- `new_zeroed` variants move to `new_zeroed_alloc`
- the `write` fn moves to `box_uninit_write`

The remainder will be stabilized in upcoming patches, as
it was decided to only stabilize `uninit*` and `assume_init`.

The current `build.rs` will automatically skip source files that don't exist.
An unfortunate side-effect is that if _no_ files could be found (e.g. because
the directory was wrong), the build fails with a mysterious linker error.

We can reduce the awkwardness of this by explicitly checking that at least one
source file was found.

also make it fail if there's a compression issue

zlib is seemingly always enabled, so we can test it unconditionally

it now checks zlib and zstd, via rustc and rust-lld

move it where it's used, and name it like the other scripts

but warn about it

Link: rust-lang#129416
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>

…avidtwco

Implement `-Z embed-source` (DWARFv5 source code embedding extension)

Implement rust-lang/compiler-team#764 MCP which adds an unstable flag that exposes LLVM's [DWARFv5 source code embedding](https://dwarfstd.org/issues/180201.1.html) support.

…jieyouxu

Migrate `libtest-thread-limit` `run-make` test to rmake

Part of rust-lang#121876 and the associated [Google Summer of Code project](https://blog.rust-lang.org/2024/05/01/gsoc-2024-selected-projects.html).

Please try, but **only if normal CI is green**:

// try-job: armhf-gnu // <- failed on this
try-job: aarch64-gnu

More work on `zstd` compression

r? `@Kobzol` as we've discussed this.

This is a draft to show the current approach of supporting zstd in compiletest, and making the tests using it unconditional.

Knowing whether llvm/lld was built with `LLVM_ENABLE_ZSTD` is quite hard, so there are two strategies. There are details in the code, and we can discuss this approach. Until we know the config used to build CI artifacts, it seems our options are somewhat limited in any case.

zlib compression seems always enabled, so we only check this in its dedicated test, allowing the test to ignore errors due to zstd not being supported.

The zstd test is made unconditional in what it tests, by relying on `needs-llvm-zstd` to be ignored when `llvm.libzstd` isn't enabled in `config.toml`.

try-job: x86_64-gnu
try-job: x86_64-msvc

Add f16 and f128 to tests/ui/consts/const-float-bits-conv.rs

Fixes rust-lang#129163

try-job: x86_64-msvc
try-job: x86_64-mingw
try-job: i686-msvc
try-job: i686-mingw

Build `library/profiler_builtins` from `ci-llvm` if appropriate

Running all of `tests/coverage` requires the LLVM profiler runtime, which requires setting `build.profiler = true`.

Historically, doing that has required checking out the entire `src/llvm-project` submodule. For compiler contributors who otherwise don't need that submodule (thanks to `download-ci-vm`), that's quite inconvenient.

However, thanks to rust-lang#129116, the downloaded CI LLVM tarball now contains a copy of LLVM's `compiler-rt` directory, which includes all the files needed to build the profiler runtime. So with a little bit of extra logic in bootstrap, we can have `library/profiler_builtins` look for the `compiler-rt` files in `ci-llvm` instead of the `src/llvm-project` submodule.

…tabilization, r=dtolnay

library: Move unstable API of new_uninit to new features

- `new_zeroed` variants move to `new_zeroed_alloc`
- the `write` fn moves to `box_uninit_write`

The remainder will be stabilized in upcoming patches, as it was decided to only stabilize `uninit*` and `assume_init`.

rustc: Simplify getting sysroot library directory

It was very non-obvious that `sess.target_tlib_path`, `make_target_lib_path(...)`, and `sess.target_filesearch(...).search_paths()` result in the same sysroot library directory paths.
They are however, indeed the same, because `sess.target_tlib_path` is initialized to `make_target_lib_path(...)` on `Session` creation, and they are used interchangeably.

There are still some redundant calls to `make_target_lib_path` and other inconsistent ways to obtain sysroot directories, but fixing that requires some behavior changes, while this PR is a pure refactoring.
Some places in the compiler even disagree on the number of sysroots - 1 (explicit `--sysroot` *or* default sysroot), 2 (explicit `--sysroot` *and* default sysroot), or an unclear number of `sysroot_candidates` every of which is considered.
The logic currently using `sess.target_tlib_path` or equivalents assumes one sysroot.