make `Instant::{duration_since, elapsed, sub}` saturating and remove workarounds
This removes all mutex/atomic-based workarounds for non-monotonic clocks and makes the previously panicking methods saturating instead. Additionally `saturating_duration_since` becomes deprecated since `duration_since` now fills that role.
Effectively this moves the fixup from `Instant` construction to the comparisons.
This has some observable effects, especially on platforms without monotonic clocks:
* Incorrectly ordered Instant comparisons no longer panic in release mode. This could hide some programming errors, but since debug mode still panics tests can still catch them.
* `checked_duration_since` will now return `None` in more cases. Previously it only happened when one compared instants obtained in the wrong order or manually created ones. Now it also does on backslides.
* non-monotonic intervals will not be transitive, i.e. `b.duration_since(a) + c.duration_since(b) != c.duration_since(a)`
The upsides are reduced complexity and lower overhead of `Instant::now`.
## Motivation
Currently we must choose between two poisons. One is high worst-case latency and jitter of `Instant::now()` due to explicit synchronization; see #83093 for benchmarks, the worst-case overhead is > 100x. The other is sporadic panics on specific, rare combinations of CPU/hypervisor/operating system due to platform bugs.
Use-cases where low-overhead, fine-grained timestamps are needed - such as syscall tracing, performance profiles or sensor data acquisition (drone flight controllers were mentioned in a libs meeting) in multi-threaded programs - are negatively impacted by the synchronization.
The panics are user-visible (program crashes), hard to reproduce and can be triggered by any dependency that might be using Instants for any reason.
A solution that is fast _and_ doesn't panic is desirable.
----
closes#84448closes#86470
This removes all mutex/atomics based workarounds for non-monotonic clocks and makes the previously panicking methods saturating instead.
Effectively this moves the monotonization from `Instant` construction to the comparisons.
This has some observable effects, especially on platforms without monotonic clocks:
* Incorrectly ordered Instant comparisons no longer panic. This may hide some programming errors until someone actually looks at the resulting `Duration`
* `checked_duration_since` will now return `None` in more cases. Previously it only happened when one compared instants obtained in the wrong order or
manually created ones. Now it also does on backslides.
The upside is reduced complexity and lower overhead of `Instant::now`.
Fix hashing for windows paths containing a CurDir component
* the logic only checked for / but not for \
* verbatim paths shouldn't skip items at all since they don't get normalized
* the extra branches get optimized out on unix since is_sep_byte is a trivial comparison and is_verbatim is always-false
* tests lacked windows coverage for these cases
That lead to equal paths not having equal hashes and to unnecessary collisions.
Rollup of 10 pull requests
Successful merges:
- #90955 (Rename `FilenameTooLong` to `InvalidFilename` and also use it for Windows' `ERROR_INVALID_NAME`)
- #91607 (Make `span_extend_to_prev_str()` more robust)
- #92895 (Remove some unused functionality)
- #93635 (Add missing platform-specific information on current_dir and set_current_dir)
- #93660 (rustdoc-json: Add some tests for typealias item)
- #93782 (Split `pauth` target feature)
- #93868 (Fix incorrect register conflict detection in asm!)
- #93888 (Implement `AsFd` for `&T` and `&mut T`.)
- #93909 (Fix typo: explicitely -> explicitly)
- #93910 (fix mention of moved function in `rustc_hir` docs)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Implement `AsFd` for `&T` and `&mut T`.
Add implementations of `AsFd` for `&T` and `&mut T`, so that users can
write code like this:
```rust
pub fn fchown<F: AsFd>(fd: F, uid: Option<u32>, gid: Option<u32>) -> io::Result<()> {
```
with `fd: F` rather than `fd: &F`.
And similar for `AsHandle` and `AsSocket` on Windows.
Also, adjust the `fchown` example to pass the file by reference. The
code can work either way now, but passing by reference is more likely
to be what users will want to do.
This is an alternative to #93869, and is a simpler way to achieve the
same goals: users don't need to pass borrowed-`BorrowedFd` arguments,
and it prevents a pitfall in the case where users write `fd: F` instead
of `fd: &F`.
r? ```@joshtriplett```
Rename `FilenameTooLong` to `InvalidFilename` and also use it for Windows' `ERROR_INVALID_NAME`
Address https://github.com/rust-lang/rust/issues/90940#issuecomment-970157931
`ERROR_INVALID_NAME` (i.e. "The filename, directory name, or volume label syntax is incorrect") happens if we pass an invalid filename, directory name, or label syntax, so mapping as `InvalidInput` is reasonable to me.
Stabilise `is_aarch64_feature_detected!` under `simd_aarch64` feature
Initial implementation, looking for feedback on the approach here. https://github.com/rust-lang/rust/issues/86941
One point I noticed was that I haven't seen different "since" versions for the same feature - does this mean that other features can't be added to to the `simd_aarch64` feature once this is in stable? If so it might need a more specific name.
r? `@Amanieu`
Add implementations of `AsFd` for `&T` and `&mut T`, so that users can
write code like this:
```rust
pub fn fchown<F: AsFd>(fd: F, uid: Option<u32>, gid: Option<u32>) -> io::Result<()> {
```
with `fd: F` rather than `fd: &F`.
