Refactor rustc_resolve::late::lifetimes to resolve per-item
There are some changes to tests that I'd like some feedback on; so this is still WIP.
The reason behind this change will (hopefully) allow us to (as part of #76814) be able to essentially use the lifetime resolve code to resolve *all* late bound vars (including those of super traits). Currently, it only resolves those that are *syntactically* in scope. In #76814, I'm essentially finding that I would essentially have to redo the passing of bound vars through scopes (i.e. when instantiating a poly trait ref), and that's what this code does anyways. However, to be able to do this (ask super traits what bound vars are in scope), we have to be able to resolve items separately.
The first commit is actually partially orthogonal. Essentially removing one use of late bound debruijn indices.
Not exactly sure who would be best to review here.
Let r? `@nikomatsakis`
coverage bug fixes and optimization support
Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to
address multiple, somewhat related issues.
Fixed a significant flaw in prior coverage solution: Every counter
generated a new counter variable, but there should have only been one
counter variable per function. This appears to have bloated .profraw
files significantly. (For a small program, it increased the size by
about 40%. I have not tested large programs, but there is anecdotal
evidence that profraw files were way too large. This is a good fix,
regardless, but hopefully it also addresses related issues.
Fixes: #82144
Invalid LLVM coverage data produced when compiled with -C opt-level=1
Existing tests now work up to at least `opt-level=3`. This required a
detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR
when compiled with coverage, and a lot of trial and error with codegen
adjustments.
The biggest hurdle was figuring out how to continue to support coverage
results for unused functions and generics. Rust's coverage results have
three advantages over Clang's coverage results:
1. Rust's coverage map does not include any overlapping code regions,
making coverage counting unambiguous.
2. Rust generates coverage results (showing zero counts) for all unused
functions, including generics. (Clang does not generate coverage for
uninstantiated template functions.)
3. Rust's unused functions produce minimal stubbed functions in LLVM IR,
sufficient for including in the coverage results; while Clang must
generate the complete LLVM IR for each unused function, even though
it will never be called.
This PR removes the previous hack of attempting to inject coverage into
some other existing function instance, and generates dedicated instances
for each unused function. This change, and a few other adjustments
(similar to what is required for `-C link-dead-code`, but with lower
impact), makes it possible to support LLVM optimizations.
Fixes: #79651
Coverage report: "Unexecuted instantiation:..." for a generic function
from multiple crates
Fixed by removing the aforementioned hack. Some "Unexecuted
instantiation" notices are unavoidable, as explained in the
`used_crate.rs` test, but `-Zinstrument-coverage` has new options to
back off support for either unused generics, or all unused functions,
which avoids the notice, at the cost of less coverage of unused
functions.
Fixes: #82875
Invalid LLVM coverage data produced with crate brotli_decompressor
Fixed by disabling the LLVM function attribute that forces inlining, if
`-Z instrument-coverage` is enabled. This attribute is applied to
Rust functions with `#[inline(always)], and in some cases, the forced
inlining breaks coverage instrumentation and reports.
FYI: `@wesleywiser`
r? `@tmandry`
2229 migration: Don't try resolve regions before writeback
In the analysis use `resolve_vars_if_possible` instead of `fully_resolve`,
because we might not have performed regionck yet.
Fixes: #83176
r? `@nikomatsakis`
This currently creates a field which is always false on GenericParamDefKind for future use when
consts are permitted to have defaults
Update const_generics:default locations
Previously just ignored them, now actually do something about them.
Fix using type check instead of value
Add parsing
This adds all the necessary changes to lower const-generics defaults from parsing.
Change P<Expr> to AnonConst
This matches the arguments passed to instantiations of const generics, and makes it specific to
just anonymous constants.
Attempt to fix lowering bugs
Revert performance-sensitive change in #82436
This change was done in #82436, as an "optimization". Unfortunately I
missed that this code is not always executed, because of the "continue"
in the conditional above it.
This commit should solve the perf regressions introduced by #82436 as I
think there isn't anything else that could affect runtime performance in
that PR. The `Pick` type grows only one word, which I doubt can cause up
to 8.8% increase in RSS in some of the benchmarks.
---
Could someone with the rights start a perf job please?
Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to
address multiple, somewhat related issues.
