When the panic handler is run, the existing Handler may be in a weird
state if it was responsible for triggering the panic. By using a freshly
created Handler, we avoid trying to re-entrantly lock a HandlerInner,
which was causing a double panic on ICEs.
Rename `subst::Kind` to `subst::GenericArg`
And `subst::UnpackedKind` to `subst::GenericArgKind`. Individual variable names (e.g. `kind`) are not renamed, which would be an infeasible mission.
Fixes https://github.com/rust-lang/rust/issues/64352.
r? @eddyb
reserve `impl<T> From<!> for T`
this is necessary for never-type stabilization.
cc #57012#35121
I think we wanted a crater run for this @nikomatsakis?
r? @nikomatsakis
or-patterns: Push `PatKind/PatternKind::Or` at top level to HIR & HAIR
Following up on work in https://github.com/rust-lang/rust/pull/64111, https://github.com/rust-lang/rust/pull/63693, and https://github.com/rust-lang/rust/pull/61708, in this PR:
- We change `hair::Arm.patterns: Vec<Pattern<'_>>` into `hir::Arm.pattern: Pattern<'_>`.
- `fn hair::Arm::top_pats_hack` is introduced as a temporary crutch in MIR building to avoid more changes.
- We change `hir::Arm.pats: HirVec<P<Pat>>` into `hir::Arm.pat: P<Pat>`.
- The hacks in `rustc::hir::lowering` are removed since the representation hack is no longer necessary.
- In some places, `fn hir::Arm::top_pats_hack` is introduced to leave some things as future work.
- Misc changes: HIR pretty printing is adjusted to behave uniformly wrt. top/inner levels, rvalue promotion is adjusted, regionck, and dead_code is also.
- Type checking is adjusted to uniformly handle or-patterns at top/inner levels.
To make things compile, `p_0 | ... | p_n` is redefined as a "reference pattern" in [`fn is_non_ref_pat`](https://doc.rust-lang.org/nightly/nightly-rustc/rustc_typeck/check/struct.FnCtxt.html#method.is_non_ref_pat) for now. This is done so that reference types are not eagerly stripped from the `expected: Ty<'tcx>`.
- Liveness is adjusted wrt. the `unused_variables` and `unused_assignments` lints to handle top/inner levels uniformly and the handling of `fn` parameters, `let` locals, and `match` arms are unified in this respect. This is not tested for now as exhaustiveness checks are reachable and will ICE.
- In `check_match`, checking `@` and by-move bindings is adjusted. However, exhaustiveness checking is not adjusted the moment and is handled by @dlrobertson in https://github.com/rust-lang/rust/pull/63688.
- AST borrowck (`construct.rs`) is not adjusted as AST borrowck will be removed soon.
r? @matthewjasper
cc @dlrobertson @varkor @oli-obk
Almost fully deprecate hir::map::Map.hir_to_node_id
- HirIdify `doctree::Module.id`
- HirIdify `hir::Crate.modules`
- introduce a `HirId` to `DefIndex` map in `map::Definitions`
The only last uses of `hir::map::Map.hir_to_node_id` in the compiler are:
- for the purposes of `driver::pretty` (in `map::nodes_matching_suffix`), but I don't know if we can remove `NodeId`s in there (I think when I attempted it previously there was some issue due to `HirId` not being representable with an integer)
- in `ty::query::on_disk_cache` (not sure about the purpose of this one)
- in `mir::transform::check_unsafety` (only important for error message order)
Any suggestions how to kill these off?
r? @Zoxc
Remove blanket silencing of "type annotation needed" errors
Remove blanket check for existence of other errors before emitting "type annotation needed" errors, and add some eager checks to avoid adding obligations when they refer to types that reference `[type error]` in order to reduce unneeded errors.
Fix#64084.
rustc: Fix mixing crates with different `share_generics`
This commit addresses #64319 by removing the `dylib` crate type from the
list of crate type that exports generic symbols. The bug in #64319
arises because a `dylib` crate type was trying to export a symbol in an
uptream crate but it miscalculated the symbol name of the uptream
symbol. This isn't really necessary, though, since `dylib` crates aren't
that heavily used, so we can just conservatively say that the `dylib`
crate type never exports generic symbols, forcibly removing them from
the exported symbol lists if were to otherwise find them.
The fix here happens in two places:
* First is in the `local_crate_exports_generics` method, indicating that
it's now `false` for the `Dylib` crate type. Only rlibs actually
export generics at this point.
* Next is when we load exported symbols from upstream crate. If, for our
compilation session, the crate may be included from a dynamic library,
then its generic symbols are removed. When the crate was linked into a
dynamic library its symbols weren't exported, so we can't consider
them a candidate to link against.
Overally this should avoid situations where we incorrectly calculate the
upstream symbol names in the face of differnet `share_generics` options,
ultimately...
Closes#64319
Recover on `const X = 42;` and infer type + Error Stash API
Here we:
1. Introduce a notion of the "error stash".
This is a map in the `Handler` to which you can `err.stash(...)` away your diagnostics and then steal them in a later "phase" of the compiler (e.g. stash in parser, steal in typeck) to enrich them with more information that isn't available in the previous "phase".
