mk_poly_existential_predicates
- _intern_poly_existential_predicates -> intern_poly_existential_predicates
- predicates
- mk_predicates -> mk_predicates_from_iter
- intern_predicates -> mk_predicates
- _intern_predicates -> intern_predicates
- projs
- intern_projs -> mk_projs
- place_elems
- mk_place_elems -> mk_place_elems_from_iter
- intern_place_elems -> mk_place_elems
- bound_variable_kinds
- mk_bound_variable_kinds -> mk_bound_variable_kinds_from_iter
- intern_bound_variable_kinds -> mk_bound_variable_kinds
direct_interners
- region
- intern_region (unchanged)
- const
- mk_const_internal -> intern_const
- const_allocation
- intern_const_alloc -> mk_const_alloc
- layout
- intern_layout -> mk_layout
- adt_def
- intern_adt_def -> mk_adt_def_from_data (unusual case, hard to avoid)
- alloc_adt_def(!) -> mk_adt_def
- external_constraints
- intern_external_constraints -> mk_external_constraints
Other
- type_list
- mk_type_list -> mk_type_list_from_iter
- intern_type_list -> mk_type_list
- tup
- mk_tup -> mk_tup_from_iter
- intern_tup -> mk_tup
Remove type-traversal trait aliases
#107924 moved the type traversal (folding and visiting) traits into the type library, but created trait aliases in `rustc_middle` to minimise both the API churn for trait consumers and the arising boilerplate. As mentioned in that PR, an alternative approach of defining subtraits with blanket implementations of the respective supertraits was also considered at that time but was ruled out as not adding much value.
Unfortunately, it has since emerged that rust-analyzer has difficulty with these trait aliases at present, resulting in a degraded contributor experience (see the recent [r-a has become useless](https://rust-lang.zulipchat.com/#narrow/stream/182449-t-compiler.2Fhelp/topic/r-a.20has.20become.20useless) topic on the #t-compiler/help Zulip stream).
This PR removes the trait aliases, and accordingly the underlying type library traits are now used directly; they are parameterised by `TyCtxt<'tcx>` rather than just the `'tcx` lifetime, and imports have been updated to reflect the fact that the trait aliases' explicitly named traits are no longer automatically brought into scope. These changes also roll-back the (no-longer required) workarounds to #107747 that were made in b409329c62.
Since this PR is just a find+replace together with the changes necessary for compilation & tidy to pass, it's currently just one mega-commit. Let me know if you'd like it broken up.
r? `@oli-obk`
Instead of loading the Fluent resources for every crate in
`rustc_error_messages`, each crate generates typed identifiers for its
own diagnostics and creates a static which are pulled together in the
`rustc_driver` crate and provided to the diagnostic emitter.
Signed-off-by: David Wood <david.wood@huawei.com>
Add rpitit queries
This is part of the changes we are making to lower RPITITs as an associated type. The rest of the stuff will follow under a `-Z` flag.
I still need to add comments to the code, explain stuff and also I'd need to avoid encoding in metadata when rpitit queries return `&[]`
r? `@compiler-errors`
Fix RPITITs in default trait methods (by assuming projection predicates in param-env)
Instead of having special projection logic that allows us to turn `ProjectionTy(RPITIT, [Self#0, ...])` into `OpaqueTy(RPITIT, [Self#0, ...])`, we can instead augment the param-env of default trait method bodies to assume these as projection predicates. This should allow us to only project where we're allowed to!
In order to make this work without introducing a bunch of cycle errors, we additionally tweak the `OpaqueTypeExpander` used by `ParamEnv::with_reveal_all_normalized` to not normalize the right-hand side of projection predicates. This should be fine, because if we use the projection predicate to normalize some other projection type, we'll continue to normalize the opaque that it gets projected to.
This also makes it possible to support default trait methods with RPITITs in an associated-type based RPITIT lowering strategy without too much extra effort.
Fixes#107002
Alternative to #108142
`InternIteratorElement` is a trait used to intern values produces by
iterators. There are three impls, corresponding to iterators that
produce different types:
- One for `T`, which operates straightforwardly.
- One for `Result<T, E>`, which is fallible, and will fail early with an
error result if any of the iterator elements are errors.
- One for `&'a T`, which clones the items as it iterates.
That last one is bad: it's extremely easy to use it without realizing
that it clones, which goes against Rust's normal "explicit is better"
approach to cloning.
