Suggest `impl Trait` return type when incorrectly using a generic return type
Address #85991
When there is a type mismatch error and the return type is generic, and that generic parameter is not used in the function parameters, suggest replacing that generic with the `impl Trait` syntax.
r? `@estebank`
Address #85991
Suggest the `impl Trait` return type syntax if the user tried to return a generic parameter and we get a type mismatch
The suggestion is not emitted if the param appears in the function parameters, and only get the bounds that actually involve `T: ` directly
It also checks whether the generic param is contained in any where bound (where it isn't the self type), and if one is found (like `Option<T>: Send`), it is not suggested.
This also adds `TyS::contains`, which recursively vistits the type and looks if the other type is contained anywhere
Suggest copying trait associated type bounds on lifetime error
Closes#92033
Kind of the most simple suggestion to make - we don't try to be fancy. Turns out, it's still pretty useful (the couple existing tests that trigger this error end up fixed - for this error - upon applying the fix).
r? ``@estebank``
cc ``@nikomatsakis``
Improve comments about type folding/visiting.
I have found this code confusing for years. I've always roughly
understood it, but never exactly. I just made my fourth(?) attempt and
finally cracked it.
This commit improves the comments. In particular, it explicitly
describes how you can't do a custom fold/visit of any type; there are
actually a handful of "types of interest" (e.g. `Ty`, `Predicate`,
`Region`, `Const`) that can be custom folded/visted, and all other types
just get a generic traversal. I think this was the part that eluded me
on all my prior attempts at understanding.
The commit also updates comments to account for some newer changes such
as the fallible/infallible folding distinction, does some minor
reorderings, and moves one `impl` to a better place.
r? `@BoxyUwU`
I have found this code confusing for years. I've always roughly
understood it, but never exactly. I just made my fourth(?) attempt and
finally cracked it.
This commit improves the comments. In particular, it explicitly
describes how you can't do a custom fold/visit of any type; there are
actually a handful of "types of interest" (e.g. `Ty`, `Predicate`,
`Region`, `Const`) that can be custom folded/visted, and all other types
just get a generic traversal. I think this was the part that eluded me
on all my prior attempts at understanding.
The commit also updates comments to account for some newer changes such
as the fallible/infallible folding distinction, does some minor
reorderings, and moves one `impl` to a better place.
Fix inconsistent symbol mangling with -Zverbose
Always skip arguments that are the defaults of their respective
parameters, to avoid generating inconsistent symbols for builds
with `-Zverbose` flag and without it.
Support pretty printing of invalid constants
Make it possible to pretty print invalid constants by introducing a
fallible variant of `destructure_const` and falling back to debug
formatting when it fails.
Closes#93688.
Treat static refs as `mir::ConstantKind::Val`
With the upcoming introduction of Valtrees we want to treat more values as `mir::ConstantKind::Val` directly.
r? `@lcnr`
cc `@oli-obk`
Always skip arguments that are the defaults of their respective
parameters, to avoid generating inconsistent symbols for builds
with `-Zverbose` flag and without it.
Make it possible to pretty print invalid constants by introducing a
fallible variant of `destructure_const` and falling back to debug
formatting when it fails.
Specifically, rename the `Const` struct as `ConstS` and re-introduce `Const` as
this:
```
pub struct Const<'tcx>(&'tcx Interned<ConstS>);
```
This now matches `Ty` and `Predicate` more closely, including using
pointer-based `eq` and `hash`.
Notable changes:
- `mk_const` now takes a `ConstS`.
- `Const` was copy, despite being 48 bytes. Now `ConstS` is not, so need a
we need separate arena for it, because we can't use the `Dropless` one any
more.
- Many `&'tcx Const<'tcx>`/`&Const<'tcx>` to `Const<'tcx>` changes
- Many `ct.ty` to `ct.ty()` and `ct.val` to `ct.val()` changes.
- Lots of tedious sigil fiddling.
The variant names are exported, so we can use them directly (possibly
with a `ty::` qualifier). Lots of places already do this, this commit
just increases consistency.
Specifically, change `Region` from this:
```
pub type Region<'tcx> = &'tcx RegionKind;
```
to this:
```
pub struct Region<'tcx>(&'tcx Interned<RegionKind>);
```
This now matches `Ty` and `Predicate` more closely.
