Use BCryptGenRandom instead of RtlGenRandom on Windows.
This removes usage of RtlGenRandom on Windows, in favour of BCryptGenRandom.
BCryptGenRandom isn't available on XP, but we dropped XP support a while ago.
Avoid allocations and copying in Vec::leak
The [`Vec::leak`] method (#62195) is currently implemented by calling `Vec::into_boxed_slice` and `Box::leak`. This shrinks the vector before leaking it, which potentially causes a reallocation and copies the vector's contents.
By avoiding the conversion to `Box`, we can instead leak the vector without any expensive operations, just by returning a slice reference and forgetting the `Vec`. Users who *want* to shrink the vector first can still do so by calling `shrink_to_fit` explicitly.
**Note:** This could break code that uses `Box::from_raw` to “un-leak” the slice returned by `Vec::leak`. However, the `Vec::leak` docs explicitly forbid this, so such code is already incorrect.
[`Vec::leak`]: https://doc.rust-lang.org/stable/std/vec/struct.Vec.html#method.leak
Optimize VecDeque::append
Optimize `VecDeque::append` to do unsafe copy rather than iterating through each element.
On my `Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz`, the benchmark shows 37% improvements:
```
Master:
custom-bench vec_deque_append 583164 ns/iter
custom-bench vec_deque_append 550040 ns/iter
Patched:
custom-bench vec_deque_append 349204 ns/iter
custom-bench vec_deque_append 368164 ns/iter
```
Additional notes on the context: this is the third attempt to implement a non-trivial version of `VecDeque::append`, the last two are reverted due to unsoundness or regression, see:
- https://github.com/rust-lang/rust/pull/52553, reverted in https://github.com/rust-lang/rust/pull/53571
- https://github.com/rust-lang/rust/pull/53564, reverted in https://github.com/rust-lang/rust/pull/54851
Both cases are covered by existing tests.
Signed-off-by: tabokie <xy.tao@outlook.com>
Rollup of 7 pull requests
Successful merges:
- #86011 (move implicit `Sized` predicate to end of list)
- #89821 (Add a strange test for `unsafe_code` lint.)
- #89859 (add dedicated error variant for writing the discriminant of an uninhabited enum variant)
- #89870 (Suggest Box::pin when Pin::new is used instead)
- #89880 (Use non-checking TLS relocation in aarch64 asm! sym test.)
- #89885 (add long explanation for E0183)
- #89894 (Remove unused dependencies from rustc_const_eval)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Use non-checking TLS relocation in aarch64 asm! sym test.
The checking variant ensures that the offset required is not larger than 12 bits - hence we wouldn't ever need the upper 12 bits.
It's unlikely to ever fail in this small test but this is technically correct.
This was noticed incidentally when we found that LLD doesn't support the `tprel_lo12` relocation, even though LLVM can apparently generate it when using `-mtls-size=12`.
Suggest Box::pin when Pin::new is used instead
This fixes an incorrect diagnostic.
**Based on #89390**; only the last commit is specific to this PR. "Ignore whitespace changes" also helps here.
add dedicated error variant for writing the discriminant of an uninhabited enum variant
This is conceptually different from hitting an `Unreachable` terminator. Also add some sanity check making sure we don't write discriminants of things that do not have discriminants.
r? ``@oli-obk``
Add a strange test for `unsafe_code` lint.
The current behavior is a little surprising to me. I'm not sure whether people would change it, but at least let me document the current behavior with a test.
I learnt about this from the [totally-speedy-transmute](https://docs.rs/totally-speedy-transmute) crate.
cc #10599 the original implementation pr.
move implicit `Sized` predicate to end of list
In `Bounds::predicates()`, move the implicit `Sized` predicate to the
end of the generated list. This means that if there is an explicit
`Sized` bound, it will be checked first, and any resulting
diagnostics will have a more useful span.
Fixes#85998, at least partially. ~~Based on #85979, but only the last 2 commits are new for this pull request.~~ (edit: rebased) A full fix would need to deal with where-clauses, and that seems difficult. Basically, predicates are being collected in multiple stages, and there are two places where implicit `Sized` predicates can be inserted: once for generic parameters, and once for where-clauses. I think this insertion is happening too early, and we should actually do it only at points where we collect all of the relevant trait bounds for a type parameter.