So this commit just removes it. In practice, there weren't many use
sites. For all but one of them `into_iter()` could be used, which avoids
the need for cloning. And for the one remaining case `copied()` is
used.
There are several `mk_foo`/`intern_foo` pairs, where the former takes an
iterator and the latter takes a slice. (This naming convention is bad,
but that's a fix for another PR.)
This commit changes several `mk_foo` occurrences into `intern_foo`,
avoiding the need for some `.iter()`/`.into_iter()` calls. Affected
cases:
- mk_type_list
- mk_tup
- mk_substs
- mk_const_list
Switch to `EarlyBinder` for `type_of` query
Part of the work to finish #105779 and implement https://github.com/rust-lang/types-team/issues/78.
Several queries `X` have a `bound_X` variant that wraps the output in `EarlyBinder`. This adds `EarlyBinder` to the return type of the `type_of` query and removes `bound_type_of`.
r? `@lcnr`
Rollup of 7 pull requests
Successful merges:
- #106347 (More accurate spans for arg removal suggestion)
- #108057 (Prevent some attributes from being merged with others on reexports)
- #108090 (`if $c:expr { Some($r:expr) } else { None }` =>> `$c.then(|| $r)`)
- #108092 (note issue for feature(packed_bundled_libs))
- #108099 (use chars instead of strings where applicable)
- #108115 (Do not ICE on unmet trait alias bounds)
- #108125 (Add new people to the compiletest review rotation)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Optimize `mk_region`
PR #107869 avoiding some interning under `mk_ty` by special-casing `Ty` variants with simple (integer) bodies. This PR does something similar for regions.
r? `@compiler-errors`
Avoid accessing HIR when it can be avoided
Experiment to see if it helps some incremental cases.
Will be rebased once https://github.com/rust-lang/rust/pull/107942 gets merged.
r? `@ghost`
Much like there are specialized variants of `mk_ty`. This will enable
some optimization in the next commit.
Also rename the existing `re_error*` functions as `mk_re_error*`, for
consistency.
a placeholder type is the same as a param as they
represent "this could be any type". A bound type
represents a type inside of a `for<T>` or `exists<T>`.
When entering a forall or exists `T` should be
instantiated as a existential (inference var) or universal
(placeholder). You should never observe a bound variable
without its binder.
make &mut !Unpin not dereferenceable, and Box<!Unpin> not noalias
See https://github.com/rust-lang/unsafe-code-guidelines/issues/381 and [this LLVM discussion](https://discourse.llvm.org/t/interaction-of-noalias-and-dereferenceable/66979). The exact semantics of how `noalias` and `dereferenceable` interact are unclear, and `@comex` found a case of LLVM actually exploiting that ambiguity for optimizations. I think for now we should treat LLVM `dereferenceable` as implying a "fake read" to happen immediately at the top of the function (standing in for the spurious reads that LLVM might introduce), and that fake read is subject to all the usual `noalias` restrictions. This means we cannot put `dereferenceable` on `&mut !Unpin` references as those references can alias with other references that are being read and written inside the function (e.g. for self-referential generators), meaning the fake read introduces aliasing conflicts with those other accesses.
For `&` this is already not a problem due to https://github.com/rust-lang/rust/pull/98017 which removed the `dereferenceable` attribute for other reasons.
Regular `&mut Unpin` references are unaffected, so I hope the impact of this is going to be tiny.
The first commit does some refactoring of the `PointerKind` enum since I found the old code very confusing each time I had to touch it. It doesn't change behavior.
Fixes https://github.com/rust-lang/miri/issues/2714
EDIT: Turns out our `Box<!Unpin>` treatment was incorrect, too, so the PR also fixes that now (in codegen and Miri): we do not put `noalias` on these boxes any more.
The code that consumes PointerKind (`adjust_for_rust_scalar` in rustc_ty_utils)
ended up using PointerKind variants to talk about Rust reference types (& and
&mut) anyway, making the old code structure quite confusing: one always had to
keep in mind which PointerKind corresponds to which type. So this changes
PointerKind to directly reflect the type.
This does not change behavior.
Especially when trying to diagnose runaway future sizes, it might be
more intuitive to sort the variants according to the control flow
(aka their yield points) rather than the size of the variants.
abi: add AddressSpace field to Primitive::Pointer
...and remove it from `PointeeInfo`, which isn't meant for this.
There are still various places (marked with FIXMEs) that assume all pointers
have the same size and alignment. Fixing this requires parsing non-default
address spaces in the data layout string (and various other changes),
which will be done in a followup.
