Use `llvm-config` instead of `download-ci-llvm` in PGO script
This should avoid CI breakage when the LLVM stamp is updated, and also it will avoid an unnecessary LLVM download from CI.
r? `@jyn514`
Dont ICE for `dyn* Trait: Trait` (built-in object) goals during selection in new trait solver
We were ICEing too eagerly during selection for `dyn*` goals -- both for dyn unsizing candidates and for built-in object candidates. The former should only be performed on `dyn` objects, but the latter are totally fine.
fix compiletest crash
### Motivation
When running compiler-tests locally for the `wasm32` platform, one test repeatedly crashed. It does not crash on the CI, only locally. Investigation shows that the `compiletest` itself crashes
> panicked-at-attempt-to-subtract-with-overflow
```rust
let mut head = replace(bytes, Vec::new());
let mut middle = head.split_off(HEAD_LEN);
// The following line will panic
let tail = middle.split_off(middle.len() - TAIL_LEN).into_boxed_slice();
let skipped = new_len - HEAD_LEN - TAIL_LEN;
```
### Background
The code in question collects the output of a process. Small output is kept completely, but larger output is kept only partially: the first 160 kB and the last 256 kB.
The code that performs this split crashes if the data size is less than 416 kB. There is an early out based on the "filtered" length, but it is possible that the filtered length is greater than the real length. It seems that this code was written with the assumption that the filtered length is larger than the real length, which is not true in general.
When running CI tests locally using `src/ci/docker/run.sh`, the filtered folder is `/checkout`, which is shorter than the placeholder length of 32 bytes.
### Note
This PR should not change any behaviour. It only adds an early our for a case which will definitely crash (at least if compiletest is build with integer checks).
Note that an early out makes sense here: If the real data is too small, it does not sense to split it.
Fix the tests-listing-format-json test on Windows
tests/ui/test-attrs/tests-listing-json-format.rs was failing on Windows because each path in the json-formatted output contained "\\\\" instead of "\\". `runtest::TestCx::normalize_output` already checks the compile flags for json-related arguments to handle this case, so I added an equivalent check for the new run flag.
Move `TyCtxt::mk_x` to `Ty::new_x` where applicable
Part of rust-lang/compiler-team#616
turns out there's a lot of places we construct `Ty` this is a ridiculously huge PR :S
r? `@oli-obk`
Rollup of 9 pull requests
Successful merges:
- #111119 (style-guide: Add chapter about formatting for nightly-only syntax)
- #112791 (llvm ffi: Expose `CallInst->setTailCallKind`)
- #113145 (style-guide: Document newline rules for assignment operators)
- #113163 (Add a regression test for #112895)
- #113332 (resolve: Use `Interned` for some interned structures)
- #113334 (Revert the lexing of `c"…"` string literals)
- #113350 (Fix the issue of wrong diagnosis for extern pub fn)
- #113371 (Fix submodule handling when the current branch is named after a tag)
- #113384 (style-guide: Clarify grammar for small patterns (not a semantic change))
r? `@ghost`
`@rustbot` modify labels: rollup
style-guide: Clarify grammar for small patterns (not a semantic change)
The grammar as written feels ambiguous and confusing, in large part
because it uses square brackets and commas in the names of
non-terminals. Rewrite it to avoid symbols in the names of
non-terminals, and to instead wrap terminals in backquotes.
Also rename "smallntp" to "small_no_tuple" to make it self-describing.
Fix submodule handling when the current branch is named after a tag
If:
1. The current branch has the same name as git tag, and
2. The current branch is set to track a remote other than `origin`, and
3. We try to update a submodule
then we'll get the following error:
```
; x c
Updating submodule src/doc/reference
remote: Total 0 (delta 0), reused 0 (delta 0), pack-reused 0
fatal: 'personal' does not appear to be a git repository
fatal: Could not read from remote repository.
```
The problem is that 1. causes `git symbolic-ref --short HEAD` to try and disambiguate the branch from the tag using `heads/branch-name`, which breaks a previous workaround for a bug in `git submodule update` that uses the wrong remote.
Adapt the workaround to strip `heads/` from the output.
Revert the lexing of `c"…"` string literals
Fixes \[after beta-backport\] #113235.
Further progress is tracked in #113333.
This PR *manually* reverts parts of #108801 (since a git-revert would've been too coarse-grained & messy)
and git-reverts #111647.