Fixed a significant flaw in prior coverage solution: Every counter
generated a new counter variable, but there should have only been one
counter variable per function. This appears to have bloated .profraw
files significantly. (For a small program, it increased the size by
about 40%. I have not tested large programs, but there is anecdotal
evidence that profraw files were way too large. This is a good fix,
regardless, but hopefully it also addresses related issues.
Fixes: #82144
Invalid LLVM coverage data produced when compiled with -C opt-level=1
Existing tests now work up to at least `opt-level=3`. This required a
detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR
when compiled with coverage, and a lot of trial and error with codegen
adjustments.
The biggest hurdle was figuring out how to continue to support coverage
results for unused functions and generics. Rust's coverage results have
three advantages over Clang's coverage results:
1. Rust's coverage map does not include any overlapping code regions,
making coverage counting unambiguous.
2. Rust generates coverage results (showing zero counts) for all unused
functions, including generics. (Clang does not generate coverage for
uninstantiated template functions.)
3. Rust's unused functions produce minimal stubbed functions in LLVM IR,
sufficient for including in the coverage results; while Clang must
generate the complete LLVM IR for each unused function, even though
it will never be called.
This PR removes the previous hack of attempting to inject coverage into
some other existing function instance, and generates dedicated instances
for each unused function. This change, and a few other adjustments
(similar to what is required for `-C link-dead-code`, but with lower
impact), makes it possible to support LLVM optimizations.
Fixes: #79651
Coverage report: "Unexecuted instantiation:..." for a generic function
from multiple crates
Fixed by removing the aforementioned hack. Some "Unexecuted
instantiation" notices are unavoidable, as explained in the
`used_crate.rs` test, but `-Zinstrument-coverage` has new options to
back off support for either unused generics, or all unused functions,
which avoids the notice, at the cost of less coverage of unused
functions.
Fixes: #82875
Invalid LLVM coverage data produced with crate brotli_decompressor
Fixed by disabling the LLVM function attribute that forces inlining, if
`-Z instrument-coverage` is enabled. This attribute is applied to
Rust functions with `#[inline(always)], and in some cases, the forced
inlining breaks coverage instrumentation and reports.
This change was done in #82436, as an "optimization". Unfortunately I
missed that this code is not always executed, because of the "continue"
in the conditional above it.
This commit should solve the perf regressions introduced by #82436 as I
think there isn't anything else that could affect runtime performance in
that PR. The `Pick` type grows only one word, which I doubt can cause up
to 8.8% increase in RSS in some of the benchmarks.
make changes to liveness to use closure_min_captures
use different span
borrow check uses new structures
rename to CapturedPlace
stop using upvar_capture in regionck
remove the bridge
cleanup from rebase + remove the upvar_capture reference from mutability_errors.rs
remove line from livenes test
make our unused var checking more consistent
update tests
adding more warnings to the tests
move is_ancestor_or_same_capture to rustc_middle/ty
update names to reflect the closures
add FIXME
check that all captures are immutable borrows before returning
add surrounding if statement like the original
move var out of the loop and rename
Co-authored-by: Logan Mosier <logmosier@gmail.com>
Co-authored-by: Roxane Fruytier <roxane.fruytier@hotmail.com>
ast/hir: Rename field-related structures
I always forget what `ast::Field` and `ast::StructField` mean despite working with AST for long time, so this PR changes the naming to less confusing and more consistent.
- `StructField` -> `FieldDef` ("field definition")
- `Field` -> `ExprField` ("expression field", not "field expression")
- `FieldPat` -> `PatField` ("pattern field", not "field pattern")
Various visiting and other methods working with the fields are renamed correspondingly too.
The second commit reduces the size of `ExprKind` by boxing fields of `ExprKind::Struct` in preparation for https://github.com/rust-lang/rust/pull/80080.
More precise spans for HIR paths
`Ty::assoc_item` is lowered to `<Ty>::assoc_item` in HIR, but `Ty` got span from the whole path.
This PR fixes that, and adjusts some diagnostic code that relied on `Ty` having the whole path span.
This is a pre-requisite for https://github.com/rust-lang/rust/pull/82868 (we cannot report suggestions like `Tr::assoc` -> `<dyn Tr>::assoc` with the current imprecise spans).
r? ````@estebank````
2229: Handle patterns within closures correctly when `capture_disjoint_fields` is enabled
This PR fixes several issues related to handling patterns within closures when `capture_disjoint_fields` is enabled.