(That is, if it's actually worth it to support multiple different pointer sizes.
There is a lot of code that would be affected by that.)
Fixes#106367
r? ``@oli-obk``
cc ``@Patryk27``
use LocalDefId instead of HirId in trait resolution to simplify
the obligation clause resolution
Signed-off-by: Vincenzo Palazzo <vincenzopalazzodev@gmail.com>
...and remove it from `PointeeInfo`, which isn't meant for this.
There are still various places (marked with FIXMEs) that assume all pointers
have the same size and alignment. Fixing this requires parsing non-default
address spaces in the data layout string, which will be done in a followup.
- Eliminates all the `get_context` calls that async lowering created.
- Replace all `Local` `ResumeTy` types with `&mut Context<'_>`.
The `Local`s that have their types replaced are:
- The `resume` argument itself.
- The argument to `get_context`.
- The yielded value of a `yield`.
The `ResumeTy` hides a `&mut Context<'_>` behind an unsafe raw pointer, and the
`get_context` function is being used to convert that back to a `&mut Context<'_>`.
Ideally the async lowering would not use the `ResumeTy`/`get_context` indirection,
but rather directly use `&mut Context<'_>`, however that would currently
lead to higher-kinded lifetime errors.
See <https://github.com/rust-lang/rust/issues/105501>.
The async lowering step and the type / lifetime inference / checking are
still using the `ResumeTy` indirection for the time being, and that indirection
is removed here. After this transform, the generator body only knows about `&mut Context<'_>`.
Put `noundef` on all scalars that don't allow uninit
Previously, it was only put on scalars with range validity invariants like bool, was uninit was obviously invalid for those.
Since then, we have normatively declared all uninit primitives to be undefined behavior and can therefore put `noundef` on them.
The remaining concern was the `mem::uninitialized` function, which cause quite a lot of UB in the older parts of the ecosystem. After #99182, this function now doesn't return uninit values anymore, making users of it safe from this change.
The only real sources of UB where people could encounter uninit primitives are `MaybeUninit::uninit().assume_init()`, which has always be clear in the docs about being UB and from heap allocations (like reading from the spare capacity of a vec). This is hopefully rare enough to not break anything.
cc `@nagisa` `@scottmcm` `@nikic`
Previously, it was only put on scalars with range validity invariants
like bool, was uninit was obviously invalid for those.
Since then, we have normatively declared all uninit primitives to be
undefined behavior and can therefore put `noundef` on them.
The remaining concern was the `mem::uninitialized` function, which cause
quite a lot of UB in the older parts of the ecosystem. This function now
doesn't return uninit values anymore, making users of it safe from this
change.
The only real sources of UB where people could encounter uninit
primitives are `MaybeUninit::uninit().assume_init()`, which has always
be clear in the docs about being UB and from heap allocations (like
reading from the spare capacity of a vec. This is hopefully rare enough
to not break anything.
Rename `hir::Map::{get_,find_}parent_node` to `hir::Map::{,opt_}parent_id`, and add `hir::Map::{get,find}_parent`
The `hir::Map::get_parent_node` function doesn't return a `Node`, and I think that's quite confusing. Let's rename it to something that sounds more like something that gets the parent hir id => `hir::Map::parent_id`. Same with `find_parent_node` => `opt_parent_id`.
Also, combine `hir.get(hir.parent_id(hir_id))` and similar `hir.find(hir.parent_id(hir_id))` function into new functions that actually retrieve the parent node in one call. This last commit is the only one that might need to be looked at closely.
do not add noalias in return position
`noalias` as a return attribute in LLVM indicates that the returned pointer does not alias anything else that is reachable from the caller, *including things reachable before this function call*. This is clearly not the case with a function like `fn id(Box<T>) -> Box<T>`, so we cannot use this attribute.
Fixes https://github.com/rust-lang/unsafe-code-guidelines/issues/385 (including an actual miscompilation that `@comex` managed to produce).
Fix Async Generator ABI
This change was missed when making async generators implement `Future` directly.
It did not cause any problems in codegen so far, as `GeneratorState<(), Output>`
happens to have the same ABI as `Poll<Output>`.
This change was missed when making async generators implement `Future` directly.
It did not cause any problems in codegen so far, as `GeneratorState<(), Output>`
happens to have the same ABI as `Poll<Output>`.
`mk_const(ty::ConstKind::X(...), ty)` can now be simplified to
`mk_cosnt(..., ty)`.
