Generalize `fn allocator` for Rc/Arc.
Split out from #119761
- For `Rc`/`Arc`, the existing associated `fn`s are changed to allow unsized pointees.
- For `Weak`s, new methods are added.
`````@rustbot````` label +A-allocators
Mark format! with must_use hint
Uses unstable feature https://github.com/rust-lang/rust/issues/94745
Part of #126475
First contribution to rust, please let me know if the blessing of tests is correct
Thanks `@bjorn3` for the help
Make casts of pointers to trait objects stricter
This is an attempt to `fix` https://github.com/rust-lang/rust/issues/120222 and https://github.com/rust-lang/rust/issues/120217.
This is done by adding restrictions on casting pointers to trait objects.
Before this PR the rules were as follows:
> When casting `*const X<dyn A>` -> `*const Y<dyn B>`, principal traits in `A` and `B` must refer to the same trait definition (or no trait).
With this PR the rules are changed to
> When casting `*const X<dyn Src>` -> `*const Y<dyn Dst>`
> - if `Dst` has a principal trait `DstP`,
> - `Src` must have a principal trait `SrcP`
> - `dyn SrcP` and `dyn DstP` must be the same type (modulo the trait object lifetime, `dyn T+'a` -> `dyn T+'b` is allowed)
> - Auto traits in `Dst` must be a subset of auto traits in `Src`
> - Not adhering to this is currently a FCW (warn-by-default + `FutureReleaseErrorReportInDeps`), instead of an error
> - if `Src` has a principal trait `Dst` must as well
> - this restriction will be removed in a follow up PR
This ensures that
1. Principal trait's generic arguments match (no `*const dyn Tr<A>` -> `*const dyn Tr<B>` casts, which are a problem for [#120222](https://github.com/rust-lang/rust/issues/120222))
2. Principal trait's lifetime arguments match (no `*const dyn Tr<'a>` -> `*const dyn Tr<'b>` casts, which are a problem for [#120217](https://github.com/rust-lang/rust/issues/120217))
3. No auto traits can be _added_ (this is a problem for arbitrary self types, see [this comment](https://github.com/rust-lang/rust/pull/120248#discussion_r1463835350))
Some notes:
- We only care about the metadata/last field, so you can still cast `*const dyn T` to `*const WithHeader<dyn T>`, etc
- The lifetime of the trait object itself (`dyn A + 'lt`) is not checked, so you can still cast `*mut FnOnce() + '_` to `*mut FnOnce() + 'static`, etc
- This feels fishy, but I couldn't come up with a reason it must be checked
The diagnostics are currently not great, to say the least, but as far as I can tell this correctly fixes the issues.
cc `@oli-obk` `@compiler-errors` `@lcnr`
Run alloc sync tests
I was browsing the code and this struck me as weird. We're not running some doc tests because, the comment says, Windows builders deadlock. That should absolutely not happen, at least with our current implementation. And if it does happen I'd like to know.
Just to be sure though I'll do some try builds.
try-job: x86_64-msvc
try-job: i686-msvc
try-job: i686-mingw
try-job: x86_64-mingw
Rollup of 8 pull requests
Successful merges:
- #127179 (Print `TypeId` as hex for debugging)
- #127189 (LinkedList's Cursor: method to get a ref to the cursor's list)
- #127236 (doc: update config file path in platform-support/wasm32-wasip1-threads.md)
- #127297 (Improve std::Path's Hash quality by avoiding prefix collisions)
- #127308 (Attribute cleanups)
- #127354 (Describe Sized requirements for mem::offset_of)
- #127409 (Emit a wrap expr span_bug only if context is not tainted)
- #127447 (once_lock: make test not take as long in Miri)
r? `@ghost`
`@rustbot` modify labels: rollup
LinkedList's Cursor: method to get a ref to the cursor's list
We're already providing `.back()` & `.front()`, for which we hold onto a reference to the parent list, so why not share it? Useful for when you got `LinkedList` -> `CursorMut` -> `Cursor` and cannot take another ref to the list, even though you should be able to. This seems to be completely safe & sound.
The name is, of course, bikesheddable.
Don't check the capacity every time (and also for `Extend` for tuples, as this is how `unzip()` is implemented).
I did this with an unsafe method on `Extend` that doesn't check for growth (`extend_one_unchecked()`). I've marked it as perma-unstable currently, although we may want to expose it in the future so collections outside of std can benefit from it. Then specialize `Extend for (A, B)` for `TrustedLen` to call it.
It may seem that an alternative way of implementing this is to have a semi-public trait (`#[doc(hidden)]` public, so collections outside of core can implement it) for `extend()` inside tuples, and specialize it from collections. However, it is impossible due to limitations of `min_specialization`.
