A `ChunkedBitSet` has to be at least 2048 bits for it to outperform a
`BitSet`, because that's the chunk size. The largest `SparseBitMatrix`
encountered when compiling the compiler and the entire rustc-perf
benchmark suite is less than 600 bits.
This change is a tiny perf win, but the motivation is more about
avoiding uses of `ChunkedBitSet` outside of `MixedBitSet`.
The test change is necessary to avoid hitting the `<BitSet<T> as
BitRelations<ChunkedBitSet<T>>>::subtract` method that has
`unimplemented!` in its body and isn't otherwise used.
It's a performance win because `MixedBitSet` is faster and uses less
memory than `ChunkedBitSet`.
Also reflow some overlong comment lines in
`lint_tail_expr_drop_order.rs`.
It just uses `BitSet` for small/medium sizes (<= 2048 bits) and
`ChunkedBitSet` for larger sizes. This is good because `ChunkedBitSet`
is slow and memory-hungry at smaller sizes.
Rollup of 8 pull requests
Successful merges:
- #133737 (Include LLDB and GDB visualizers in MSVC distribution)
- #133774 (Make CoercePointee errors translatable)
- #133831 (Don't try and handle unfed `type_of` on anon consts)
- #133847 (Remove `-Zshow-span`.)
- #133849 (coverage: Use a separate counter type and simplification step during counter creation)
- #133850 (Avoid `opaque type not constrained` errors in the presence of other errors)
- #133851 (Stop git from merging generated files)
- #133856 (Update sysinfo version to 0.33.0)
r? `@ghost`
`@rustbot` modify labels: rollup
Stop git from merging generated files
See the relevant documentation for details: https://git-scm.com/docs/gitattributes#Documentation/gitattributes.txt-Unset-1-1-1
This will cause `.stderr`, `.stdout`, `.mir`, and `.fixed` files to stop generating conflict markers and instead just pick the version from your branch, not the one you're rebasing over. Git will still raise a conflict and thus stop a rebase at that commit, but it avoids having to deal with conflict markers in files where we just bless them away anyway.
Avoid `opaque type not constrained` errors in the presence of other errors
pulled out of https://github.com/rust-lang/rust/pull/128440
These errors carry no new information if the opaque type was actually used in a constraining (but erroneous) way somewhere.
coverage: Use a separate counter type and simplification step during counter creation
When instrumenting a function's MIR for coverage, there is a point where we need to decide, for each node in the control-flow graph, whether its execution count will be tracked by a physical counter, or by an expression that combines physical counters from other parts of the graph.
Currently the code for doing that is heavily tied to the final form of the LLVM coverage mapping format, and performs some important simplification steps on-the-fly. These factors make the code extremely difficult to modify without breaking or massively worsening the resulting coverage-instrumentation metadata.
---
This PR aims to improve that situation somewhat by adding an extra intermediate representation between the code that chooses how each node will be counted, and the code that converts those decisions into actual tables of physical counters and trees of counter expressions.
As part of doing that, some of the simplifications that are currently performed during the main counter creation step have been pulled out into a separate step.
In most cases the resulting coverage metadata is equivalent, slightly better, or slightly worse. The biggest outlier is `counters.rs`, where the coverage metadata ends up about 10% larger. This seems to be the result of the new approach having less subexpression sharing (because it relies on flatten-sort-cancel), and therefore being less effective at taking advantage of MIR optimizations to replace counters for unused control-flow with zeroes. I think the modest downside is acceptable in light of the future possibilities opened up by this decoupling.
Remove `-Zshow-span`.
It's very old (added in #12087). It's strange, and it's not clear what its use cases are. It only works with the crate root file because it runs before expansion. I suspect it won't be missed.
r? `@estebank`
Don't try and handle unfed `type_of` on anon consts
The `type_of` query for anon consts in the type system is actually implemented by feeding the return value during hir ty lowering, not the hir-based logic in `const_arg_anon_type_of`. The HIR based logic is incomplete (doesn't handle all hir nodes) and also generally wrong to call (re-lowers HIR or invokes typeck which can result in query cycles).
r? `@compiler-errors`
Make CoercePointee errors translatable
Tracked by #123430
Just in case that a translatable error message would become a blocker to stabilization, this PR switches over to fluent error messages, which also slightly improve the wordings and use more accurate span information.
cc `@Darksonn` `@traviscross`
Include LLDB and GDB visualizers in MSVC distribution
MSVC distributions currently don't include the lldb or GDB python files. MSVC and LLDB/GDB are not mutually exclusive (and end up being a common case with vscode + codelldb/lldb-dap), so they should probably be included.
the existing visualizers currently only partially work on MSVC due to the differences in how the debug info is generated, but they also only partially work on GNU anyway - both of which are actively being fixed.
It's very old (added in #12087). It's strange, and it's not clear what
its use cases are. It only works with the crate root file because it
runs before expansion. I suspect it won't be missed.