Things to note
- Regions have always been interned, but we haven't been using pointer-based
`Eq` and `Hash`. This is now happening.
- I chose to impl `Deref` for `Region` because it makes pattern matching a lot
nicer, and `Region` can be viewed as just a smart wrapper for `RegionKind`.
- Various methods are moved from `RegionKind` to `Region`.
- There is a lot of tedious sigil changes.
- A couple of types like `HighlightBuilder`, `RegionHighlightMode` now have a
`'tcx` lifetime because they hold a `Ty<'tcx>`, so they can call `mk_region`.
- A couple of test outputs change slightly, I'm not sure why, but the new
outputs are a little better.
Specifically, change `Ty` from this:
```
pub struct Predicate<'tcx> { inner: &'tcx PredicateInner<'tcx> }
```
to this:
```
pub struct Predicate<'tcx>(&'tcx Interned<PredicateS<'tcx>>)
```
where `PredicateInner` is renamed as `PredicateS`.
This (plus a few other minor changes) makes the parallels with `Ty` and
`TyS` much clearer, and makes the uniqueness more explicit.
Specifically, change `Ty` from this:
```
pub type Ty<'tcx> = &'tcx TyS<'tcx>;
```
to this
```
pub struct Ty<'tcx>(Interned<'tcx, TyS<'tcx>>);
```
There are two benefits to this.
- It's now a first class type, so we can define methods on it. This
means we can move a lot of methods away from `TyS`, leaving `TyS` as a
barely-used type, which is appropriate given that it's not meant to
be used directly.
- The uniqueness requirement is now explicit, via the `Interned` type.
E.g. the pointer-based `Eq` and `Hash` comes from `Interned`, rather
than via `TyS`, which wasn't obvious at all.
Much of this commit is boring churn. The interesting changes are in
these files:
- compiler/rustc_middle/src/arena.rs
- compiler/rustc_middle/src/mir/visit.rs
- compiler/rustc_middle/src/ty/context.rs
- compiler/rustc_middle/src/ty/mod.rs
Specifically:
- Most mentions of `TyS` are removed. It's very much a dumb struct now;
`Ty` has all the smarts.
- `TyS` now has `crate` visibility instead of `pub`.
- `TyS::make_for_test` is removed in favour of the static `BOOL_TY`,
which just works better with the new structure.
- The `Eq`/`Ord`/`Hash` impls are removed from `TyS`. `Interned`s impls
of `Eq`/`Hash` now suffice. `Ord` is now partly on `Interned`
(pointer-based, for the `Equal` case) and partly on `TyS`
(contents-based, for the other cases).
- There are many tedious sigil adjustments, i.e. adding or removing `*`
or `&`. They seem to be unavoidable.
make `find_similar_impl_candidates` even fuzzier
continues the good work of `@BGR360` in #92223. I might have overshot a bit and we're now slightly too fuzzy 😅
with this we can now also simplify `simplify_type`, which is nice :3
Make `Res::SelfTy` a struct variant and update docs
I found pattern matching on a `(Option<DefId>, Option<(DefId, bool)>)` to not be super readable, additionally the doc comments on the types in a tuple variant aren't visible anywhere at use sites as far as I can tell (using rust analyzer + vscode)
The docs incorrectly assumed that the `DefId` in `Option<(DefId, bool)>` would only ever be for an impl item and I also found the code examples to be somewhat unclear about which `DefId` was being talked about.
r? `@lcnr` since you reviewed the last PR changing these docs
Improve chalk integration
- Support subtype bounds in chalk lowering
- Handle universes in canonicalization
- Handle type parameters in chalk responses
- Use `chalk_ir::LifetimeData::Empty` for `ty::ReEmpty`
- Remove `ignore-compare-mode-chalk` for tests that no longer hang (they may still fail or ICE)
This is enough to get a hello world program to compile with `-Zchalk` now. Some of the remaining issues that are needed to get Chalk integration working on larger programs are:
- rust-lang/chalk#234
- rust-lang/chalk#548
- rust-lang/chalk#734
- Generators are handled differently in chalk and rustc
r? `@jackh726`
Apply noundef attribute to &T, &mut T, Box<T>, bool
This doesn't handle `char` because it's a bit awkward to distinguish it from `u32` at this point in codegen.