A concern that may arise with the current approach is that implementing `extend_one_unchecked()` correctly must also incur implementing `extend_reserve()`, otherwise you can have UB. This is a somewhat non-local safety invariant. However, I believe this is fine, since to have actual UB you must have unsafe code inside your `extend_one_unchecked()` that makes incorrect assumption, *and* not implement `extend_reserve()`. I've also documented this requirement.
The rules for casting `*mut X<dyn A>` -> `*mut Y<dyn B>` are as follows:
- If `B` has a principal
- `A` must have exactly the same principal (including generics)
- Auto traits of `B` must be a subset of autotraits in `A`
Note that `X<_>` and `Y<_>` can be identity, or arbitrary structs with last field being the dyn type.
The lifetime of the trait object itself (`dyn ... + 'a`) is not checked.
This prevents a few soundness issues with `#![feature(arbitrary_self_types)]` and trait upcasting.
Namely, these checks make sure that vtable is always valid for the pointee.
In 126578 we ended up with more binary size increases than expected.
This change attempts to avoid inlining large things into small things, to avoid that kind of increase, in cases when top-down inlining will still be able to do that inlining later.
Cleanup bootstrap check-cfg
This PR cleanup many custom `check-cfg` in bootstrap that have been accumulated over the years.
As well as updating some outdated comments.
Remove `__rust_force_expr`.
This was added (with a different name) to improve an error message. It is no longer needed -- removing it changes the error message, but overall I think the new message is no worse:
- the mention of `#` in the first line is a little worse,
- but the extra context makes it very clear what the problem is, perhaps even clearer than the old message,
- and the removal of the note about the `expr` fragment (an internal detail of `__rust_force_expr`) is an improvement.
Overall I think the error is quite clear and still far better than the old message that prompted #61933, which didn't even mention patterns.
The motivation for this is #124141, which will cause pasted metavariables to be tokenized and reparsed instead of the AST node being cached. This change in behaviour occasionally has a non-zero perf cost, and `__rust_force_expr` causes the tokenize/reparse step to occur twice. Removing `__rust_force_expr` greatly reduces the extra overhead for the `deep-vector` benchmark.
r? ```@oli-obk```
Detect unused structs which derived Default
<!--
If this PR is related to an unstable feature or an otherwise tracked effort,
please link to the relevant tracking issue here. If you don't know of a related
tracking issue or there are none, feel free to ignore this.
This PR will get automatically assigned to a reviewer. In case you would like
a specific user to review your work, you can assign it to them by using
r? <reviewer name>
-->
Fixes#98871
This was added (with a different name) to improve an error message. It
is no longer needed -- removing it changes the error message, but overall
I think the new message is no worse:
- the mention of `#` in the first line is a little worse,
- but the extra context makes it very clear what the problem is, perhaps
even clearer than the old message,
- and the removal of the note about the `expr` fragment (an internal
detail of `__rust_force_expr`) is an improvement.
Overall I think the error is quite clear and still far better than the
old message that prompted #61933, which didn't even mention patterns.
The motivation for this is #124141, which will cause pasted
metavariables to be tokenized and reparsed instead of the AST node being
cached. This change in behaviour occasionally has a non-zero perf cost,
and `__rust_force_expr` causes the tokenize/reparse step to occur twice.
Removing `__rust_force_expr` greatly reduces the extra overhead for the
`deep-vector` benchmark.
This is possible now that inline const blocks are stable; the idea was
even mentioned as an alternative when `uninit_array()` was added:
<https://github.com/rust-lang/rust/pull/65580#issuecomment-544200681>
> if it’s stabilized soon enough maybe it’s not worth having a
> standard library method that will be replaceable with
> `let buffer = [MaybeUninit::<T>::uninit(); $N];`
Const array repetition and inline const blocks are now stable (in the
next release), so that circumstance has come to pass, and we no longer
have reason to want `uninit_array()` other than convenience. Therefore,
let’s evaluate the inconvenience by not using `uninit_array()` in
the standard library, before potentially deleting it entirely.
Rollup of 3 pull requests
Successful merges:
- #126140 (Rename `std::fs::try_exists` to `std::fs::exists` and stabilize fs_try_exists)
- #126318 (Add a `x perf` command for integrating bootstrap with `rustc-perf`)
- #126552 (Remove use of const traits (and `feature(effects)`) from stdlib)
r? `@ghost`
`@rustbot` modify labels: rollup
Generalize `{Rc,Arc}::make_mut()` to unsized types.
* `{Rc,Arc}::make_mut()` now accept any type implementing the new unstable trait `core::clone::CloneToUninit`.
* `CloneToUninit` is implemented for `T: Clone` and for `[T] where T: Clone`.
* `CloneToUninit` is a generalization of the existing internal trait `alloc::alloc::WriteCloneIntoRaw`.