CC `@fee1-dead` (#108801) `@klensy` (#111647)
r? `@compiler-errors`
`@rustbot` label F-c_str_literals beta-nominated
style-guide: Document newline rules for assignment operators
The style guide gives general rules for binary operators including
assignment, and one of those rules says to put the operator on the
subsequent line; the style guide needs to explicitly state the exception
of breaking *after* assignment operators rather than before.
This is already what rustfmt does and what users do; this fixes the
style guide to match the expected default style.
Specialize `try_destructure_mir_constant` for its sole user (pretty printing)
We can't remove the query, as we need to invoke it from rustc_middle, but can only implement it in mir interpretation/const eval.
r? `@RalfJung` for a first round.
While we could move all the logic into pretty printing, that would end up duplicating a bit of code with const eval, which doesn't seem great either.
The grammar as written feels ambiguous and confusing, in large part
because it uses square brackets and commas in the names of
non-terminals. Rewrite it to avoid symbols in the names of
non-terminals, and to instead wrap terminals in backquotes.
enable test_join test in Miri
Miri for quite a while now has a hack to support self-referential generators: non-`Unique` mutable references are exempt from aliasing conditions. So we can run this test now. (It passes.)
Also extend a comment in a Vec test, while I am at it.
Rollup of 8 pull requests
Successful merges:
- #113010 (rust-installer & rls: remove exclusion from rustfmt & tidy )
- #113317 ( -Ztrait-solver=next: stop depending on old solver)
- #113319 (`TypeParameterDefinition` always require a `DefId`)
- #113320 (Add some extra information to opaque type cycle errors)
- #113321 (Move `ty::ConstKind` to `rustc_type_ir`)
- #113337 (Winnow specialized impls during selection in new solver)
- #113355 (Move most coverage code out of `rustc_codegen_ssa`)
- #113356 (Add support for NetBSD/riscv64 aka. riscv64gc-unknown-netbsd.)
r? `@ghost`
`@rustbot` modify labels: rollup
If:
1. The current branch has the same name as git tag, and
2. The current branch is set to track a remote other than `origin`, and
3. We try to update a submodule
then we'll get the following error:
```
; x c
Updating submodule src/doc/reference
remote: Total 0 (delta 0), reused 0 (delta 0), pack-reused 0
fatal: 'personal' does not appear to be a git repository
fatal: Could not read from remote repository.
```
The problem is that 1. causes `git symbolic-ref --short HEAD` to try and disambiguate the branch
from the tag using `heads/branch-name`, which breaks a previous workaround for a bug in `git
submodule update` that uses the wrong remote.
Adapt the workaround to strip `heads/` from the output.
Move most coverage code out of `rustc_codegen_ssa`
*This is one step in my larger coverage refactoring ambitions described at <https://github.com/rust-lang/compiler-team/issues/645>.*
The backend implementation of coverage instrumentation was originally split between SSA and LLVM, perhaps in the hopes that it could be used by other backends.
In practice, this split mostly just makes the coverage implementation harder to navigate and harder to modify. It seems unlikely that any backend will actually implement coverage instrumentation in the foreseeable future, especially since many parts of the existing implementation (outside the LLVM backend) are heavily tied to the specific details of LLVM's coverage instrumentation features.
The current shared implementation of `codegen_coverage` is heavily tied to the details of `StatementKind::Coverage`, which makes those details difficult to change. I have reason to want to change those details as part of future fixes/improvements, so this will reduce the amount of interface churn caused by those later changes.
---
This is intended to be a pure refactoring change, with no changes to actual behaviour. All of the “added” code has really just been moved from other files.
Winnow specialized impls during selection in new solver
We need to be able to winnow impls that are specialized by more specific impls in order for codegen to be able to proceed.
r? ``@lcnr``
Move `ty::ConstKind` to `rustc_type_ir`
Needed this in another PR for custom debug impls, and this will also be required to move the new solver into a separate crate that does not use `TyCtxt` so that r-a and friends can depend on the trait solver.
Rebased on top of #113325, only the second and third commits needs reviewing
Add some extra information to opaque type cycle errors
Plus a bunch of cleanups.
This should help users debug query cycles due to auto trait checking. We'll probably want to fix cycle errors in most (or all?) cases by looking at the current item's hidden types (new solver does this), and by delaying the auto trait checks to after typeck.
`TypeParameterDefinition` always require a `DefId`
the `None` case never actually reaches diagnostics so it feels better for diagnostics to be able to rely on the `DefId` being there, cc #113310