I searched with the following regex: \mk_const\([\n\s]*(ty::)?ConstKind\
I've left `ty::ConstKind::{Bound, Error}` as-is, they seem clearer this
way.
stricter alignment enforcement for ScalarPair
`@eddyb` [indicated](https://github.com/rust-lang/rust/pull/103926#discussion_r1033315005) that alignment violating this check might be a bug. So let's see what the test suite says.
(Only the 2nd commit actually changes behavior... but I couldn't not do that other cleanup.^^)
Does the PR CI runner even enable debug assertions though...?
Add `ConstKind::Expr`
Starting to implement `ty::ConstKind::Abstract`, most of the match cases are stubbed out, some I was unsure what to add, others I didn't want to add until a more complete implementation was ready.
r? `@lcnr`
Initial pass at expr/abstract const/s
Address comments
Switch to using a list instead of &[ty::Const], rm `AbstractConst`
Remove try_unify_abstract_consts
Update comments
Add edits
Recurse more
More edits
Prevent equating associated consts
Move failing test to ui
Changes this test from incremental to ui, and mark it as failing and a known bug.
Does not cause the compiler to ICE, so should be ok.
Make rustc_target usable outside of rustc
I'm working on showing type size in rust-analyzer (https://github.com/rust-lang/rust-analyzer/pull/13490) and I currently copied rustc code inside rust-analyzer, which works, but is bad. With this change, I would become able to use `rustc_target` and `rustc_index` directly in r-a, reducing the amount of copy needed.
This PR contains some feature flag to put nightly features behind them to make crates buildable on the stable compiler + makes layout related types generic over index type + removes interning of nested layouts.
Previously, async constructs would be lowered to "normal" generators,
with an additional `from_generator` / `GenFuture` shim in between to
convert from `Generator` to `Future`.
The compiler will now special-case these generators internally so that
async constructs will *directly* implement `Future` without the need
to go through the `from_generator` / `GenFuture` shim.
The primary motivation for this change was hiding this implementation
detail in stack traces and debuginfo, but it can in theory also help
the optimizer as there is less abstractions to see through.
Support using `Self` or projections inside an RPIT/async fn
I reuse the same idea as https://github.com/rust-lang/rust/pull/103449 to use variances to encode whether a lifetime parameter is captured by impl-trait.
The current implementation of async and RPIT replace all lifetimes from the parent generics by `'static`. This PR changes the scheme
```rust
impl<'a> Foo<'a> {
fn foo<'b, T>() -> impl Into<Self> + 'b { ... }
}
opaque Foo::<'_a>::foo::<'_b, T>::opaque<'b>: Into<Foo<'_a>> + 'b;
impl<'a> Foo<'a> {
// OLD
fn foo<'b, T>() -> Foo::<'static>::foo::<'static, T>::opaque::<'b> { ... }
^^^^^^^ the `Self` becomes `Foo<'static>`
// NEW
fn foo<'b, T>() -> Foo::<'a>::foo::<'b, T>::opaque::<'b> { ... }
^^ the `Self` stays `Foo<'a>`
}
```
There is the same issue with projections. In the example, substitute `Self` by `<T as Trait<'b>>::Assoc` in the sugared version, and `Foo<'_a>` by `<T as Trait<'_b>>::Assoc` in the desugared one.
This allows to support `Self` in impl-trait, since we do not replace lifetimes by `'static` any more. The same trick allows to use projections like `T::Assoc` where `Self` is allowed. The feature is gated behind a `impl_trait_projections` feature gate.
The implementation relies on 2 tweaking rules for opaques in 2 places:
- we only relate substs that correspond to captured lifetimes during TypeRelation;
- we only list captured lifetimes in choice region computation.
For simplicity, I encoded the "capturedness" of lifetimes as a variance, `Bivariant` vs `Invariant` for unused vs captured lifetimes. The `variances_of` query used to ICE for opaques.
Impl-trait that do not reference `Self` or projections will have their variances as:
- `o` (invariant) for each parent type or const;
- `*` (bivariant) for each parent lifetime --> will not participate in borrowck;
- `o` (invariant) for each own lifetime.
Impl-trait that does reference `Self` and/or projections will have some parent lifetimes marked as `o` (as the example above), and participate in type relation and borrowck. In the example above, `variances_of(opaque) = ['_a: o, '_b: *, T: o, 'b: o]`.
r? types
cc `@compiler-errors` , as you asked about the issue with `Self` and projections.
