bootstrap: Set the dylib path when building books with rustdoc
The library path is needed when the toolchain has been configured with
`[rust] rpath = false`. Otherwise, building the reference book will get
an error when it tries to run rustdoc, like:
rustdoc: error while loading shared libraries: librustc_driver-2ec457c3b8826b72.so
A couple tests were using `BOOTSTRAP_CARGO` with `-Zbuild-std`, but that
stage0 cargo might not always be in sync with in-tree changes. In
particular, those tests started failing on the beta branch because the
older cargo couldn't find the library `Cargo.lock`, and then couldn't
build the latest version of `compiler_builtins` that had nightly changes.
The library path is needed when the toolchain has been configured with
`[rust] rpath = false`. Otherwise, building the reference book will get
an error when it tries to run rustdoc, like:
rustdoc: error while loading shared libraries: librustc_driver-2ec457c3b8826b72.so
set `download-ci-llvm = true` by default on "library" and "tools" profiles
It's very rare for developers to need to modify LLVM, so "if-unchanged" isn't a good default for "tools" and "library" profiles since it fetches the LLVM submodule to track changes.
Rollup of 5 pull requests
Successful merges:
- #130138 (bootstrap: Print more debug info when `find_initial_libdir` fails)
- #130199 (Don't call closure_by_move_body_def_id on FnOnce async closures in MIR validation)
- #130302 (add llvm-bitcode-linker and llvm-tools bins to ci-rustc's sysroot)
- #130306 (avoid updating LLVM submodule during bootstrap unit tests)
- #130317 (`ProjectionElem` and `UnOp`/`BinOp` dont need to be `PartialOrd`/`Ord`)
r? `@ghost`
`@rustbot` modify labels: rollup
avoid updating LLVM submodule during bootstrap unit tests
To test this, make sure you don't have `src/llvm-project` fetched and then set `llvm.download-ci-llvm=true` and run `x test bootstrap`.
bootstrap: Print more debug info when `find_initial_libdir` fails
From looking at the failure messages printed by #129775, my hypothesis is that `rustc --print=sysroot` sometimes prints the wrong path when the rustc executable is hardlinked in multiple places, at least on my macOS system.
However, currently I don't have any concrete evidence of this actually happening. This PR therefore expands on #129775 by printing even more information on failure (including the actual rustc path), in the hope that when the failure next occurs we can confirm or reject the hypothesis that `rustc --print=sysroot` is printing the wrong path.
[bootstrap] Add support for building gcc and libgccjit
As `@eholk` summarized below:
> From my understanding, this change would add libgccjit as an optional component to the Rust distribution. This library is licensed under GPLv2 and currently we do not have any other components under that license so it would be a new license, and one that is generally more restrictive than the other licenses we use.
It'll greatly improve the experience for anyone wanting to work on the GCC backend from the compiler.
Should help with https://github.com/rust-lang/rust/issues/124172.
Will unblock #124353.
r? `@Kobzol`
skip target sanity check when it's a `local-rebuild`
Running the stage0 target sanity check on the newly built compiler can result in errors and incorrect assumptions.
Resolves#130242
Running the stage0 target sanity check on the newly built compiler can result
in errors and incorrect assumptions.
Signed-off-by: onur-ozkan <work@onurozkan.dev>
refactor merge base logic and fix `x fmt`
When remote upstream is not configured, using [get_git_modified_files](38e3a5771c/src/tools/build_helper/src/git.rs (L114)) to find modified files fails because [get_rust_lang_rust_remote](38e3a5771c/src/tools/build_helper/src/git.rs (L46-L48)) can not resolve "rust-lang/rust" from the git output. The changes in this PR makes bootstrap to find the latest bors commit, treating it as the "closest upstream commit" so that the change tracker logic can use it to find the diffs.
In addition, [skips formatting](e392454483) if there are no modified files.
Fixes#130147
It's very rare for developers to need to modify LLVM,
so "if-unchanged" isn't a good default since it fetches
the LLVM submodule to track changes.
Signed-off-by: onur-ozkan <work@onurozkan.dev>
miri-test-libstd: add missing BOOTSTRAP_ARGS
Note sure if BOOTSTRAP_ARGS will make any difference here, but all the other x.py invocations have it and I did not *deliberately* leave it away when I added these, so... probably best to add them?
Also don't unnecessarily set BOOTSTRAP_SKIP_TARGET_SANITY while we are at it.
use `download-ci-llvm=true` in the default compiler config
1ca2708e77 made it so that the `src/llvm-project` submodule has to be checkout for `download-ci-llvm = "if-unchanged"` to know if the submodule has been changed, but that is not required, if the submodule hasn't been checkout it cannot have been modified.
~~This PR restore the previous behavior by only updating the submodule if it has already been checkout.~~
This PR makes `download-ci-llvm = true` check if CI llvm is available and make it the default for the compiler profile, as to prevent unnecessarily checking out `src/llvm-project` with `"if-unchanged"`.
r? `````@onur-ozkan`````
Rollup of 9 pull requests
Successful merges:
- #129929 (`rustc_mir_transform` cleanups, round 2)
- #130022 (Dataflow/borrowck lifetime cleanups)
- #130064 (fix ICE in CMSE type validation)
- #130067 (Remove redundant check in `symlink_hard_link` test)
- #130131 (Print a helpful message if any tests were skipped for being up-to-date)
- #130137 (Fix ICE caused by missing span in a region error)
- #130153 (use verbose flag as a default value for `rust.verbose-tests`)
- #130154 (Stabilize `char::MIN`)
- #130158 (Update books)
r? `@ghost`
`@rustbot` modify labels: rollup
Print a helpful message if any tests were skipped for being up-to-date
When running tests without the `--force-rerun` flag, compiletest will automatically skip any tests that (in its judgement) don't need to be run again since the last time they were run.
This is normally very useful, but can occasionally be confusing, especially in edge-cases where up-to-date checking is not completely accurate (or the test is flaky).
This PR makes bootstrap count the number of tests that were ignored for being up-to-date (via a hard-coded check on the ignore reason), and prints a helpful message when that number is nonzero.
---
Sample output:
```text
test result: ok. 4 passed; 0 failed; 17578 ignored; 0 measured; 0 filtered out; finished in 463.79ms
help: ignored 17295 up-to-date tests; use `--force-rerun` to prevent this
Build completed successfully in 0:00:07
```