* New feature gate: `clone_to_uninit`
This allows performing `make_mut()` on `Rc<[T]>` and `Arc<[T]>`, which was not previously possible.
---
Previous PR description, now obsolete:
> Add `{Rc, Arc}::make_mut_slice()`
>
> These functions behave identically to `make_mut()`, but operate on `Arc<[T]>` instead of `Arc<T>`.
>
> This allows performing the operation on slices, which was not previously possible because `make_mut()` requires `T: Clone` (and slices, being `!Sized`, do not and currently cannot implement `Clone`).
>
> Feature gate: `make_mut_slice`
try-job: test-various
This requires introducing a new internal type `RcUninit` (and
`ArcUninit`), which can own an `RcBox<T>` without requiring it to be
initialized, sized, or a slice. This is similar to `UniqueRc`, but
`UniqueRc` doesn't support the allocator parameter, and there is no
`UniqueArc`.
Replace sort implementations
This PR replaces the sort implementations with tailor-made ones that strike a balance of run-time, compile-time and binary-size, yielding run-time and compile-time improvements. Regressing binary-size for `slice::sort` while improving it for `slice::sort_unstable`. All while upholding the existing soft and hard safety guarantees, and even extending the soft guarantees, detecting strict weak ordering violations with a high chance and reporting it to users via a panic.
* `slice::sort` -> driftsort [design document](https://github.com/Voultapher/sort-research-rs/blob/main/writeup/driftsort_introduction/text.md), includes detailed benchmarks and analysis.
* `slice::sort_unstable` -> ipnsort [design document](https://github.com/Voultapher/sort-research-rs/blob/main/writeup/ipnsort_introduction/text.md), includes detailed benchmarks and analysis.
#### Why should we change the sort implementations?
In the [2023 Rust survey](https://blog.rust-lang.org/2024/02/19/2023-Rust-Annual-Survey-2023-results.html#challenges), one of the questions was: "In your opinion, how should work on the following aspects of Rust be prioritized?". The second place was "Runtime performance" and the third one "Compile Times". This PR aims to improve both.
#### Why is this one big PR and not multiple?
* The current documentation gives performance recommendations for `slice::sort` and `slice::sort_unstable`. If for example only one of them were to be changed, this advice would be misleading for some Rust versions. By replacing them atomically, the advice remains largely unchanged, and users don't have to change their code.
* driftsort and ipnsort share a substantial part of their implementations.
* The implementation of `select_nth_unstable` uses internals of `slice::sort_unstable`, which makes it impractical to split changes.
---
This PR is a collaboration with `@orlp.`
Most modules have such a blank line, but some don't. Inserting the blank
line makes it clearer that the `//!` comments are describing the entire
module, rather than the `use` declaration(s) that immediately follows.
This makes their intent and expected location clearer. We see some
examples where these comments were not clearly separate from `use`
declarations, which made it hard to understand what the comment is
describing.
The changes made only a limited improvement for the current small
miri coverage and in general test coverage of the sort implementations.
But they exploded test times from ~13s to ~240s, which is not deemed
worth it.
`UniqueRc`: support allocators and `T: ?Sized`.
Added the following (all unstable):
* Defaulted type pararameter `A: Allocator`.
* `UniqueRc::new_in()`.
* `T: ?Sized` where possible.
* `impl CoerceUnsized for UniqueRc`.
These changes are motivated by supporting the implementation of unsized `Rc::make_mut()` (PR #116113), but are also intended to be obvious generalizations of `UniqueRc` to support the things `Rc` does.
r? ``````@the8472``````
Added the following (all unstable):
* Defaulted type pararameter `A: Allocator`.
* `UniqueRc::new_in()`.
* `T: ?Sized` where possible.
* `impl CoerceUnsized for UniqueRc`.
* Drive-by doc polish: links and periods at the end of sentences.
These changes are motivated by supporting the implementation of unsized
`Rc::make_mut()` (PR #116113), but are also intended to be obvious
generalizations of `UniqueRc` to support the things `Rc` does.
Add `size_of` and `size_of_val` and `align_of` and `align_of_val` to the prelude
(Note: need to update the PR to add `align_of` and `align_of_val`, and remove the second commit with the myriad changes to appease the lint.)
Many, many projects use `size_of` to get the size of a type. However,
it's also often equally easy to hardcode a size (e.g. `8` instead of
`size_of::<u64>()`). Minimizing friction in the use of `size_of` helps
ensure that people use it and make code more self-documenting.
The name `size_of` is unambiguous: the name alone, without any prefix or
path, is self-explanatory and unmistakeable for any other functionality.
Adding it to the prelude cannot produce any name conflicts, as any local
definition will silently shadow the one from the prelude. Thus, we don't
need to wait for a new edition prelude to add it.