Note that this _does not_ change whether or not it is UB for `&`, `&mut`, or `Box` to point to undef. It only applies to the pointer itself, not the pointed-to memory.
Fixes (partially) #74378.
r? `@nikic` cc `@RalfJung`
This is required to avoid creating large numbers of universes from each
Chalk query, while still having enough universe information for lifetime
errors.
Make all `hir::Map` methods consistently by-value
`hir::Map` only consists of a single reference (as part of the contained `TyCtxt`) anyways, so copying is literally zero overhead compared to passing a reference
Ensure that queries only return Copy types.
This should pervent the perf footgun of returning a result with an expensive `Clone` impl (like a `Vec` of a hash map).
I went for the stupid solution of allocating on an arena everything that was not `Copy`. Some query results could be made Copy easily, but I did not really investigate.
Refactor query system to maintain a global job id counter
This replaces the per-shard counters with a single global counter, simplifying
the JobId struct down to just a u64 and removing the need to pipe a DepKind
generic through a bunch of code. The performance implications on non-parallel
compilers are likely minimal (this switches to `Cell<u64>` as the backing
storage over a `u64`, but the latter was already inside a `RefCell` so it's not
really a significance divergence). On parallel compilers, the cost of a single
global u64 counter may be more significant: it adds a serialization point in
theory. On the other hand, we can imagine changing the counter to have a
thread-local component if it becomes worrisome or some similar structure.
The new design is sufficiently simpler that it warrants the potential for slight
changes down the line if/when we get parallel compilation to be more of a
default.
A u64 counter, instead of u32 (the old per-shard width), is chosen to avoid
possibly overflowing it and causing problems; it is effectively impossible that
we would overflow a u64 counter in this context.
This replaces the per-shard counters with a single global counter, simplifying
the JobId struct down to just a u64 and removing the need to pipe a DepKind
generic through a bunch of code. The performance implications on non-parallel
compilers are likely minimal (this switches to `Cell<u64>` as the backing
storage over a `u64`, but the latter was already inside a `RefCell` so it's not
really a significance divergence). On parallel compilers, the cost of a single
global u64 counter may be more significant: it adds a serialization point in
theory. On the other hand, we can imagine changing the counter to have a
thread-local component if it becomes worrisome or some similar structure.
The new design is sufficiently simpler that it warrants the potential for slight
changes down the line if/when we get parallel compilation to be more of a
default.
A u64 counter, instead of u32 (the old per-shard width), is chosen to avoid
possibly overflowing it and causing problems; it is effectively impossible that
we would overflow a u64 counter in this context.
Add more *-unwind ABI variants
The following *-unwind ABIs are now supported:
- "C-unwind"
- "cdecl-unwind"
- "stdcall-unwind"
- "fastcall-unwind"
- "vectorcall-unwind"
- "thiscall-unwind"
- "aapcs-unwind"
- "win64-unwind"
- "sysv64-unwind"
- "system-unwind"
cc `@rust-lang/wg-ffi-unwind`
Lazy type-alias-impl-trait
Previously opaque types were processed by
1. replacing all mentions of them with inference variables
2. memorizing these inference variables in a side-table
3. at the end of typeck, resolve the inference variables in the side table and use the resolved type as the hidden type of the opaque type
This worked okayish for `impl Trait` in return position, but required lots of roundabout type inference hacks and processing.
This PR instead stops this process of replacing opaque types with inference variables, and just keeps the opaque types around.
Whenever an opaque type `O` is compared with another type `T`, we make the comparison succeed and record `T` as the hidden type. If `O` is compared to `U` while there is a recorded hidden type for it, we grab the recorded type (`T`) and compare that against `U`. This makes implementing
* https://github.com/rust-lang/rfcs/pull/2515
much simpler (previous attempts on the inference based scheme were very prone to ICEs and general misbehaviour that was not explainable except by random implementation defined oddities).
r? `@nikomatsakis`
fixes#93411fixes#88236
use `fold_list` in `try_super_fold_with` for `SubstsRef`
split out from #93505 as this by itself is responsible for most of the perf improvements there
r? `@michaelwoerister`
This doesn't handle `char` because it's a bit awkward to distinguish it
from u32 at this point in codegen.