Use `eprintln` instead of `println` in bootstrap/compiletest/tidy
A big unconditional CTRL-F replace to start with to check if there's anything that CI expects to be on stdout
r? `@jieyouxu`
Improve code for FileName retrieval in rustdoc
Some calls were performed twice (first in `is_real_and_local` and then in the function calling it). Also the `FileName` was matched on a few times too.
r? `@notriddle`
stop replacing bivariant args with `'static` when computing closure requirements
It is unnecessary, these get constrained when checking that the opaque type is well-formed.
It also results in the opaque type no longer being well formed. If you've got `fn foo<'a>() -> impl Sized + 'a` the opaque is `type Opaque<'a, 'aDummy> where 'a: 'aDummy, 'aDummy: 'a` where `'aDummy` is bivariant. If we call `foo::<'b>()` inside of a closure and its return type ends up in a type test, we start out with the WF `Opaque<'b, 'b>`, and then replace the bivariant `'b` with `'static`. `Opaque<'b, 'static>` is no longer well-formed. Given how these type tests are used, I don't think this caused any practical issues.
r? types
Fix MutVisitor's default implementations to visit Stmt's and BinOp's spans
The `Stmt` case is a bug introduced almost certainly unintentionally by https://github.com/rust-lang/rust/pull/126993. The code _used_ to visit and mutate `span` correctly, but got changed as follows by that PR. Notice how `span` is **copied** into the output by `|kind| Stmt { id, kind, span }` which happens after the mutation in the correct code (red) and before the mutation in the incorrect code (green).
```diff
pub fn noop_flat_map_stmt<T: MutVisitor>(
Stmt { kind, mut span, mut id }: Stmt,
vis: &mut T,
) -> SmallVec<[Stmt; 1]> {
vis.visit_id(&mut id);
- vis.visit_span(&mut span);
let stmts: SmallVec<_> = noop_flat_map_stmt_kind(kind, vis)
.into_iter()
.map(|kind| Stmt { id, kind, span })
.collect();
if stmts.len() > 1 {
panic!(...);
}
+ vis.visit_span(&mut span);
stmts
}
```
rustdoc: Rename `set_back_info` to `restore_module_data`.
Follow-up to #133345, r? `@GuillaumeGomez`
Most of the references to `info` got removed as it was clear that `module_data` makes more sense here. Makes it clearer that `save_module_data` and `restore_module_data` are a pair.
Update `NonZero` and `NonNull` to not field-project (per MCP#807)
https://github.com/rust-lang/compiler-team/issues/807#issuecomment-2506098540 was accepted, so this is the first PR towards moving the library to not using field projections into `[rustc_layout_scalar_valid_range_*]` types.
`NonZero` was already using `transmute` nearly everywhere, so there are very few changes to it.
`NonNull` needed more changes, but they're mostly simple, changing `.pointer` to `.as_ptr()`.
r? libs
cc #133324, which will tidy up some of the MIR from this a bit more, but isn't a blocker.
compiletest: explain that UI tests are expected not to compile by default
```
error: ui test compiled successfully!
```
is not a very useful message for someone new to the test suite, so change the wording and add a note to explain
Update wasm-component-ld to 0.5.11
This pulls in an update that supports ``@`-files` used to pass arguments to linkers to fix invocations on Windows that are large.
Closes#133649
[AIX] change AIX default codemodel=large
On AIX, for most libraries, we run out of Table of Contents (TOC) offsets very quickly due to the default 16-bit offset limit. We want the large code model should be used as the default to provide more TOC entries so the end user does not have to specify `-Ccode-model=large` for all their packages. This is even more of an issue with ThinLTO as the ThinLTO globals can very quickly use all available TOC entry.
In addition, on AIX, code with different code-model across different compilation units will not cause undefined behavior, so this is safe to do.
Remove `generic_associated_types_extended` feature gate
This PR retires nightly support for the `generic_associated_types_extended` feature. This feature hasn't received much attention in the last two years or so, and I believe the feature still remains both unsound and ICEy to use. I think that if we were to redesign and reimplement it, we'd want to first figure out how to implement it soundly, but in the mean time I'd prefer to clean this up.
r? ``@lcnr`` cc ``@jackh726`` who added this feature gate I think
improve TagEncoding::Niche docs, sanity check, and UB checks
Turns out the `niche_variants` range can actually contain the `untagged_variant`. We should report this as UB in Miri, so this PR implements that.
Also rename `partially_check_layout` to `layout_sanity_check` for better consistency with how similar functions are called in other parts of the compiler.
Turns out my adjustments to the transmutation logic also fix https://github.com/rust-lang/rust/issues/126267.
a release operation synchronizes with an acquire operation
Change:
1. `Calls to park _synchronize-with_ calls to unpark` to `Calls to unpark _synchronize-with_ calls to park`
2. `park synchronizes-with _all_ prior unpark operations` to `_all_ prior unpark operations synchronize-with park`