I believe I've covered all the bases to make sure these diagnostics are actually emitted eventually even under `#[cfg(FALSE)]` but please check my logic.
2. Recover when parsing `[const | static mut?] $ident = $expr;` which has a missing type.
Use the "error stash" to stash away the error and later steal the error in typeck where we emit the error as `MachineApplicable` with the actual inferred type. This builds on https://github.com/rust-lang/rust/pull/62804.
cc https://github.com/rust-lang/rfcs/pull/2545
r? @estebank
Add a cycle detector for generic `Graph`s and `mir::Body`s
Cycle detection is one way to differentiate the upcoming `const_loop` feature flag (#52000) from the `const_if_match` one (#49146). It would be possible to use the existing implementation of strongly-connected components for this but less efficient.
The ["tri-color" terminology](http://www.cs.cornell.edu/courses/cs2112/2012sp/lectures/lec24/lec24-12sp.html) is common in introductory data structures and algorithms courses: black nodes are settled, grey nodes are visited, and white nodes have no state. This particular implementation is iterative and uses a well-known technique where "node settled" events are kept on the stack alongside nodes to visit. When a settled event is popped, we know that all successors of that node have been visited and themselves settled. If we encounter a successor node that has been visited (is on the stack) but not yet settled, we have found a cycle.
r? @eddyb
Remove blanket check for existence of other errors before emitting
"type annotation needed" errors, and add some eager checks to avoid
adding obligations when they refer to types that reference
`[type error]` in order to reduce unneded errors.
Fixes#63677
RFC #2071 (impl-trait-existential-types) does not explicitly state how
impl trait type alises should interact with coherence. However, there's
only one choice which makes sense - coherence should look at the
underlying type (i.e. the 'defining' type of the impl trait) of the type
alias, just like we do for non-impl-trait type aliases.
Specifically, impl trait type alises which resolve to a local type
should be treated like a local type with respect to coherence (e.g.
impl trait type aliases which resolve to a forieign type should be
treated as a foreign type, and those that resolve to a local type should
be treated as a local type).
Since neither inherent impls nor direct trait impl (i.e. `impl MyType`
or `impl MyTrait for MyType`) are allowd for type aliases, this
usually does not come up. Before we ever attempt to do coherence
checking, we will have errored out if an impl trait type alias was used
directly in an 'impl' clause.
However, during trait selection, we sometimes need to prove bounds like
'T: Sized' for some type 'T'. If 'T' is an impl trait type alias, this
requires to know the coherence behavior for impl trait type aliases when
we perform coherence checking.
Note: Since determining the underlying type of an impl trait type alias
requires us to perform body type checking, this commit causes us to type
check some bodies easlier than we otherwise would have. However, since
this is done through a query, this shouldn't cause any problems
For completeness, I've added an additional test of the coherence-related
behavior of impl trait type aliases.
This commit addresses #64319 by removing the `dylib` crate type from the
list of crate type that exports generic symbols. The bug in #64319
arises because a `dylib` crate type was trying to export a symbol in an
uptream crate but it miscalculated the symbol name of the uptream
symbol. This isn't really necessary, though, since `dylib` crates aren't
that heavily used, so we can just conservatively say that the `dylib`
crate type never exports generic symbols, forcibly removing them from
the exported symbol lists if were to otherwise find them.
The fix here happens in two places:
* First is in the `local_crate_exports_generics` method, indicating that
it's now `false` for the `Dylib` crate type. Only rlibs actually
export generics at this point.
* Next is when we load exported symbols from upstream crate. If, for our
compilation session, the crate may be included from a dynamic library,
then its generic symbols are removed. When the crate was linked into a
dynamic library its symbols weren't exported, so we can't consider
them a candidate to link against.
Overally this should avoid situations where we incorrectly calculate the
upstream symbol names in the face of differnet `share_generics` options,
ultimately...
Closes#64319
This commit converts a field of `Session`, `dependency_formats`, into a
query of `TyCtxt`. This information then also needed to be threaded
through to other remaining portions of the linker, but it's relatively
straightforward. The only change here is that instead of
`HashMap<CrateType, T>` the data structure changed to `Vec<(CrateType,
T)>` to make it easier to deal with in queries.
Refactor librustc_errors::Handler API
This should be reviewed by-commit.
The last commit moves all fields into an inner struct behind a single lock; this is done to prevent possible deadlocks in a multi-threaded compiler, as well as inconsistent state observation.
Infer consts more consistently
Moved some duplicated logic in `TypeRelation` methods into `super_combined_consts`. Before some `TypeRelation`s like `Lub` wasn't using `replace_if_possible`, meaning some inference types were staying around longer than they should be.
Fixes https://github.com/rust-lang/rust/issues/64519
r? @varkor
record fewer adjustment types in generator witnesses, avoid spurious drops in MIR construction
Don't record all intermediate adjustment types -- That's way more than is needed, and winds up recording types that will never appear in MIR.
Note: I'm like 90% sure that this logic is correct, but this stuff is subtle and can be hard to keep straight. However, the risk of this PR is fairly low -- if we miss types here, I believe the most common outcome is an ICE.
This fixes the original issue cited by #64477, but I'm leaving the issue open for now since there may be other cases we can detect and improve in a targeted way.
r? @Zoxc