Note that for some types (like `&Struct` and `&mut Struct`),
we already apply `dereferenceable`, which implies `noundef`,
so the IR does not change.
Stabilize `-Z instrument-coverage` as `-C instrument-coverage`
(Tracking issue for `instrument-coverage`: https://github.com/rust-lang/rust/issues/79121)
This PR stabilizes support for instrumentation-based code coverage, previously provided via the `-Z instrument-coverage` option. (Continue supporting `-Z instrument-coverage` for compatibility for now, but show a deprecation warning for it.)
Many, many people have tested this support, and there are numerous reports of it working as expected.
Move the documentation from the unstable book to stable rustc documentation. Update uses and documentation to use the `-C` option.
Addressing questions raised in the tracking issue:
> If/when stabilized, will the compiler flag be updated to -C instrument-coverage? (If so, the -Z variant could also be supported for some time, to ease migrations for existing users and scripts.)
This stabilization PR updates the option to `-C` and keeps the `-Z` variant to ease migration.
> The Rust coverage implementation depends on (and automatically turns on) -Z symbol-mangling-version=v0. Will stabilizing this feature depend on stabilizing v0 symbol-mangling first? If so, what is the current status and timeline?
This stabilization PR depends on https://github.com/rust-lang/rust/pull/90128 , which stabilizes `-C symbol-mangling-version=v0` (but does not change the default symbol-mangling-version).
> The Rust coverage implementation implements the latest version of LLVM's Coverage Mapping Format (version 4), which forces a dependency on LLVM 11 or later. A compiler error is generated if attempting to compile with coverage, and using an older version of LLVM.
Given that LLVM 13 has now been released, requiring LLVM 11 for coverage support seems like a reasonable requirement. If people don't have at least LLVM 11, nothing else breaks; they just can't use coverage support. Given that coverage support currently requires a nightly compiler and LLVM 11 or newer, allowing it on a stable compiler built with LLVM 11 or newer seems like an improvement.
The [tracking issue](https://github.com/rust-lang/rust/issues/79121) and the [issue label A-code-coverage](https://github.com/rust-lang/rust/labels/A-code-coverage) link to a few open issues related to `instrument-coverage`, but none of them seem like showstoppers. All of them seem like improvements and refinements we can make after stabilization.
The original `-Z instrument-coverage` support went through a compiler-team MCP at https://github.com/rust-lang/compiler-team/issues/278 . Based on that, `@pnkfelix` suggested that this needed a stabilization PR and a compiler-team FCP.
Fix ret > 1 bound if shadowed by const
Prior to a change, it would only look at types in bounds. When it started looking for consts,
shadowing type variables with a const would cause an ICE, so now defer looking at consts only if
there are no types present.
cc ``````@compiler-errors``````
Should Fix#93553
Prior to a change, it would only look at types in bounds. When it started looking for consts,
shadowing type variables with a const would cause an ICE, so now defer looking at consts only if
there are no types present.
Temporary fix for the layout of aligned enums
Fix for the issue #92464
~~I was after this issue for quite some time now, I have a temporary fix for it.
I think the current problem is [here](e75f96763f/compiler/rustc_middle/src/ty/layout.rs (L1305-L1310)) created `tag` value might be wrong, because when I checked `min` and `max` values it's always between 0..1, which results in wrong size comparison in a few lines down below.
I think `min` and `max` values don't take `#[repr(aligned(8))]` into consideration and just act from base values assigned inside the enum. If what I am saying is true, aligned enums were created with the wrong layout for some time.~~
~~As stated in the title this is only a temporary fix and I think this needs further investigation, if someone wants to mentor it I would like to work on that too.~~ 😸
**Edit: Weird some tests fail now going to close this for now...**
**Edit2: I made it work again.**
I think I figured out the main problem of the issue, layout types of aligned enums with custom discriminant types were not handled, which resulted in confusing(such as this issue) behavior down the line, this is a kinda hacky fix for the issue.
by using an opaque type obligation to bubble up comparisons between opaque types and other types
Also uses proper obligation causes so that the body id works, because out of some reason nll uses body ids for logic instead of just diagnostics.
