Fix parallel rustc not being reproducible due to unstable sorts of items
Currently, A tuple `(DefId, SymbolName)` is used to determine the order of items in the final binary. However `DefId` is expected as non-deterministic, which leads to some not reproducible issues under parallel compilation. (See https://github.com/rust-lang/rust/issues/140425#issuecomment-3111802148)
Theoretically, we don't need the sorting because the order of these items is already deterministic.
However, codegen tests reply on the same order of items between in binary and source.
So here we added a new option `codegen-source-order` to indicate whether sorting based on the order in source. For codegen tests, items are sorted according to the order in the source code, whereas in the normal path, no sorting is performed.
Specially, for codegen tests, in preparation for parallel compilation potentially being enabled by default in the future, we use `Span` replacing `DefId` to make the order deterministic.
This PR is purposed to fixrust-lang/rust#140425, but seemly works on rust-lang/rust#140413 too.
This behavior hasn't added into any test until we have a test suit for the parallel frontend. (See https://github.com/rust-lang/rust/pull/143953)
Related discussion: [Zulip](https://rust-lang.zulipchat.com/#narrow/channel/187679-t-compiler.2Fparallel-rustc/topic/Async.20closures.20not.20reproducible.28.23140425.29) https://github.com/rust-lang/rust/pull/144576
Update rust-lang/rust#113349
r? `@oli-obk`
cc `@lqd` `@cramertj` `@matthiaskrgr` `@Zoxc` `@SparrowLii` `@bjorn3` `@cjgillot` `@joshtriplett`
Add `cast_init` and `cast_uninit` methods for pointers
ACP: rust-lang/libs-team#627
Tracking issue: rust-lang/rust#145036
This includes an incredibly low-effort search to find uses that could be switched to using these methods. I only searched for `cast::<\w>` and `cast::<MaybeUninit` because there would otherwise be way too much to look through, and I also didn't modify anything inside submodules/subtrees.
Adjust documentation of `dangling`
I believe the current doc of `dangling` is slightly off as it indicates:
`Note that the pointer value may potentially represent a valid pointer to a T`
The returned pointer has no provenance, so it may not be a valid pointer (except in the case of ZSTs, but I don't think this is what the documentation is trying to warn about).
See: https://rust-lang.zulipchat.com/#narrow/channel/136281-t-opsem/topic/Dangling.20pointers.3A.20definition
The value returned by dangling may never be used to dereference a value that isn't a ZST, even if address equality is detected with that of a valid pointer.
This is a minor fix, but this doc still got me confused for a second
Deprecate RUST_TEST_* env variables
Like with rust-lang/rust#139224, this is a documentation-only deprecation for now. Over time, we can
- warn and then remove on use of unstable environment variables
- warn on use of stable environment variables (no plan to remove due to compatibility)
Longer term, we expect test runners, like `cargo test`, to provide the necessary mechanisms for environmental or persistent configuration (e.g. using cargo config which supports `.cargo/config.toml` as well as environment variables).
This would include:
- `RUST_TEST_THREADS`
- `RUST_TEST_NOCAPTURE`
- `RUST_TEST_SHUFFLE` (unstable)
- `RUST_TEST_SHUFFLE_SEED` (unstable)
The primary outcomes for this change are
- Reducing the scope of what is expected for custom test harnesses to implement
- Reduce the mechanisms that test runners, like `cargo test`, are expected to track when they are being bypassed to protect against negative interactions, e.g. `RUST_TEST_NOCAPTURE=1` when json output is being read.
For testing-devex FCP, see rust-lang/testing-devex-team#10Fixesrust-lang/testing-devex-team#10
History
-------
At each step, I could not find evidence of design discussions on whether to support CLI, env, or both. The first env variable seems to come from the fact that it was being forked out of an existing env variable that had a much wider scope.
At best, this seems like a way to offer a more persistent configuration for these flags but environment variables hidden away in libtest is a bit clunky and this seems like the wrong layer to handle this problem.
**Originally:** `RUST_THREADS` was respected by the Rust runtime and libextra/test got this for free
**2013:** rust-lang/rust#7335 suggested splitting `RUST_TEST_TASKS` out of `RUST_THREADS`. In that issue and the implementation (rust-lang/rust#8823).
**2014:** rust-lang/rust#13374 ask for support to disable capturing of stdout/stderr. `--nocapture` and `RUST_TEST_NOCAPTURE` were added together.
