| Age | Commit message (Collapse) | Author | Lines |
|---|
|
r=WaffleLapkin,JakobDegen
Ensure `ptr::read` gets all the same LLVM `load` metadata that dereferencing does
I was looking into `array::IntoIter` optimization, and noticed that it wasn't annotating the loads with `noundef` for simple things like `array::IntoIter<i32, N>`. Trying to narrow it down, it seems that was because `MaybeUninit::assume_init_read` isn't marking the load as initialized (<https://rust.godbolt.org/z/Mxd8TPTnv>), which is unfortunate since that's basically its reason to exist.
The root cause is that `ptr::read` is currently implemented via the *untyped* `copy_nonoverlapping`, and thus the `load` doesn't get any type-aware metadata: no `noundef`, no `!range`. This PR solves that by lowering `ptr::read(p)` to `copy *p` in MIR, for which the backends already do the right thing.
Fortuitiously, this also improves the IR we give to LLVM for things like `mem::replace`, and fixes a couple of long-standing bugs where `ptr::read` on `Copy` types was worse than `*`ing them.
Zulip conversation: <https://rust-lang.zulipchat.com/#narrow/stream/219381-t-libs/topic/Move.20array.3A.3AIntoIter.20to.20ManuallyDrop/near/341189936>
cc `@erikdesjardins` `@JakobDegen` `@workingjubilee` `@the8472`
Fixes #106369
Fixes #73258
|
|
Apparently in CI it's getting generated in the opposite order, one function per file will make the test pass either way.
|
|
|
|
remove obsolete `givens` from regionck
Fixes #106567
r? `@lcnr` (feel free to reassign)
|
|
Rollup of 9 pull requests
Successful merges:
- #108722 (Support for Fuchsia RISC-V target)
- #108880 (Remove tests/ui/impl-trait/in-trait/new-lowering-strategy in favor of using revisions on existing tests)
- #108909 (Fix object safety checks for new RPITITs)
- #108915 (Remove some direct calls to local_def_id_to_hir_id on diagnostics)
- #108923 (Make fns from other crates with RPITIT work for -Zlower-impl-trait-in-trait-to-assoc-ty)
- #109101 (Fall back to old metadata computation when type references errors)
- #109105 (Don't ICE for late-bound consts across `AnonConstBoundary`)
- #109110 (Don't codegen impossible to satisfy impls)
- #109116 (Emit diagnostic when calling methods on the unit type in method chains)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
|
|
MaciejWas:add-modifies-receiver-diagn-when-method-not-found, r=petrochenkov
Emit diagnostic when calling methods on the unit type in method chains
Fixes #104204.
What this PR does: If a method is not found somewhere in a call chain, we check if we called earlier a method with signature `(&mut T, ...) -> ()`. If this is the case then we emit a diagnostic message.
For example given input:
```
vec![1, 2, 3].into_iter().collect::<Vec<i32>>().sort_by_key(|i| i).sort();
```
the current output is:
```
error[E0599]: no method named `sort` found for unit type `()` in the current scope
--> hello.rs:3:72
|
3 | vec![1, 2, 3].into_iter().collect::<Vec<i32>>().sort_by_key(|i| i).sort();
| ^^^^ method not found in `()`
```
after this PR it will be:
```
error[E0599]: no method named `sort` found for unit type `()` in the current scope
--> ./hello.rs:3:72
|
3 | vec![1, 2, 3].into_iter().collect::<Vec<i32>>().sort_by_key(|i| i).sort();
| ^^^^ method not found in `()`
|
note: method `sort_by_key` modifies its receiver in-place, it is not meant to be used in method chains.
--> ./hello.rs:3:53
|
3 | vec![1, 2, 3].into_iter().collect::<Vec<i32>>().sort_by_key(|i| i).sort();
| ^^^^^^^^^^^ this call modifies its receiver in-place
```
|
|
Don't codegen impossible to satisfy impls
Fixes #109098
|
|
Don't ICE for late-bound consts across `AnonConstBoundary`
Fixes #108194
|
|
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
|
|
Make fns from other crates with RPITIT work for -Zlower-impl-trait-in-trait-to-assoc-ty
Only the last two commits are meaningful.
r? `@compiler-errors`
|
|
Fix object safety checks for new RPITITs
This one goes on top of #108869
r? `@compiler-errors`
|
|
Remove tests/ui/impl-trait/in-trait/new-lowering-strategy in favor of using revisions on existing tests
r? `@compiler-errors`
This one again sits on top of existing approved PRs and it still needs to add revisions to tests in `tests/ui/impl-trait/in-trait` as it only does so for async in traits.
|
|
|
|
If no method is found when checking method call, we check if we called a method with signature (&mut T, ...) -> (). If this is the case then we emit a diagnostic message
|
|
r=petrochenkov
Properly allow macro expanded `format_args` invocations to uses captures
Originally, this was kinda half-allowed. There were some primitive checks in place that looked at the span to see whether the input was likely a literal. These "source literal" checks are needed because the spans created during `format_args` parsing only make sense when it is indeed a literal that was written in the source code directly.