I could use some help interpreting the changes to the stderr output. It looks like reordering the predicates changed some diagnostics that don't obviously have anything to do with `Sized` bounds. Possibly some error reporting code is making assumptions about ordering of predicates? The diagnostics for src/test/ui/derives/derives-span-Hash-*.rs seem to have improved, no longer pointing at the type parameter identifier, but src/test/ui/type-alias-impl-trait/generic_duplicate_param_use9.rs became less verbose for some reason.
I also ran into an instance of #84970 while working on this, but I kind of expected that could happen, because I'm reordering predicates. I can open a separate issue on that if it would be helpful.
``@estebank`` this seems likely to conflict (slightly?) with your work on #85947; how would you like to resolve that?
Allow static linking LLVM with ThinLTO
There's no reason not to allow this if the user wants it. It works, at least in a local build on linux host.
For our use case, we're happy to spend more time building the compiler if it creates a speedup every time we run it, and we've observed speedups like this with clang.
Fix ctrl-c causing reads of stdin to return empty on Windows.
Pressing ctrl+c (or ctrl+break) on Windows caused a blocking read of stdin to unblock and return empty, unlike other platforms which continue to block.
On ctrl-c, `ReadConsoleW` will return success, but also set `LastError` to `ERROR_OPERATION_ABORTED`.
This change detects this case, and re-tries the call to `ReadConsoleW`.
Fixes#89177. See issue for further details.
Tested on Windows 7 and Windows 10 with both MSVC and GNU toolchains
Fix incorrect Box::pin suggestion
The suggestion checked if `Pin<Box<T>>` could be coeerced to the expected
type, but did not check predicates created by the coercion. We now
look for predicates that definitely cannot be satisfied before giving
the suggestion.
The suggestion is still marked MaybeIncorrect because we allow predicates that
are still ambiguous and can't be proven.
Fixes#72117.
Add check that live_region is live in sanitize_promoted
This pull request fixes#88434 by adding a check in `sanitize_promoted` to ensure that only regions which are actually live are added to the `liveness_constraints` of the `BorrowCheckContext`.
To implement this change, I needed to add a method to `LivenessValues` which gets the elements contained by a region:
/// Returns an iterator of all the elements contained by the region `r`
crate fn get_elements(&self, row: N) -> impl Iterator<Item = Location> + '_
Then, inside `sanitize_promoted`, we check whether the iterator returned by this method is non-empty to ensure that the region is actually live at at least one location before adding that region to the `liveness_constraints` of the `BorrowCheckContext`.
This is my first pull request to the Rust repo, so any feedback on how I can improve this pull request or if there is a better way to fix this issue would be very appreciated.
Add `const_eval_select` intrinsic
Adds an intrinsic that calls a given function when evaluated at compiler time, but generates a call to another function when called at runtime.
See https://github.com/rust-lang/const-eval/issues/7 for previous discussion.
r? `@oli-obk.`
The suggestion checked if Pin<Box<T>> could be coeerced to the expected
type, but did not check predicates created by the coercion. We now
look for predicates that definitely cannot be satisfied before giving
the suggestion.
The suggestion is marked MaybeIncorrect because we allow predicates that
are still ambiguous and can't be proven.
Assemble the compiler when running `x.py build`
Previously, there was no way to actually get binaries in
`build/$TARGET/stage1/bin` without building the standard library. This
makes it possible to build just the compiler. This can be useful when
the standard library isn't actually necessary for trying out your tests
(e.g. a bug that can be reproduced with only a `no_core` crate).
Closes https://github.com/rust-lang/rust/issues/73519.
Use shallow clones for submodules
This reduces the amount of git history downloaded for submodules from ~67M to ~11M. For comparison, a shallow clone of rust-lang/rust is 103M and a deep clone is 740M, so this almost halves the amount of history necessary if you made a shallow clone to start, and it's a significant reduction even if not.
Closes https://github.com/rust-lang/rust/issues/63978. r? `@Mark-Simulacrum`