1. Matching is always considered a use of the place, even with `_` patterns
2. Compiler ICE when capturing fields in closures through `let` assignments
To do so, we
- Introduced new Fake Reads
- Delayed use of `Place` in favor of `PlaceBuilder`
- Ensured that `PlaceBuilder` can be resolved before attempting to extract `Place` in any of the pattern matching code
Closes rust-lang/project-rfc-2229/issues/27
Closes rust-lang/project-rfc-2229/issues/24
r? `@nikomatsakis`
StructField -> FieldDef ("field definition")
Field -> ExprField ("expression field", not "field expression")
FieldPat -> PatField ("pattern field", not "field pattern")
Also rename visiting and other methods working on them.
Add a `min_type_alias_impl_trait` feature gate
This new feature gate only permits type alias impl trait to be constrained by function and trait method return types. All other possible constraining sites like const/static types, closure return types and binding types are now forbidden and gated under the `type_alias_impl_trait` and `impl_trait_in_bindings` feature gates (which are both marked as incomplete, as they have various ways to ICE the compiler or cause query cycles where they shouldn't).
r? `@nikomatsakis`
This is best reviewed commit-by-commit
Avoid sorting predicates by `DefId`
Fixes issue #82920
Even if an item does not change between compilation sessions, it may end
up with a different `DefId`, since inserting/deleting an item affects
the `DefId`s of all subsequent items. Therefore, we use a `DefPathHash`
in the incremental compilation system, which is stable in the face of
changes to unrelated items.
In particular, the query system will consider the inputs to a query to
be unchanged if any `DefId`s in the inputs have their `DefPathHash`es
unchanged. Queries are pure functions, so the query result should be
unchanged if the query inputs are unchanged.
Unfortunately, it's possible to inadvertantly make a query result
incorrectly change across compilations, by relying on the specific value
of a `DefId`. Specifically, if the query result is a slice that gets
sorted by `DefId`, the precise order will depend on how the `DefId`s got
assigned in a particular compilation session. If some definitions end up
with different `DefId`s (but the same `DefPathHash`es) in a subsequent
compilation session, we will end up re-computing a *different* value for
the query, even though the query system expects the result to unchanged
due to the unchanged inputs.
It turns out that we have been sorting the predicates computed during
`astconv` by their `DefId`. These predicates make their way into the
`super_predicates_that_define_assoc_type`, which ends up getting used to
compute the vtables of trait objects. This, re-ordering these predicates
between compilation sessions can lead to undefined behavior at runtime -
the query system will re-use code built with a *differently ordered*
vtable, resulting in the wrong method being invoked at runtime.
This PR avoids sorting by `DefId` in `astconv`, fixing the
miscompilation. However, it's possible that other instances of this
issue exist - they could also be easily introduced in the future.
To fully fix this issue, we should
1. Turn on `-Z incremental-verify-ich` by default. This will cause the
compiler to ICE whenver an 'unchanged' query result changes between
compilation sessions, instead of causing a miscompilation.
2. Remove the `Ord` impls for `CrateNum` and `DefId`. This will make it
difficult to introduce ICEs in the first place.
Fixes issue #82920
Even if an item does not change between compilation sessions, it may end
up with a different `DefId`, since inserting/deleting an item affects
the `DefId`s of all subsequent items. Therefore, we use a `DefPathHash`
in the incremental compilation system, which is stable in the face of
changes to unrelated items.
In particular, the query system will consider the inputs to a query to
be unchanged if any `DefId`s in the inputs have their `DefPathHash`es
unchanged. Queries are pure functions, so the query result should be
unchanged if the query inputs are unchanged.
Unfortunately, it's possible to inadvertantly make a query result
incorrectly change across compilations, by relying on the specific value
of a `DefId`. Specifically, if the query result is a slice that gets
sorted by `DefId`, the precise order will depend on how the `DefId`s got
assigned in a particular compilation session. If some definitions end up
with different `DefId`s (but the same `DefPathHash`es) in a subsequent
compilation session, we will end up re-computing a *different* value for
the query, even though the query system expects the result to unchanged
due to the unchanged inputs.
It turns out that we have been sorting the predicates computed during
`astconv` by their `DefId`. These predicates make their way into the
`super_predicates_that_define_assoc_type`, which ends up getting used to
compute the vtables of trait objects. This, re-ordering these predicates
between compilation sessions can lead to undefined behavior at runtime -
the query system will re-use code built with a *differently ordered*
vtable, resulting in the wrong method being invoked at runtime.