Return an indexmap in `all_local_trait_impls` query
The data structure previously used here required that `DefId` be `Ord`. As part of #90317, we do not want `DefId` to implement `Ord`.
Fix two incorrect "it's" (typos in comments)
Found one of these while reading the documentation online. The other came up because it's in the same file.
Make dead code check a query.
Dead code check is run for each invocation of the compiler, even if no modifications were involved.
This PR makes dead code check a query keyed on the module. This allows to skip the check when a module has not changed.
To perform this, a query `live_symbols_and_ignored_derived_traits` is introduced to encapsulate the global analysis of finding live symbols. The second query `check_mod_deathness` outputs diagnostics for each module based on this first query's results.
Continue work on associated const equality
This actually implements some more complex logic for assigning associated consts to values.
Inside of projection candidates, it now defers to a separate function for either consts or
types. To reduce amount of code, projections are now generic over T, where T is either a Type or
a Const. I can add some comments back later, but this was the fastest way to implement it.
It also now finds the correct type of consts in type_of.
---
The current main TODO is finding the const of the def id for the LeafDef.
Right now it works if the function isn't called, but once you use the trait impl with the bound it fails inside projection.
I was hoping to get some help in getting the `&'tcx ty::Const<'tcx>`, in addition to a bunch of other `todo!()`s which I think may not be hit.
r? `@oli-obk`
Updates #92827
rustc_errors: only box the `diagnostic` field in `DiagnosticBuilder`.
I happened to need to do the first change (replacing `allow_suggestions` with equivalent functionality on `Diagnostic` itself) as part of a larger change, and noticed that there's only two fields left in `DiagnosticBuilderInner`.
So with this PR, instead of a single pointer, `DiagnosticBuilder` is two pointers, which should work just as well for passing *it* by value (and may even work better wrt some operations, though probably not by much).
But anything that was already taking advantage of `DiagnosticBuilder` being a single pointer, and wrapping it further (e.g. `Result<T, DiagnosticBuilder>` w/ non-ZST `T`), ~~will probably see a slowdown~~, so I want to do a perf run before even trying to propose this.
Store def_id_to_hir_id as variant in hir_owner.
If hir_owner is Owner(_), the LocalDefId is pointing to an owner, so the ItemLocalId is 0.
If the HIR node does not exist, we store Phantom.
Otherwise, we store the HirId associated to the LocalDefId.
Related to #89278
r? `@oli-obk`
Add note suggesting that predicate may be satisfied, but is not `const`
Not sure if we should be printing this in addition to, or perhaps _instead_ of the help message:
```
help: the trait `~const Add` is not implemented for `NonConstAdd`
```
Also added `ParamEnv::is_const` and `PolyTraitPredicate::is_const_if_const` and, in a separate commit, used those in other places instead of `== hir::Constness::Const`, etc.
r? ````@fee1-dead````
If hir_owner is Owner(_), the LocalDefId is pointing to an owner, so the ItemLocalId is 0.
If the HIR node does not exist, we store Phantom.
Otherwise, we store the HirId associated to the LocalDefId.
Store a `Symbol` instead of an `Ident` in `AssocItem`
This is the same idea as #92533, but for `AssocItem` instead
of `VariantDef`/`FieldDef`.
With this change, we no longer have any uses of
`#[stable_hasher(project(...))]`
Check `const Drop` impls considering `~const` Bounds
This PR adds logic to trait selection to account for `~const` bounds in custom `impl const Drop` for types, elaborates the `const Drop` check in `rustc_const_eval` to check those bounds, and steals some drop linting fixes from #92922, thanks `@DrMeepster.`
r? `@fee1-dead` `@oli-obk` <sup>(edit: guess I can't request review from two people, lol)</sup>
since each of you wrote and reviewed #88558, respectively.
Since the logic here is more complicated than what existed, it's possible that this is a perf regression. But it works correctly with tests, and that makes me happy.
Fixes#92881
rustc_lint: Some early linting refactorings
The first one removes and renames some fields and methods from `EarlyContext`.
The second one uses the set of registered tools (for tool attributes and tool lints) in a more centralized way.
The third one removes creation of a fake `ast::Crate` from `fn pre_expansion_lint`.
Pre-expansion linting is done with per-module granularity on freshly loaded modules, and it previously synthesized an `ast::Crate` to visit non-root modules, now they are visited as modules.
