Replace some once(x).chain(once(y)) with [x, y] IntoIter
Now that we have by-value array iterators that are [already used](25c8c53dd9/compiler/rustc_hir/src/def.rs (L305-L307))...
For example,
```diff
- once(self.type_ns).chain(once(self.value_ns)).chain(once(self.macro_ns)).filter_map(|it| it)
+ IntoIter::new([self.type_ns, self.value_ns, self.macro_ns]).filter_map(|it| it)
```
BTreeMap: admit the existence of leaf edges in comments
The btree code is ambiguous about leaf edges (i.e., edges within leaf nodes). Iteration relies on them heavily, but some of the comments suggest there are no leaf edges (extracted from #77025)
r? @Mark-Simulacrum
Use more intra-doc-links in `core::fmt`
This is a follow-up to #75819, which encountered some broken links due to #75176, so this PR contains the links that are blocked on #75176.
r? @jyn514
Hint the maximum length permitted by invariant of slices
One of the safety invariants of references, and in particular of references to slices, is that they may not cover more than `isize::MAX` bytes. The unsafe `from_raw_parts` constructors of slices explicitly requires the caller to guarantee this fact. Violating it would also be UB with regards to the semantics of generated llvm code.
This effectively bounds the length of a (non-ZST) slice from above by a compile time constant. But when the length is loaded from a function argument it appears llvm is not aware of this requirement. The additional value range assertions allow some further elision of code branches, including overflow checks, especially in the presence of artithmetic on the indices.
This may have a performance impact, adding more code to a common method but allowing more optimization. I'm not quite sure, is the Rust side of const-prop strong enough to elide the irrelevant match branches?
Fixes: #67186
Uses assume to check the length against a constant upper bound. The
inlined result then informs the optimizer of the sound value range.
This was tried with unreachable_unchecked before which introduces a
branch. This has the advantage of not being executed in sound code but
complicates basic blocks. It resulted in ~2% increased compile time in
some worst cases.
Add a codegen test for the assumption, testing the issue from #67186
Rollup of 8 pull requests
Successful merges:
- #77072 (Minor `hash_map` doc adjustments + item attribute orderings)
- #77368 (Backport LLVM apfloat commit to rustc_apfloat)
- #77445 (BTreeMap: complete the compile-time test_variance test case)
- #77504 (Support vectors with fewer than 8 elements for simd_select_bitmask)
- #77513 (Change DocFragments from enum variant fields to structs with a nested enum)
- #77518 (Only use Fira Sans for the first `td` in item lists)
- #77521 (Move target feature whitelist from cg_llvm to cg_ssa)
- #77525 (Enable RenameReturnPlace MIR optimization on mir-opt-level >= 2)
Failed merges:
r? `@ghost`
BTreeMap: complete the compile-time test_variance test case
Some of the items added to the new `test_sync` belonged in the old `test_variance` as well. And fixed inconsistent paths to nearby modules.
r? @Mark-Simulacrum
Minor `hash_map` doc adjustments + item attribute orderings
This PR is really a couple visual changes glued together:
1. Some of the doc comments for items in `std::collections::hash_map` referenced the names of types without escaping their formatting (e.g. using "VacantEntry" instead of "`VacantEntry`") - the ones I could find were changed to the latter
2. The vast majority of pre-item attributes seem to place doc comments as the first attribute (instead of things like `#[feature(...)]`), so the few that had the other order were changed.
3. Also ordering related: the general trend seems to be that `#[feature]` attributes follow `#[inline]`, so I swapped the two lines in places where that ordering was reversed. This is primarily a change based on stylistic continuity and aesthetics - I'm not sure how important that actually is / should be.
I figured this would be pretty uncontroversial, but some of these might have been intentional for reasons I don't know about - if so, I'd be happy to remove the relevant changes. Of these, the final set of changes is probably the most unnecessary, so it also might be better to leave those out (in favor of reducing code churn).
Implement as_ne_bytes() for integers and floats
This is related to issue #64464.
I am pretty sure that these functions are actually const-ify-able, and technically as_bits() can also be implemented for floats, but I might need some comments on both.
