On usize=u64 platforms, the 4th iteration would overflow the `mod_gate`
back to 0. Similarly for usize=u32 platforms, the 3rd iteration would
overflow much the same way.
I tested various approaches to resolving this, including approaches with
`saturating_mul` and `widening_mul` to a double usize. Turns out LLVM
likes `mul_with_overflow` the best. In fact now, that LLVM can see the
iteration count is limited, it will happily unroll the loop into a nice
linear sequence.
You will also notice that the code around the loop got simplified
somewhat. Now that LLVM is handling the loop nicely, there isn’t any
more reasons to manually unroll the first iteration out of the loop
(though looking at the code today I’m not sure all that complexity was
necessary in the first place).
Fixes#103361
Add default trait implementations for "c-unwind" ABI function pointers
Following up on #92964, only add default trait implementations for the `c-unwind` family of function pointers. The previous attempt in #92964 added trait implementations for many more ABIs and ran into concerns regarding the increase in size of the libcore rlib.
An attempt to abstract away function pointer types behind a unified trait to reduce the duplication of trait impls is being discussed in #99531 but this change looks to be blocked on a lang MCP.
Following `@RalfJung's` suggestion in https://github.com/rust-lang/rust/pull/99531#issuecomment-1233440142, this commit is another cut at #92964 but it _only_ adds the impls for `extern "C-unwind" fn` and `unsafe extern "C-unwind" fn`.
I am interested in landing this patch to unblock the stabilization of the `c_unwind` feature.
RFC: https://github.com/rust-lang/rfcs/pull/2945
Tracking Issue: https://github.com/rust-lang/rust/issues/74990
Adjust `transmute{,_copy}` to be clearer about which of `T` and `U` is input vs output
This is essentially a documentation-only change (although it does touch code in an irrelevant way).
Following up on #92964, only add default trait implementations for the
`c-unwind` family of function pointers. The previous attempt in #92964
added trait implementations for many more ABIs and ran into concerns
regarding the increase in size of the libcore rlib.
An attempt to abstract away function pointer types behind a unified
trait to reduce the duplication of trait impls is being discussed in #99531
but this change looks to be blocked on a lang MCP.
Following @RalfJung's suggestion in
https://github.com/rust-lang/rust/pull/99531#issuecomment-1233440142,
this commit is another cut at #92964 but it _only_ adds the impls for
`extern "C-unwind" fn` and `unsafe extern "C-unwind" fn`.
I am interested in landing this patch to unblock the stabilization of
the `c_unwind` feature.
RFC: https://github.com/rust-lang/rfcs/pull/2945
Tracking Issue: https://github.com/rust-lang/rust/issues/74990
Make transpose const and inline
r? `@scottmcm`
- These should have been const from the beginning since we're never going to do more than a transmute.
- Inline these always because that's what every other method in MaybeUninit which simply casts does. :) Ok, but a stronger justification is that because we're taking in arrays by `self`, not inlining would defeat the whole purpose of using `MaybeUninit` due to the copying.
Optimize `slice_iter.copied().next_chunk()`
```
OLD:
test iter::bench_copied_array_chunks ... bench: 371 ns/iter (+/- 7)
NEW:
test iter::bench_copied_array_chunks ... bench: 31 ns/iter (+/- 0)
```
The default `next_chunk` implementation suffers from having to assemble the array byte by byte via `next()`, checking the `Option<&T>` and then dereferencing `&T`. The specialization copies the chunk directly from the slice.
More slice::partition_point examples
After seeing the discussion of `binary_search` vs `partition_point` in #101999, I thought some more example code could be helpful.
doc: rewrite doc for uint::{carrying_add,borrowing_sub}
Reword the documentation for bigint helper methods `uint::{carrying_add,borrowing_sub}` (#85532).
The examples were also rewritten to demonstrate how the methods can be used in bignum arithmetic. No loops are used in the examples, but the variable names were chosen to include indices so that it is clear how this can be used in a loop if required.
Also, previously `carrying_add` had an example to say that if the input carry is false, the method is equivalent to `overflowing_add`. While the note was kept, the example was removed and an extra note was added to make sure this equivalence is not assumed for signed integers as well.
Remove the redundant `Some(try_opt!(..))` in `checked_pow`
The final return value doesn't need to be tried at all -- we can just
return the checked option directly. The optimizer can probably figure
this out anyway, but there's no need to make it work here.
The final return value doesn't need to be tried at all -- we can just
return the checked option directly. The optimizer can probably figure
this out anyway, but there's no need to make it work here.
Clarify the possible return values of `len_utf16`
`char::len_utf16` always return 1 or 2. Clarify this in the docs, in the same way as `char::len_utf8`.