This is orthogonal to the restriction that the first argument must be a "direct literal", not being exanpanded from macros. This restriction was imposed by [RFC 2795] on the basis of being too confusing. But this was only concerned with the argument of the invocation being a literal, not whether it was a source literal (maybe in spirit it meant it being a source literal, this is not clear to me).
Since the original check only really cared about source literals (which is good enough to deny the `format_args!(concat!())` example), macros expanding to `format_args` invocations were able to use implicit captures if they spanned the string in a way that lead back to a source string.
The "source literal" checks were not strict enough and caused ICEs in certain cases (see #106191). So I tightened it up in #106195 to really only work if it's a direct source literal.
This caused the `indoc` crate to break. `indoc` transformed the source literal by removing whitespace, which made it not a "source literal" anymore (which is required to fix the ICE). But since `indoc` spanned the literal in ways that made the old check think that it's a literal, it was able to use implicit captures (which is useful and nice for the users of `indoc`).
This commit properly seperates the previously introduced concepts of "source literal" and "direct literal" and therefore allows `indoc` invocations, which don't create "source literals" to use implicit captures again.
Fixes #106191
[RFC 2795]: https://rust-lang.github.io/rfcs/2795-format-args-implicit-identifiers.html#macro-hygiene
|
|
|
|
Originally, this was kinda half-allowed. There were some primitive
checks in place that looked at the span to see whether the input was
likely a literal. These "source literal" checks are needed because the
spans created during `format_args` parsing only make sense when it is
indeed a literal that was written in the source code directly.
This is orthogonal to the restriction that the first argument must be a
"direct literal", not being exanpanded from macros. This restriction was
imposed by [RFC 2795] on the basis of being too confusing. But this was
only concerned with the argument of the invocation being a literal, not
whether it was a source literal (maybe in spirit it meant it being a
source literal, this is not clear to me).
Since the original check only really cared about source literals (which
is good enough to deny the `format_args!(concat!())` example), macros
expanding to `format_args` invocations were able to use implicit
captures if they spanned the string in a way that lead back to a source
string.
The "source literal" checks were not strict enough and caused ICEs in
certain cases (see # 106191 (the space is intended to avoid spammy
backreferences)). So I tightened it up in # 106195 to really only work
if it's a direct source literal.
This caused the `indoc` crate to break. `indoc` transformed the source
literal by removing whitespace, which made it not a "source literal"
anymore (which is required to fix the ICE). But since `indoc` spanned
the literal in ways that made the old check think that it's a literal,
it was able to use implicit captures (which is useful and nice for the
users of `indoc`).
This commit properly seperates the previously introduced concepts of
"source literal" and "direct literal" and therefore allows `indoc`
invocations, which don't create "source literals" to use implicit
captures again.
[RFC 2795]: https://rust-lang.github.io/rfcs/2795-format-args-implicit-identifiers.html#macro-hygiene
|
|
Remove `identity_future` indirection
This was previously needed because the indirection used to hide some unexplained lifetime errors, which it turned out were related to the `min_choice` algorithm.
Removing the indirection also solves a couple of cycle errors, large moves and makes async blocks support the `#[track_caller]`annotation.
Fixes https://github.com/rust-lang/rust/issues/104826.
|
|
resolve: Querify most cstore access methods (subset)
A subset of https://github.com/rust-lang/rust/pull/108346 that is not on a hot path in any way.
|
|
|
|
|
|
r=compiler-errors
Gracefully handle `#[target_feature]` on statics
The was careful around not calling `fn_sig` on not-functions but well, it wasn't careful enough. This commit makes it a little more careful and also adds tests for a bunch more item kinds.
