`slice.get(i)` should use a slice projection in MIR, like `slice[i]` does
`slice[i]` is built-in magic, so ends up being quite different from `slice.get(i)` in MIR, even though they're both doing nearly identical operations -- checking the length of the slice then getting a ref/ptr to the element if it's in-bounds.
This PR adds a `slice_get_unchecked` intrinsic for `impl SliceIndex for usize` to use to fix that, so it no longer needs to do a bunch of lines of pointer math and instead just gets the obvious single statement. (This is *not* used for the range versions, since `slice[i..]` and `slice[..k]` can't use the mir Slice projection as they're using fenceposts, not indices.)
I originally tried to do this with some kind of GVN pattern, but realized that I'm pretty sure it's not legal to optimize `BinOp::Offset` to `PlaceElem::Index` without an extremely complicated condition. Basically, the problem is that the `Index` projection on a dereferenced slice pointer *cares about the metadata*, since it's UB to `PlaceElem::Index` outside the range described by the metadata. But then you cast the fat pointer to a thin pointer then offset it, that *ignores* the slice length metadata, so it's possible to write things that are legal with `Offset` but would be UB if translated in the obvious way to `Index`. Checking (or even determining) the necessary conditions for that would be complicated and error-prone, whereas this intrinsic-based approach is quite straight-forward.
Zero backend changes, because it just lowers to MIR, so it's already supported naturally by CTFE/Miri/cg_llvm/cg_clif.
atomic_load intrinsic: use const generic parameter for ordering
We have a gazillion intrinsics for the atomics because we encode the ordering into the intrinsic name rather than making it a parameter. This is particularly bad for those operations that take two orderings. Let's fix that!
This PR only converts `load`, to see if there's any feedback that would fundamentally change the strategy we pursue for the const generic intrinsics.
The first two commits are preparation and could be a separate PR if you prefer.
`@BoxyUwU` -- I hope this is a use of const generics that is unlikely to explode? All we need is a const generic of enum type. We could funnel it through an integer if we had to but an enum is obviously nicer...
`@bjorn3` it seems like the cranelift backend entirely ignores the ordering?
Specifically `TyAlias`, `Enum`, `Struct`, `Union`. So the fields match
the textual order in the source code.
The interesting part of the change is in
`compiler/rustc_hir/src/hir.rs`. The rest is extremely mechanical
refactoring.
GCI: At their def site, actually wfcheck the where-clause & always eval free lifetime-generic constants
* 1st commit: Partially addresses [#136204](https://github.com/rust-lang/rust/issues/136204) by turning const eval errors from post to pre-mono for free lifetime-generic constants.
* As the linked issue/comment states, on master there's a difference between `const _: () = panic!();` (pre-mono error) and `const _<'a>: () = panic!();` (post-mono error) which feels wrong.
* With this PR, both become pre-mono ones!
* 2nd commit: Oof, yeah, I missed that in the initial impl!
This doesn't fully address #136204 because I still haven't figured out how & where to properly & best suppress const eval of free constants whose predicates don't hold at the def site. The motivating example is `const _UNUSED: () = () where for<'_delay> String: Copy;` which can also be found over at the tracking issue #113521.
r? compiler-errors or reassign
Rename `{GenericArg,Term}::unpack()` to `kind()`
A well-deserved rename IMO.
r? `@oli-obk` or `@lcnr` (or anyone)
cc `@rust-lang/types,` but I'd be surprised if this is controversial.
Defer evaluating type system constants when they use infers or params
Split out of #137972, the parts necessary for associated const equality and min generic const args to make progress and have correct semantics around when CTFE is invoked. According to a [previous perf run](https://perf.rust-lang.org/compare.html?start=93257e2d20809d82d1bc0fcc1942480d1a66d7cd&end=01b4cbf0f47c3f782330db88fa5ba199bba1f8a2&stat=instructions:u) of adding the new `const_arg_kind` query we should expect minor regressions here.