The node ID used for pre-expansion linting is also made more precise (the loaded module ID is used).
Make `Decodable` and `Decoder` infallible.
`Decoder` has two impls:
- opaque: this impl is already partly infallible, i.e. in some places it
currently panics on failure (e.g. if the input is too short, or on a
bad `Result` discriminant), and in some places it returns an error
(e.g. on a bad `Option` discriminant). The number of places where
either happens is surprisingly small, just because the binary
representation has very little redundancy and a lot of input reading
can occur even on malformed data.
- json: this impl is fully fallible, but it's only used (a) for the
`.rlink` file production, and there's a `FIXME` comment suggesting it
should change to a binary format, and (b) in a few tests in
non-fundamental ways. Indeed #85993 is open to remove it entirely.
And the top-level places in the compiler that call into decoding just
abort on error anyway. So the fallibility is providing little value, and
getting rid of it leads to some non-trivial performance improvements.
Much of this PR is pretty boring and mechanical. Some notes about
a few interesting parts:
- The commit removes `Decoder::{Error,error}`.
- `InternIteratorElement::intern_with`: the impl for `T` now has the same
optimization for small counts that the impl for `Result<T, E>` has,
because it's now much hotter.
- Decodable impls for SmallVec, LinkedList, VecDeque now all use
`collect`, which is nice; the one for `Vec` uses unsafe code, because
that gave better perf on some benchmarks.
r? `@bjorn3`
`Decoder` has two impls:
- opaque: this impl is already partly infallible, i.e. in some places it
currently panics on failure (e.g. if the input is too short, or on a
bad `Result` discriminant), and in some places it returns an error
(e.g. on a bad `Option` discriminant). The number of places where
either happens is surprisingly small, just because the binary
representation has very little redundancy and a lot of input reading
can occur even on malformed data.
- json: this impl is fully fallible, but it's only used (a) for the
`.rlink` file production, and there's a `FIXME` comment suggesting it
should change to a binary format, and (b) in a few tests in
non-fundamental ways. Indeed #85993 is open to remove it entirely.
And the top-level places in the compiler that call into decoding just
abort on error anyway. So the fallibility is providing little value, and
getting rid of it leads to some non-trivial performance improvements.
Much of this commit is pretty boring and mechanical. Some notes about
a few interesting parts:
- The commit removes `Decoder::{Error,error}`.
- `InternIteratorElement::intern_with`: the impl for `T` now has the same
optimization for small counts that the impl for `Result<T, E>` has,
because it's now much hotter.
- Decodable impls for SmallVec, LinkedList, VecDeque now all use
`collect`, which is nice; the one for `Vec` uses unsafe code, because
that gave better perf on some benchmarks.
I have found this code very confusing at times. This commit clarifies
things.
In particular, the commit explains the requirements that the `Borrow`
impls put on the `Eq` and `Hash` impls, which are non-obvious. And it
puts the `Borrow` impls first, since they force `Eq` and `Hash` to have
particular forms.
The commit also notes `TyS`'s uniqueness requirements.
Rollup of 17 pull requests
Successful merges:
- #91032 (Introduce drop range tracking to generator interior analysis)
- #92856 (Exclude "test" from doc_auto_cfg)
- #92860 (Fix errors on blanket impls by ignoring the children of generated impls)
- #93038 (Fix star handling in block doc comments)
- #93061 (Only suggest adding `!` to expressions that can be macro invocation)
- #93067 (rustdoc mobile: fix scroll offset when jumping to internal id)
- #93086 (Add tests to ensure that `let_chains` works with `if_let_guard`)
- #93087 (Fix src/test/run-make/raw-dylib-alt-calling-convention)
- #93091 (⬆ chalk to 0.76.0)
- #93094 (src/test/rustdoc-json: Check for `struct_field`s in `variant_tuple_struct.rs`)
- #93098 (Show a more informative panic message when `DefPathHash` does not exist)
- #93099 (rustdoc: auto create output directory when "--output-format json")
- #93102 (Pretty printer algorithm revamp step 3)
- #93104 (Support --bless for pp-exact pretty printer tests)
- #93114 (update comment for `ensure_monomorphic_enough`)
- #93128 (Add script to prevent point releases with same number as existing ones)
- #93136 (Backport the 1.58.1 release notes to master)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Show a more informative panic message when `DefPathHash` does not exist
This should hopefully make it easier to debug incremental compilation
bugs like #93096 without affecting performance.
