For `check` builds, since we only need to perform a sanity check on
the host target, this patch skips target sanity checks on non-host targets.

Signed-off-by: onur-ozkan <work@onurozkan.dev>

This interning is pointless and only makes the code more complex.

The only remaining use of interning is `TargetSelection`, for which I
left a comment.

I don't think it is necessary anymore. As I understand it from issue
39504 the original problem was that rustbuild changed a hardlink in the
cargo build dir to point to copy in the sysroot while cargo may have
hardlinked it to the original first. I don't think this happens anymore
and as such this workaround is no longer necessary.

This will make it easier to report rustc versions for older metadata
formats.

…larsan68

Avoid some interning in bootstrap

This interning is pointless and only makes the code more complex.

The only remaining use of interning is `TargetSelection`, for which I left a comment.

…cjgillot

Misc improvements to non local defs lint implementation

This PR is a collection of small improvements I found when I [needlessly tried](https://www.github.com/rust-lang/rust/pull/120393#issuecomment-1971787475) to fix a "perf-regression" in the lint implementation.

I recommend looking at each commit individually.

Add a tidy check that checks whether the fluent slugs only appear once

As ``@Nilstrieb`` said in rust-lang#121828 (comment):
> Might make sense to have a tidy check that checks whether the fluent slugs only appear once in the source code and lint for that
there's a tidy check already for sorting

We can get the tidy check error:
```
tidy check
tidy error: /path/to/rust/compiler/rustc_const_eval/messages.ftl: message `const_eval_invalid_align` is not used
tidy error: /path/to/rust/compiler/rustc_lint/messages.ftl: message `lint_trivial_untranslatable_diag` is not used
tidy error: /path/to/rust/compiler/rustc_parse/messages.ftl: message `parse_invalid_literal_suffix` is not used
tidy error: /path/to/rust/compiler/rustc_infer/messages.ftl: message `infer_need_type_info_in_coroutine` is not used
tidy error: /path/to/rust/compiler/rustc_passes/messages.ftl: message `passes_expr_not_allowed_in_context` is not used
tidy error: /path/to/rust/compiler/rustc_passes/messages.ftl: message `passes_layout` is not used
tidy error: /path/to/rust/compiler/rustc_parse/messages.ftl: message `parse_not_supported` is not used
```

r? ``@Nilstrieb``

…eck, r=albertlarsan68

skip sanity check for non-host targets in `check` builds

For `check` builds, since we only need to perform a sanity check on the host target, this patch skips target sanity checks on non-host targets.

For more context, see rust-lang#121519 (comment)

cc `@saethlin`

…latest-edition, r=cjgillot

Eagerly translate `HelpUseLatestEdition` in parser diagnostics

Fixes rust-lang#122130.

This makes me suspicious of these other two usage of  `add_to_diagnostic()`. Would they *also* crash? I haven't attempted to construct test cases for them.

```
compiler/rustc_parse/src/parser/expr.rs
3453:            errors::HelpUseLatestEdition::new().add_to_diagnostic(e);

compiler/rustc_hir_typeck/src/expr.rs
2603:            HelpUseLatestEdition::new().add_to_diagnostic(&mut err);
```

This also seems like a footgun?

ci: add a runner for vanilla LLVM 18

For CI cost, this can be seen as replacing the llvm-15 runner we dropped in rust-lang#117947.

Also, I've set `IS_NOT_LATEST_LLVM` in the llvm-17 runner, since that's not the latest anymore.

…trochenkov

Remove a workaround for a bug

I don't think it is necessary anymore. As I understand it from issue 39504 the original problem was that rustbuild changed a hardlink in the cargo build dir to point to copy in the sysroot while cargo may have hardlinked it to the original first. I don't think this happens anymore and as such this workaround is no longer necessary.

Split out of rust-lang#120855

…, r=petrochenkov

Move metadata header and version checks together

This will make it easier to report rustc versions for older metadata formats.

Split out of rust-lang#120855

Some tweaks to the parallel query cycle handler

This renames `deadlock` to `break_query_cycles`. The abort logic is moved next to the thread spawning and gives the thread a name.

Fixes rust-lang#122035.

r? ``@oli-obk``

Fix typo in `VisitorResult`

r? ``@oli-obk``

library/core: fix a comment, and a cfg(miri) warning

Just two things I noticed while working on another PR.