I was sadly not able to fully bless the test locally because I'm on an aarch64 machine but I hope some manual editing made it work 😅
Fix #109079
|
|
simd-wide-sum test: adapt for LLVM 17 codegen change
After https://github.com/llvm/llvm-project/commit/0d4a709bb876824a0afa5f86e138e8ffdcaf7661 LLVM becomes more clever and turns ```@wider_reduce_loop``` into an alias:
https://buildkite.com/llvm-project/rust-llvm-integrate-prototype/builds/17806#0186da6b-582c-46bf-a227-1565fa0859ac/743-766
This adapts the test to prevent this.
|
|
Treat projections with infer as placeholder during fast reject in new solver
r? ``@lcnr``
Kind of a shame that we need to change all of the call sites for `for_each_relevant_impl`, etc. to pass an extra parameter. I guess I could have the "default" fn which calls a configurable fn?
|
|
The was careful around not calling `fn_sig` on not-functions but well,
it wasn't careful enough. This commit makes it a little more careful and
also adds tests for a bunch more item kinds.
|
|
|
|
|
|
After https://github.com/llvm/llvm-project/commit/0d4a709bb876824a0afa5f86e138e8ffdcaf7661
LLVM becomes more clever and turns `@wider_reduce_loop` into an alias:
https://buildkite.com/llvm-project/rust-llvm-integrate-prototype/builds/17806#0186da6b-582c-46bf-a227-1565fa0859ac/743-766
This adapts the test to prevent this.
|
|
Move `Option::as_slice` to an always-sound implementation
This approach depends on CSE to not have any branches or selects when the guessed offset is correct -- which it always will be right now -- but to also be *sound* (just less efficient) if the layout algorithms change such that the guess is incorrect.
The codegen test confirms that CSE handles this as expected, leaving the optimal codegen.
cc JakobDegen #108545
|
|
|
|
Remove `box_syntax`
r? `@Nilstrieb`
This removes the feature `box_syntax`, which allows the use of `box <expr>` to create a Box, and finalises removing use of the feature from the compiler. `box_patterns` (allowing the use of `box <pat>` in a pattern) is unaffected.
It also removes `ast::ExprKind::Box` - the only way to create a 'box' expression now is with the rustc-internal `#[rustc_box]` attribute.
As a temporary measure to help users move away, `box <expr>` now parses the inner expression, and emits a `MachineApplicable` lint to replace it with `Box::new`
Closes #49733
|
|
Strengthen state tracking in const-prop
Some/many of the changes are replicated between both the const-prop lint and the const-prop optimization.
Behaviour changes:
- const-prop opt does not give a span to propagated values. This was useless as that span's primary purpose is to diagnose evaluation failure in codegen.
- we remove the `OnlyPropagateInto` mode. It was only used for function arguments, which are better modeled by a write before entry.
- the tracking of assignments and discriminants make clearer that we do nothing in `NoPropagation` mode or on indirect places.
|
|
|
|
|
|
Gate usages of `dyn*` and const closures in macros
We silently accepted `dyn*` and const closures in macros as long as they didn't expand to anything containing these experimental features, unlike other gated features such as `for<'a>` binders on closures, etc. Let's not do that, to make sure nobody begins relying on this.
|
|
rustdoc: use restricted Damerau-Levenshtein distance for search
Based on https://github.com/rust-lang/rust/pull/108200, for the same rationale.
> This replaces the existing Levenshtein algorithm with the Damerau-Levenshtein algorithm. This means that "ab" to "ba" is one change (a transposition) instead of two (a deletion and insertion). More specifically, this is a restricted implementation, in that "ca" to "abc" cannot be performed as "ca" → "ac" → "abc", as there is an insertion in the middle of a transposition. I believe that errors like that are sufficiently rare that it's not worth taking into account.
Before this change, searching [`prinltn!`] listed `print!` first, followed by `println!`. With this change, `println!` matches more closely.
[`prinltn!`]: https://doc.rust-lang.org/nightly/std/?search=prinltn!
|
|
|
|
|
|
|
|
|
|
|
|
-Zlower-impl-trait-in-trait-to-assoc-ty
|
|
|
|
revisions on existing tests
|
|
|
|
|
|
|
|
Commit some tests for the new solver + lazy norm
Also consolidate `typeck/lazy-norm` into `traits/new-solver`, since it's not really useful to maintain a distinction, like when a test really is due to "lazy norm" or "the new solver" (usually both!)
|
|
Add suggestion to diagnostic when user has array but trait wants slice. (rebased)
Rebase of #91314, except for change to multipart suggestion
Resolves #90528
r? ``@compiler-errors`` since you requested the multipart suggestion
|
|
This approach depends on CSE to not have any branches or selects when the guessed offset is correct -- which it always will be right now -- but to also be *sound* (just less efficient) if the layout algorithms change such that the guess is incorrect.
|