I think this is acceptable as we should be able to remove this query relatively soon once mgca is more complete as we'll then be able to implement GCE in terms of mgca and rip out `GCEConst` at which point it's trivial to determine what kind of anon const we're dealing with (either it has generics and is a repeat expr hack, or it doesnt and is a normal anon const).
This should only affect unstable code as we handle repeat exprs specially and those are the only kinds of type system consts that are allowed to make use of generic parameters.
Fixes#133066Fixes#133199Fixes#136894Fixes#137813
r? compiler-errors
HIR ty lowering: Clean up & refactor the lowering of type-relative paths
While rebasing #126651 I realized that HIR ty lowering could benefit from some *spring cleaning* now that it's been extended to handle RTN and mGCA paths.
More seriously, similar to my merged PR #118668 which unified the handling of all *associated item constraints* (assoc ty, const (ACE) & fn (RTN)), this PR (commit 695fcf517d) partially[^1] deduplicates the resolution code for all *type-relative paths* (assoc ty, const (mGCA) & fn (RTN)).
**Why**? DRY'ing that part of the code means PR #126651 will automatically support RTN paths like `Ty::AssocTy::assoc_fn(..)` and it also implies shared diagnostic code and thus better diagnostics for RTN.
---
The other commits represent cleanups, renamings, moves. More notably, I've renamed path lowering methods to be a lot more descriptive, so ones lowering `QPath(Resolved)` paths now have `_resolved_` in their name and ones lowering `QPath(TypeRelative)` paths now have `_type_relative_` in their name. This should make it stupidly obvious what their purpose is.
---
Best reviewed commit by commit. The changes are close to trivial but the diff might make it look hairier.
r? compiler-errors
[^1]: Sadly, I couldn't unify as much compared to the other PR without introducing unnecessary `unreachable!()`s or rendering the code otherwise illegible with flags and micro helper traits.
make `rustc_attr_parsing` less dominant in the rustc crate graph
It has/had a glob re-export of `rustc_attr_data_structures`, which is a crate much lower in the graph, and a lot of crates were using it *just* (or *mostly*) for that re-export, while they can rely on `rustc_attr_data_structures` directly.
Previous graph:
Graph with this PR:
The first commit keeps the re-export, and just changes the dependency if possible. The second commit is the "breaking change" which removes the re-export, and "explicitly" adds the `rustc_attr_data_structures` dependency where needed. It also switches over some src/tools/*.
The second commit is actually a lot more involved than I expected. Please let me know if it's a better idea to back it out and just keep the first commit.
Merge mir query analysis invocations
r? `@ghost`
same thing as https://github.com/rust-lang/rust/pull/140854 just a different set of queries
Doing this in general has some bad cache coherence issues because the query caches are laid out in Vec<QueryResult> lists per query where each index refers to a DefId in the same order as we're iterating. Iterating two or more lists at the same time does have cache issues, so I want to poke a bit at it to see if we can't merge just a few of them at a time.
MIR borrowck taints its output if an obligation fails. This could then cause
`check_coroutine_obligations` to silence its error, causing us to not emit
and actual error and ICE.
Remove manual WF hack
We do not need this hack anymore since we fixed the candidate selection problems with `Sized` bounds. We prefer built-in sized bounds now since #138176, which fixes the only regression this hack was intended to fix.
While this theoretically is broken for some code, for example, when there a param-env bound that shadows an impl or built-in trait, we don't see it in practice and IMO it's not worth the burden of having to maintain this wart in `compare_method_predicate_entailment`.
The code that regresses is, for example:
```rust
trait Bar<'a> {}
trait Foo<'a, T> {
fn method(&self)
where
Self: Bar<'a>;
}
struct W<'a, T>(&'a T)
where
Self: Bar<'a>;
impl<'a, 'b, T> Bar<'a> for W<'b, T> {}
impl<'a, 'b, T> Foo<'a, T> for W<'b, T> {
fn method(&self) {}
}
```
Specifically, I don't believe this is really going to be encountered in practice. For this to fail, there must be a where clause in the *trait method* that would shadow an impl or built-in (non-`Sized`) candidate in the trait, and this shadowing would need to be encountered when solving a nested WF goal from the impl self type.