Add documentation about the memory layout of `UnsafeCell<T>`
The documentation for `UnsafeCell<T>` currently does not make any promises about its memory layout. This PR adds this documentation, namely that the memory layout of `UnsafeCell<T>` is the same as the memory layout of its inner `T`.
# Use case
Without this layout promise, the following cast would not be legally possible:
```rust
fn example<T>(ptr: *mut T) -> *const UnsafeCell<T> {
ptr as *const UnsafeCell<T>
}
```
A use case where this can come up involves FFI. If Rust receives a pointer over a FFI boundary which provides shared read-write access (with some form of custom synchronization), and this pointer is managed by some Rust struct with lifetime `'a`, then it would greatly simplify its (internal) API and safety contract if a `&'a UnsafeCell<T>` can be created from a raw FFI pointer `*mut T`. A lot of safety checks can be done when receiving the pointer for the first time through FFI (non-nullness, alignment, initialize uninit bytes, etc.) and these properties can then be encoded into the `&UnsafeCell<T>` type. Without this documentation guarantee, this is not legal today outside of the standard library.
# Caveats
Casting in the opposite direction is still not valid, even with this documentation change:
```rust
fn example2<T>(ptr: &UnsafeCell<T>) -> &mut T {
let t = ptr as *const UnsafeCell<T> as *mut T;
unsafe { &mut *t }
}
```
This is because the only legal way to obtain a mutable pointer to the contents of the shared reference is through [`UnsafeCell::get`](https://doc.rust-lang.org/std/cell/struct.UnsafeCell.html#method.get) and [`UnsafeCell::raw_get`](https://doc.rust-lang.org/std/cell/struct.UnsafeCell.html#method.raw_get). Although there might be a desire to also make this legal at some point in the future, that part is outside the scope of this PR. Also see this relevant [Zulip thread](https://rust-lang.zulipchat.com/#narrow/stream/136281-t-lang.2Fwg-unsafe-code-guidelines/topic/transmuting.20.26.20-.3E.20.26mut).
# Alternatives
Instead of adding a new documentation promise, it's also possible to add a new method to `UnsafeCell<T>` with signature `pub fn from_ptr_bikeshed(ptr: *mut T) -> *const UnsafeCell<T>` which indirectly only allows one-way casting to `*const UnsafeCell<T>`.
Fix `Duration::{try_,}from_secs_f{32,64}(-0.0)`
Make `Duration::{try_,}from_secs_f{32,64}(-0.0)` return `Duration::ZERO` (as they did before #90247) instead of erroring/panicking.
I'll update this PR to remove the `#![feature(duration_checked_float)]` if #102271 is merged before this PR.
Tracking issue for `try_from_secs_f{32,64}`: #83400
nicer errors from assert_unsafe_precondition
This makes the errors shown by cargo-careful nicer, and since `panic_no_unwind` is `nounwind noreturn` it hopefully doesn't have bad codegen impact. Thanks to `@bjorn3` for the hint!
Would be nice if we could somehow supply our own (static) message to print, currently it always prints `panic in a function that cannot unwind`. But still, this is better than before.
Warn about safety of `fetch_update`
Specifically as it relates to the ABA problem.
`fetch_update` is a useful function, and one that isn't provided by, say, C++. However, this does not mean the function is magic. It is implemented in terms of `compare_exchange_weak`, and in particular, suffers from the ABA problem. See the following code, which is a naive implementation of `pop` in a lock-free queue:
```rust
fn pop(&self) -> Option<i32> {
self.front.fetch_update(Ordering::Relaxed, Ordering::Acquire, |front| {
if front == ptr::null_mut() {
None
}
else {
Some(unsafe { (*front).next })
}
}.ok()
}
```
This code is unsound if called from multiple threads because of the ABA problem. Specifically, suppose nodes are allocated with `Box`. Suppose the following sequence happens:
```
Initial: Queue is X -> Y.
Thread A: Starts popping, is pre-empted.
Thread B: Pops successfully, twice, leaving the queue empty.
Thread C: Pushes, and `Box` returns X (very common for allocators)
Thread A: Wakes up, sees the head is still X, and stores Y as the new head.
```
But `Y` is deallocated. This is undefined behaviour.
Adding a note about this problem to `fetch_update` should hopefully prevent users from being misled, and also, a link to this common problem is, in my opinion, an improvement to our docs on atomics.
slice: #[inline] a couple iterator methods.
The one I care about and actually saw in the wild not getting inlined is
clone(). We ended up doing a whole function call for something that just
copies two pointers.
I ended up marking as_slice / as_ref as well because make_slice is
inline(always) itself, and is also the kind of think that can kill
performance in hot loops if you expect it to get inlined. But happy to
undo those.