This PR avoids sorting by `DefId` in `astconv`, fixing the
miscompilation. However, it's possible that other instances of this
issue exist - they could also be easily introduced in the future.
To fully fix this issue, we should
1. Turn on `-Z incremental-verify-ich` by default. This will cause the
compiler to ICE whenver an 'unchanged' query result changes between
compilation sessions, instead of causing a miscompilation.
2. Remove the `Ord` impls for `CrateNum` and `DefId`. This will make it
difficult to introduce ICEs in the first place.
2229: Handle capturing a reference into a repr packed struct
RFC 1240 states that it is unsafe to capture references into a
packed-struct. This PR ensures that when a closure captures a precise
path, we aren't violating this safety constraint.
To acheive so we restrict the capture precision to the struct itself.
An interesting edge case where we decided to restrict precision:
```rust
struct Foo(String);
let foo: Foo;
let c = || {
println!("{}", foo.0);
let x = foo.0;
}
```
Given how closures get desugared today, foo.0 will be moved into the
closure, making the `println!`, safe. However this can be very subtle
and also will be unsafe if the closure gets inline.
Closes: https://github.com/rust-lang/project-rfc-2229/issues/33
r? `@nikomatsakis`
Allow calling *const methods on *mut values
This allows `*const` methods to be called on `*mut` values.
TODOs:
- [x] ~~Remove debug logs~~ Done.
- [x] ~~I haven't tested, but I think this currently won't work when the `self` value has type like `&&&&& *mut X` because I don't do any autoderefs when probing. To fix this the new code in `rustc_typeck::check::method::probe` needs to reuse `pick_method` somehow as I think that's the function that autoderefs.~~ This works, because autoderefs are done before calling `pick_core`, in `method_autoderef_steps`, called by `probe_op`.
- [x] ~~I should probably move the new `Pick` to `pick_autorefd_method`. If not, I should move it to its own function.~~ Done.
- [ ] ~~Test this with a `Pick` with `to_ptr = true` and `unsize = true`.~~ I think this case cannot happen, because we don't have any array methods with `*mut [X]` receiver. I should confirm that this is true and document this. I've placed two assertions about this.
- [x] ~~Maybe give `(Mutability, bool)` a name and fields~~ I now have a `to_const_ptr` field in `Pick`.
- [x] ~~Changes in `adjust_self_ty` is quite hacky. The problem is we can't deref a pointer, and even if we don't have an adjustment to get the address of a value, so to go from `*mut` to `*const` we need a special case.~~ There's still a special case for `to_const_ptr`, but I'm not sure if we can avoid this.
- [ ] Figure out how `reached_raw_pointer` stuff is used. I suspect only for error messages.
Fixes#80258
Implement RFC 2945: "C-unwind" ABI
## Implement RFC 2945: "C-unwind" ABI
This branch implements [RFC 2945]. The tracking issue for this RFC is #74990.
The feature gate for the issue is `#![feature(c_unwind)]`.
This RFC was created as part of the ffi-unwind project group tracked at rust-lang/lang-team#19.
### Changes
Further details will be provided in commit messages, but a high-level overview
of the changes follows:
* A boolean `unwind` payload is added to the `C`, `System`, `Stdcall`,
and `Thiscall` variants, marking whether unwinding across FFI boundaries is
acceptable. The cases where each of these variants' `unwind` member is true
correspond with the `C-unwind`, `system-unwind`, `stdcall-unwind`, and
`thiscall-unwind` ABI strings introduced in RFC 2945 [3].
* This commit adds a `c_unwind` feature gate for the new ABI strings.
Tests for this feature gate are included in `src/test/ui/c-unwind/`, which
ensure that this feature gate works correctly for each of the new ABIs.
A new language features entry in the unstable book is added as well.
* We adjust the `rustc_middle::ty::layout::fn_can_unwind` function,
used to compute whether or not a `FnAbi` object represents a function that
should be able to unwind when `panic=unwind` is in use.
* Changes are also made to
`rustc_mir_build::build::should_abort_on_panic` so that the function ABI is
used to determind whether it should abort, assuming that the `panic=unwind`
strategy is being used, and no explicit unwind attribute was provided.