**2015:** rust-lang/rust#23525 renamed `RUST_TEST_TASKS` to `RUST_TEST_THREADS`
**2016:** rust-lang/rust#25636 asked to configure `RUST_TEST_THREADS` via `--test-threads` which was implemented in rust-lang/rust#35414
**2021:** rust-lang/rust#85440 asked for test randomization which was implemented in rust-lang/rust#89082, adding `--shuffle` / RUST_TEST_SHUFFLE` and `--shuffle-seed SEED` / `RUST_TEST_SHUFFLE_SEED`
Potentially relevant issues
---------------------------
- rust-lang/rust#74845
editorconfig: don't trim trailing whitespace in tests
some test snapshot files require trailing whitespace, and previously manually editing those snapshot files (as is required for run-make tests and some platform-specific tests) in an editor with editorconfig support would cause that whitespace to be removed, [causing CI failures like this one](https://github.com/rust-lang/rust/pull/144596#issuecomment-3130442996)
Constify `SystemTime` methods
This is separated out from rust-lang/rust#143949 due to the fact that it touches nontrivial system code. While the same arithmetic methods on `Instant` could be made const, since that type explicitly cannot be constructed at const-time, we don't bother. However, due to the fact that `SystemTime::UNIX_EPOCH` exists, it can be useful to create other anchors at other points in time, and thus these methods should be possible to use in const context.
> Side comment: I would honestly like to just move every single implementation which is a thin wrapper over `Duration` or some integer clock into their own module which is tested on all targets, so that we don't have to worry about code for lower-tier targets not being tested. However, the comments in `std::sys_common` pointing out the desire to move things to `std::sys::common` confused me, particularly due to the fact that `std::sys::common` is taken from the hellish asterisk-import of `std::sys::pal::common` and others.
>
> I think that, for trivial types like this, we should just have a platform-independent module that can be tested on all platforms, to both ensure that the code is properly tested regardless of whether higher-tier platforms needed, and to avoid the extreme mental gymnastics required to determine where in the code it lies.
>
> However, since I'm not on any of the teams maintaining this code and am not involved, I'll settle for just copy-pasting like everyone else has been doing. I just want to express how I'm not happy about that.
**Reviewer note: please only pay attention to the last commit of this PR, until its blocker is merged.**
Since this depends on the previous PR:
`@rustbot` blocked
Add ASCII-related methods from `u8` and `MIN`/`MAX` to `core::ascii::Char`
* Add ASCII-related methods from `u8` to `core::ascii::Char`.
* Add `core::ascii::Char::MIN` and `core::ascii::Char::MAX`.
Set dead_on_return attribute for indirect arguments
Set the dead_on_return attribute (added in LLVM 21) for arguments that are passed indirectly, but not byval.
This indicates that the value of the argument on return does not matter, enabling additional dead store elimination.
From LangRef:
> This attribute indicates that the memory pointed to by the argument is dead upon function return, both upon normal return and if the calls unwinds, meaning that the caller will not depend on its contents. Stores that would be observable either on the return path or on the unwind path may be elided.
>
> Specifically, the behavior is as-if any memory written through the pointer during the execution of the function is overwritten with a poison value upon function return. The caller may access the memory, but any load not preceded by a store will return poison.
>
> This attribute does not imply aliasing properties. For pointer arguments that do not alias other memory locations, noalias attribute may be used in conjunction. Conversely, this attribute always implies dead_on_unwind.
>
> This attribute cannot be applied to return values.
This fixes parts of https://github.com/rust-lang/rust/issues/96497.
Consolidate stage directories and group logs in bootstrap
My post-stage-0-redesign bootstrap fixes aren't done yet, but I think that enough steps have been migrated to the new system that it makes sense to actually modify the directories on disk, and what gets printed when bootstrap runs, so that it actually corresponds to the new system. Before, the printed stages didn't always make sense.
This PR:
- Fixes the numbering of `stageN` directories in the build directory. It was not corresponding to the correct stages before; notice that I did not modify `bootstrap/README.md`, as it was essentially describing what happens after this PR (first commit).
- Unifies all steps that output a build group to use the `Builder::msg` method. It's probably not the final stage, and some of the test steps might not be fully accurate yet, because I didn't fix test step numbering yet, but I think that it's a clear improvement from before, and now that everything uses the same method, we can easily make changes across the board, to ensure that it stays unified (second commit).
r? `@jieyouxu`
try-job: dist-x86_64-msvc
try-job: dist-x86_64-linux