See #108544 for the original regression. Crater run is clean!
r? lcnr
Stage0 bootstrap update
This PR [follows the release process](https://forge.rust-lang.org/release/process.html#master-bootstrap-update-tuesday) to update the stage0 compiler.
The only thing of note is 58651d1b31, which was flagged by clippy as a correctness fix. I think allowing that lint in our case makes sense, but it's worth to have a second pair of eyes on it.
r? `@Mark-Simulacrum`
Use intrinsics for `{f16,f32,f64,f128}::{minimum,maximum}` operations
This PR creates intrinsics for `{f16,f32,f64,f64}::{minimum,maximum}` operations.
This wasn't done when those operations were added as the LLVM support was too weak but now that LLVM has libcalls for unsupported platforms we can finally use them.
Cranelift and GCC[^1] support are partial, Cranelift doesn't support `f16` and `f128`, while GCC doesn't support `f16`.
r? `@tgross35`
try-job: aarch64-gnu
try-job: dist-various-1
try-job: dist-various-2
[^1]: https://www.gnu.org/software///gnulib/manual/html_node/Functions-in-_003cmath_002eh_003e.html
Merge typeck loop with static/const item eval loop
r? `@ghost`
Let's try a small one first. Doing this in general has some bad cache coherence issues because the query caches are laid out in `Vec<QueryResult>` lists per query where each index refers to a `DefId` in the same order as we're iterating. Iterating two or more lists at the same time does have cache issues, so I want to poke a bit at it to see if we can't merge just a few of them at a time.
Only include `dyn Trait<Assoc = ...>` associated type bounds for `Self: Sized` associated types if they are provided
Since #136458, we began filtering out associated types with `Self: Sized` bounds when constructing the list of associated type bounds to put into our `dyn Trait` types. For example, given:
```rust
trait Trait {
type Assoc where Self: Sized;
}
```
After #136458, even if a user writes `dyn Trait<Assoc = ()>`, the lowered ty would have an empty projection list, and thus be equivalent to `dyn Trait`. However, this has the side effect of no longer constraining any types in the RHS of `Assoc = ...`, not implying any WF implied bounds, and not requiring that they hold when unsizing.
After this PR, we include these bounds, but (still) do not require that they are provided. If the are not provided, they are skipped from the projections list.
This results in `dyn Trait` types that have differing numbers of projection bounds. This will lead to re-introducing type mismatches e.g. between `dyn Trait` and `dyn Trait<Assoc = ()>`. However, this is expected and doesn't suffer from any of the deduplication unsoundness from before #136458.
We may want to begin to ignore thse bounds in the future by bumping `unused_associated_type_bounds` to an FCW. I don't want to tangle that up into the fix that was originally intended in #136458, so I'm doing a "fix-forward" in this PR and deferring thinking about this for the future.
Fixes#140645
r? lcnr
Add `DefPathData::OpaqueLifetime` to avoid conflicts for remapped opaque lifetimes
This adds `DefPathData::OpaqueLifetime` to ensure the def paths for remapped opaque lifetimes remain unique.
Fixes https://github.com/rust-lang/rust/issues/140731.
r? ``@oli-obk``
Better error message for late/early lifetime param mismatch
Rework the way we report early-/late-bound lifetime param mismatches to equate the trait and impl signatures using region variables, so that we can detect when a late-bound param is present in the signature in place of an early-bound param, or vice versa.
The diagnostic is a bit more technical, but it's more obviously clear to see what the problem is, even if it's not great at explaining how to fix it. I think this could be improved further, but I still think it's much better than what exists today.
Note to reviewer(s): I'd appreciate if we didn't bikeshed *too* much about this verbiage, b/c I hope it's clear that the old message sucked a lot. I'm happy to file bugs for interested new contributors to improve the messaging further.
Edit(fmease): Fixes https://github.com/rust-lang/rust/issues/33624.