Implement Make `handle_alloc_error` default to panic (for no_std + liballoc)
Related: https://github.com/rust-lang/rust/issues/66741
Guarded with `#![feature(default_alloc_error_handler)]` a default
`alloc_error_handler` is called, if a custom allocator is used and no
other custom `#[alloc_error_handler]` is defined.
Unbox mutexes and condvars on some platforms
Both mutexes and condition variables contained a Box containing the actual os-specific object. This was done because moving these objects may cause undefined behaviour on some platforms.
However, this is not needed on Windows[1], Wasm[2], cloudabi[2], and 'unsupported'[3], were the box was only needlessly making them less efficient.
This change gets rid of the box on those platforms.
On those platforms, `Condvar` can no longer verify it is only used with one `Mutex`, as mutexes no longer have a stable address. This was addressed and considered acceptable in #76932.
[1]\: https://docs.microsoft.com/en-us/windows/win32/api/synchapi/nf-synchapi-initializesrwlock
[2]\: These are just a single atomic integer together with futex wait/wake calls/instructions.
[3]\: The `unsupported` platform doesn't support multiple threads at all.
Use less divisions in display u128/i128
This PR is an absolute mess, and I need to test if it improves the speed of fmt::Display for u128/i128, but I think it's correct.
It hopefully is more efficient by cutting u128 into at most 2 u64s, and also chunks by 1e16 instead of just 1e4.
Also I specialized the implementations for uints to always be non-false because it bothered me that it was checked at all
Do not merge until I benchmark it and also clean up the god awful mess of spaghetti.
Based on prior work in #44583
cc: `@Dylan-DPC`
Due to work on `itoa` and suggestion in original issue:
r? `@dtolnay`
Allow Weak::as_ptr and friends for unsized T
Relaxes `impl<T> Weak<T>` to `impl<T: ?Sized> Weak<T>` for the methods `rc::Weak::as_ptr`, `into_raw`, and `from_raw`.
Follow-up to #73845, which did most of the impl work to make these functions work for `T: ?Sized`.
We still have to adjust the implementation of `Weak::from_raw` here, however, because I missed a use of `ptr.is_null()` previously. This check was necessary when `into`/`from_raw` were first implemented, as `into_raw` returned `ptr::null()` for dangling weak. However, we now just (wrapping) offset dangling weaks' pointers the same as nondangling weak, so the null check is no longer necessary (or even hit). (I can submit just 17a928f as a separate PR if desired.)
As a nice side effect, moves the `fn is_dangling` definition closer to `Weak::new`, which creates the dangling weak.
This technically stabilizes that "something like `align_of_val_raw`" is possible to do. However, I believe the part of the functionality required by these methods here -- specifically, getting the alignment of a pointee from a pointer where it may be dangling iff the pointee is `Sized` -- is uncontroversial enough to stabilize these methods without a way to implement them on stable Rust.
r? `@RalfJung,` who reviewed #73845.
ATTN: This changes (relaxes) the (input) generic bounds on stable fn!
Remove --cfg dox from rustdoc.rs
This was added in https://github.com/rust-lang/rust/pull/53076 because
several dependencies were using `cfg(dox)` instead of `cfg(rustdoc)` (now `cfg(doc)`).
I ran `rg 'cfg\(dox\)'` on the source tree with no matches, so I think
this is now safe to remove.
r? `@Mark-Simulacrum`
cc `@QuietMisdreavus` :)
Only use LOCAL_{STDOUT,STDERR} when set_{print/panic} is used.
The thread local `LOCAL_STDOUT` and `LOCAL_STDERR` are only used by the `test` crate to capture output from tests when running them in the same process in differen threads. However, every program will check these variables on every print, even outside of testing.
This involves allocating a thread local key, and registering a thread local destructor. This can be somewhat expensive.
This change keeps a global flag (`LOCAL_STREAMS`) which will be set to `true` when either of these local streams is used. (So, effectively only in test and benchmark runs.) When this flag is off, these thread locals are not even looked at and therefore will not be initialized on the first output on every thread, which also means no thread local destructors will be registered.