Introduce drop range tracking to generator interior analysis
This PR addresses cases such as this one from #57478:
```rust
struct Foo;
impl !Send for Foo {}
let _: impl Send = || {
let guard = Foo;
drop(guard);
yield;
};
```
Previously, the `generator_interior` pass would unnecessarily include the type `Foo` in the generator because it was not aware of the behavior of `drop`. We fix this issue by introducing a drop range analysis that finds portions of the code where a value is guaranteed to be dropped. If a value is dropped at all suspend points, then it is no longer included in the generator type. Note that we are using "dropped" in a generic sense to include any case in which a value has been moved. That is, we do not only look at calls to the `drop` function.
There are several phases to the drop tracking algorithm, and we'll go into more detail below.
1. Use `ExprUseVisitor` to find values that are consumed and borrowed.
2. `DropRangeVisitor` uses consume and borrow information to gather drop and reinitialization events, as well as build a control flow graph.
3. We then propagate drop and reinitialization information through the CFG until we reach a fix point (see `DropRanges::propagate_to_fixpoint`).
4. When recording a type (see `InteriorVisitor::record`), we check the computed drop ranges to see if that value is definitely dropped at the suspend point. If so, we skip including it in the type.
## 1. Use `ExprUseVisitor` to find values that are consumed and borrowed.
We use `ExprUseVisitor` to identify the places where values are consumed. We track both the `hir_id` of the value, and the `hir_id` of the expression that consumes it. For example, in the expression `[Foo]`, the `Foo` is consumed by the array expression, so after the array expression we can consider the `Foo` temporary to be dropped.
In this process, we also collect values that are borrowed. The reason is that the MIR transform for generators conservatively assumes anything borrowed is live across a suspend point (see `rustc_mir_transform::generator::locals_live_across_suspend_points`). We match this behavior here as well.
## 2. Gather drop events, reinitialization events, and control flow graph
After finding the values of interest, we perform a post-order traversal over the HIR tree to find the points where these values are dropped or reinitialized. We use the post-order index of each event because this is how the existing generator interior analysis refers to the position of suspend points and the scopes of variables.
During this traversal, we also record branching and merging information to handle control flow constructs such as `if`, `match`, and `loop`. This is necessary because values may be dropped along some control flow paths but not others.
## 3. Iterate to fixed point
The previous pass found the interesting events and locations, but now we need to find the actual ranges where things are dropped. Upon entry, we have a list of nodes ordered by their position in the post-order traversal. Each node has a set of successors. For each node we additionally keep a bitfield with one bit per potentially consumed value. The bit is set if we the value is dropped along all paths entering this node.
To compute the drop information, we first reverse the successor edges to find each node's predecessors. Then we iterate through each node, and for each node we set its dropped value bitfield to the intersection of all incoming dropped value bitfields.
If any bitfield for any node changes, we re-run the propagation loop again.
## 4. Ignore dropped values across suspend points
At this point we have a data structure where we can ask whether a value is guaranteed to be dropped at any post order index for the HIR tree. We use this information in `InteriorVisitor` to check whether a value in question is dropped at a particular suspend point. If it is, we do not include that value's type in the generator type.
Note that we had to augment the region scope tree to include all yields in scope, rather than just the last one as we did before.
r? `@nikomatsakis`
improve `_` constants in item signature handling
removing the "type" from the error messages does slightly worsen the error messages for types, but figuring out whether the placeholder is for a type or a constant and correctly dealing with that seemed fairly difficult to me so I took the easy way out ✨ Imo the error message is still clear enough.
r? `@BoxyUwU` cc `@estebank`
- Also rename a trivial_const_drop to match style of other functions in
the util module.
- Also add a test for `const Drop` that doesn't depend on a `~const`
bound.
- Also comment a bit why we remove the const bound during dropck impl
check.
This is the same idea as #92533, but for `AssocItem` instead
of `VariantDef`/`FieldDef`.
With this change, we no longer have any uses of
`#[stable_hasher(project(...))]`