[RFC 2945]: https://github.com/rust-lang/rfcs/blob/master/text/2945-c-unwind-abi.md
Store HIR attributes in a side table
Same idea as #72015 but for attributes.
The objective is to reduce incr-comp invalidations due to modified attributes.
Notably, those due to modified doc comments.
Implementation:
- collect attributes during AST->HIR lowering, in `LocalDefId -> ItemLocalId -> &[Attributes]` nested tables;
- access the attributes through a `hir_owner_attrs` query;
- local refactorings to use this access;
- remove `attrs` from HIR data structures one-by-one.
Change in behaviour:
- the HIR visitor traverses all attributes at once instead of parent-by-parent;
- attribute arrays are sometimes duplicated: for statements and variant constructors;
- as a consequence, attributes are marked as used after unused-attribute lint emission to avoid duplicate lints.
~~Current bug: the lint level is not correctly applied in `std::backtrace_rs`, triggering an unused attribute warning on `#![no_std]`. I welcome suggestions.~~
RFC 1240 states that it is unsafe to capture references into a
packed-struct. This PR ensures that when a closure captures a precise
path, we aren't violating this safety constraint.
To acheive so we restrict the capture precision to the struct itself.
An interesting edge case:
```rust
struct Foo(String);
let foo: Foo;
let c = || {
println!("{}", foo.0);
let x = foo.0;
}
```
Given how closures get desugared today, foo.0 will be moved into the
closure, making the `println!`, safe. However this can be very subtle
and also will be unsafe if the closure gets inline.
Closes: https://github.com/rust-lang/project-rfc-2229/issues/33
Don't hardcode the `v1` prelude in diagnostics, to allow for new preludes.
Instead of looking for `std::prelude::v1`, this changes the two places where that was hardcoded to look for `std::prelude::<anything>` instead.
This is needed for https://github.com/rust-lang/rust/pull/82217.
r? `@estebank`
### Overview
This commit begins the implementation work for RFC 2945. For more
information, see the rendered RFC [1] and tracking issue [2].
A boolean `unwind` payload is added to the `C`, `System`, `Stdcall`,
and `Thiscall` variants, marking whether unwinding across FFI
boundaries is acceptable. The cases where each of these variants'
`unwind` member is true correspond with the `C-unwind`,
`system-unwind`, `stdcall-unwind`, and `thiscall-unwind` ABI strings
introduced in RFC 2945 [3].
### Feature Gate and Unstable Book
This commit adds a `c_unwind` feature gate for the new ABI strings.
Tests for this feature gate are included in `src/test/ui/c-unwind/`,
which ensure that this feature gate works correctly for each of the
new ABIs.
A new language features entry in the unstable book is added as well.
### Further Work To Be Done
This commit does not proceed to implement the new unwinding ABIs,
and is intentionally scoped specifically to *defining* the ABIs and
their feature flag.
### One Note on Test Churn
This will lead to some test churn, in re-blessing hash tests, as the
deleted comment in `src/librustc_target/spec/abi.rs` mentioned,
because we can no longer guarantee the ordering of the `Abi`
variants.
While this is a downside, this decision was made bearing in mind
that RFC 2945 states the following, in the "Other `unwind` Strings"
section [3]:
> More unwind variants of existing ABI strings may be introduced,
> with the same semantics, without an additional RFC.
Adding a new variant for each of these cases, rather than specifying
a payload for a given ABI, would quickly become untenable, and make
working with the `Abi` enum prone to mistakes.
This approach encodes the unwinding information *into* a given ABI,
to account for the future possibility of other `-unwind` ABI
strings.
### Ignore Directives
`ignore-*` directives are used in two of our `*-unwind` ABI test
cases.
Specifically, the `stdcall-unwind` and `thiscall-unwind` test cases
ignore architectures that do not support `stdcall` and
`thiscall`, respectively.
These directives are cribbed from
`src/test/ui/c-variadic/variadic-ffi-1.rs` for `stdcall`, and
`src/test/ui/extern/extern-thiscall.rs` for `thiscall`.
This would otherwise fail on some targets, see:
fcf697f902
### Footnotes
[1]: https://github.com/rust-lang/rfcs/blob/master/text/2945-c-unwind-abi.md
[2]: https://github.com/rust-lang/rust/issues/74990
[3]: https://github.com/rust-lang/rfcs/blob/master/text/2945-c-unwind-abi.md#other-unwind-abi-strings