---
Together with https://github.com/rust-lang/rust/pull/77154, this should make output a little bit more efficient.
Fix Debug implementations of some of the HashMap and BTreeMap iterator types
HashMap's `ValuesMut`, BTreeMaps `ValuesMut`, IntoValues and `IntoKeys` structs were printing both keys and values on their Debug implementations. But they are iterators over either keys or values. Irrelevant values should not be visible. With this PR, they only show relevant fields.
This fixes#75297.
[Here's an example code.](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=0c79356ed860e347a0c1a205616f93b7) This prints this on nightly:
```
ValuesMut { inner: IterMut { range: [(1, "hello"), (2, "goodbye")], length: 2 } }
IntoKeys { inner: [(1, "hello"), (2, "goodbye")] }
IntoValues { inner: [(1, "hello"), (2, "goodbye")] }
[(2, "goodbye"), (1, "hello")]
```
After the patch this example prints these instead:
```
["hello", "goodbye"]
["hello", "goodbye"]
[1, 2]
["hello", "goodbye"]
```
I didn't add test cases for them, since I couldn't see any tests for Debug implementations anywhere. But please let me know if I should add it to a specific place.
r? @dtolnay
Use posix_spawn on musl targets
The posix_spawn had been available in a form suitable for use in a
Command implementation since musl 0.9.12. Use it in a preference to a
fork when possible, to benefit from CLONE_VM|CLONE_VFORK used there.
Previously `Command::spawn` would fall back to the non-posix_spawn based
implementation if the `PATH` environment variable was possibly changed.
On systems with a modern (g)libc `posix_spawn()` can be significantly
faster. If program is a path itself the `PATH` environment variable is
not used for the lookup and it should be safe to use the
`posix_spawnp()` method. [1]
We found this, because we have a cli application that effectively runs a
lot of subprocesses. It would sometimes noticeably hang while printing
output. Profiling showed that the process was spending the majority of
time in the kernel's `copy_page_range` function while spawning
subprocesses. During this time the process is completely blocked from
running, explaining why users were reporting the cli app hanging.
Through this we discovered that `std::process::Command` has a fast and
slow path for process execution. The fast path is backed by
`posix_spawnp()` and the slow path by fork/exec syscalls being called
explicitly. Using fork for process creation is supposed to be fast, but
it slows down as your process uses more memory. It's not because the
kernel copies the actual memory from the parent, but it does need to
copy the references to it (see `copy_page_range` above!). We ended up
using the slow path, because the command spawn implementation in falls
back to the slow path if it suspects the PATH environment variable was
changed.
Here is a smallish program demonstrating the slowdown before this code
change:
```
use std::process::Command;
use std::time::Instant;
fn main() {
let mut args = std::env::args().skip(1);
if let Some(size) = args.next() {
// Allocate some memory
let _xs: Vec<_> = std::iter::repeat(0)
.take(size.parse().expect("valid number"))
.collect();
let mut command = Command::new("/bin/sh");
command
.arg("-c")
.arg("echo hello");
if args.next().is_some() {
println!("Overriding PATH");
command.env("PATH", std::env::var("PATH").expect("PATH env var"));
}
let now = Instant::now();
let child = command
.spawn()
.expect("failed to execute process");
println!("Spawn took: {:?}", now.elapsed());
let output = child.wait_with_output().expect("failed to wait on process");
println!("Output: {:?}", output);
} else {
eprintln!("Usage: prog [size]");
std::process::exit(1);
}
()
}
```
Running it and passing different amounts of elements to use to allocate
memory shows that the time taken for `spawn()` can differ quite
significantly. In latter case the `posix_spawnp()` implementation is 30x
faster:
```
$ cargo run --release 10000000
...
Spawn took: 324.275µs
hello
$ cargo run --release 10000000 changepath
...
Overriding PATH
Spawn took: 2.346809ms
hello
$ cargo run --release 100000000
...
Spawn took: 387.842µs
hello
$ cargo run --release 100000000 changepath
...
Overriding PATH
Spawn took: 13.434677ms
hello
```
[1]: 5f72f9800b/posix/execvpe.c (L81)