Accept `TyCtxt` instead of `TyCtxtAt` in `Ty::is_*` functions
Functions in answer:
- `Ty::is_freeze`
- `Ty::is_sized`
- `Ty::is_unpin`
- `Ty::is_copy_modulo_regions`
This allows to remove a lot of useless `.at(DUMMY_SP)`, making the code a bit nicer :3
r? `@compiler-errors`
spastorino noticed some silly expressions like `item_id.def_id.def_id`.
This commit renames several `def_id: OwnerId` fields as `owner_id`, so
those expressions become `item_id.owner_id.def_id`.
`item_id.owner_id.local_def_id` would be even clearer, but the use of
`def_id` for values of type `LocalDefId` is *very* widespread, so I left
that alone.
indirect immutable freeze by-value function parameters.
Right now, `rustc` only examines function signatures and the platform ABI when
determining the LLVM attributes to apply to parameters. This results in missed
optimizations, because there are some attributes that can be determined via
analysis of the MIR making up the function body. In particular, `readonly`
could be applied to most indirectly-passed by-value function arguments
(specifically, those that are freeze and are observed not to be mutated), but
it currently is not.
This patch introduces the machinery that allows `rustc` to determine those
attributes. It consists of a query, `deduced_param_attrs`, that, when
evaluated, analyzes the MIR of the function to determine supplementary
attributes. The results of this query for each function are written into the
crate metadata so that the deduced parameter attributes can be applied to
cross-crate functions. In this patch, we simply check the parameter for
mutations to determine whether the `readonly` attribute should be applied to
parameters that are indirect immutable freeze by-value. More attributes could
conceivably be deduced in the future: `nocapture` and `noalias` come to mind.
Adding `readonly` to indirect function parameters where applicable enables some
potential optimizations in LLVM that are discussed in [issue 103103] and [PR
103070] around avoiding stack-to-stack memory copies that appear in functions
like `core::fmt::Write::write_fmt` and `core::panicking::assert_failed`. These
functions pass a large structure unchanged by value to a subfunction that also
doesn't mutate it. Since the structure in this case is passed as an indirect
parameter, it's a pointer from LLVM's perspective. As a result, the
intermediate copy of the structure that our codegen emits could be optimized
away by LLVM's MemCpyOptimizer if it knew that the pointer is `readonly
nocapture noalias` in both the caller and callee. We already pass `nocapture
noalias`, but we're missing `readonly`, as we can't determine whether a
by-value parameter is mutated by examining the signature in Rust. I didn't have
much success with having LLVM infer the `readonly` attribute, even with fat
LTO; it seems that deducing it at the MIR level is necessary.
No large benefits should be expected from this optimization *now*; LLVM needs
some changes (discussed in [PR 103070]) to more aggressively use the `noalias
nocapture readonly` combination in its alias analysis. I have some LLVM patches
for these optimizations and have had them looked over. With all the patches
applied locally, I enabled LLVM to remove all the `memcpy`s from the following
code:
```rust
fn main() {
println!("Hello {}", 3);
}
```
which is a significant codegen improvement over the status quo. I expect that
if this optimization kicks in in multiple places even for such a simple
program, then it will apply to Rust code all over the place.
[issue 103103]: https://github.com/rust-lang/rust/issues/103103
[PR 103070]: https://github.com/rust-lang/rust/pull/103070
Unify `tcx.constness` query and param env constness checks
The checks that we do in the `constness` query seem inconsistent with the checks that we do to determine if an item's param-env is const, so I merged them into the `constness` query and call that from the `param_env` query.
I'm not sure if this totally makes sense -- is there a case where `tcx.param_env()` would return a const param-env for an item whose `tcx.constness()` is `Constness::NotConst`? Because if not, it seems a bit dangerous that these two differ.
Luckily, not many places actually use `tcx.constness()`, and the checks in `tcx.param_env()` seem stricter than the checks in `tcx.constness()` (at least for the types of items we type-check).
Also, due to the way that `tcx.param_env()` is implemented, it _never_ used to return a const param-env for a item coming from a different crate, which also seems dangerous (though also probably not weaponizable currently, because we seldom actually compute the param-env for a non-local item).
rename `ImplItemKind::TyAlias` to `ImplItemKind::Type`
The naming of this variant seems inconsistent given that this is not really a "type alias", and the associated type variant for `TraitItemKind` is just called `Type`.