And similar for `AsHandle` and `AsSocket` on Windows.
Also, adjust the `fchown` example to pass the file by reference. The
code can work either way now, but passing by reference is more likely
to be what users will want to do.
This is an alternative to #93869, and is a simpler way to achieve the
same goals: users don't need to pass borrowed-`BorrowedFd` arguments,
and it prevents a pitfall in the case where users write `fd: F` instead
of `fd: &F`.
kmc-solid: Fix wait queue manipulation errors in the `Condvar` implementation
This PR fixes a number of bugs in the `Condvar` wait queue implementation used by the [`*-kmc-solid_*`](https://doc.rust-lang.org/nightly/rustc/platform-support/kmc-solid.html) Tier 3 targets. These bugs can occur when there are multiple threads waiting on the same `Condvar` and sometimes manifest as an `unwrap` failure.
Use `NtCreateFile` instead of `NtOpenFile` to open a file
Generally the internal `Nt*` functions should be avoided but when we do need to use one we should stick to the most commonly used for the job. To that end, this PR replaces `NtOpenFile` with `NtCreateFile`.
NOTE: The initial version of this comment hypothesised that this may help with some recent false positives from malware scanners. This hypothesis proved wrong. Sorry for the distraction.
Make io::Error use 64 bits on targets with 64 bit pointers.
I've wanted this for a long time, but didn't see a good way to do it without having extra allocation. When looking at it yesterday, it was more clear what to do for some reason.
This approach avoids any additional allocations, and reduces the size by half (8 bytes, down from 16). AFAICT it doesn't come additional runtime cost, and the compiler seems to do a better job with code using it.
Additionally, this `io::Error` has a niche (still), so `io::Result<()>` is *also* 64 bits (8 bytes, down from 16), and `io::Result<usize>` (used for lots of io trait functions) is 2x64 bits (16 bytes, down from 24 — this means on x86_64 it can use the nice rax/rdx 2-reg struct return). More generally, it shaves a whole 64 bit integer register off of the size of basically any `io::Result<()>`.
(For clarity: Improving `io::Result` (rather than io::Error) was most of the motivation for this)
On 32 bit (or other non-64bit) targets we still use something equivalent the old repr — I don't think think there's improving it, since one of the fields it stores is a `i32`, so we can't get below that, and it's already about as close as we can get to it.
---
### Isn't Pointer Tagging Dodgy?
The details of the layout, and why its implemented the way it is, are explained in the header comment of library/std/src/io/error/repr_bitpacked.rs. There's probably more details than there need to be, but I didn't trim it down that much, since there's a lot of stuff I did deliberately, that might have not seemed that way.
There's actually only one variant holding a pointer which gets tagged. This one is the (holder for the) user-provided error.
I believe the scheme used to tag it is not UB, and that it preserves pointer provenance (even though often pointer tagging does not) because the tagging operation is just `core::ptr::add`, and untagging is `core::ptr::sub`. The result of both operations lands inside the original allocation, so it would follow the safety contract of `core::ptr::{add,sub}`.
The other pointer this had to encode is not tagged — or rather, the tagged repr is equivalent to untagged (it's tagged with 0b00, and has >=4b alignment, so we can reuse the bottom bits). And the other variants we encode are just integers, which (which can be untagged using bitwise operations without worry — they're integers).
CC `@RalfJung` for the stuff in repr_bitpacked.rs, as my comments are informed by a lot of the UCG work, but it's possible I missed something or got it wrong (even if the implementation is okay, there are parts of the header comment that says things like "We can't do $x" which could be false).
---
### Why So Many Changes?
The repr change was mostly internal, but changed one widely used API: I had to switch how `io::Error::new_const` works.
This required switching `io::Error::new_const` to take the full message data (including the kind) as a `&'static`, rather than just the string. This would have been really tedious, but I made a macro that made it much simpler, but it was a wide change since `io::Error::new_const` is used everywhere.
This included changing files for a lot of targets I don't have easy access to (SGX? Haiku? Windows? Who has heard of these things), so I expect there to be spottiness in CI initially, unless luck is on my side.
Anyway this large only tangentially-related change is all in the first commit (although that commit also pulls the previous repr out into its own file), whereas the packing stuff is all in commit 2.
---
P.S. I haven't looked at all of this since writing it, and will do a pass over it again later, sorry for any obvious typos or w/e. I also definitely repeat myself in comments and such.
(It probably could use more tests too. I did some basic testing, and made it so we `debug_assert!` in cases the decode isn't what we encoded, but I don't know the degree which I can assume libstd's testing of IO would exercise this. That is: it wouldn't be surprising to me if libstds IO testing were minimal, especially around error cases, although I have no idea).
* the logic only checked for / but not for \
* verbatim paths shouldn't skip items at all since they don't get normalized
* the extra branches get optimized out on unix since is_sep_byte is a trivial comparison and is_verbatim is always-false
* tests lacked windows coverage for these cases
That lead to equal paths not having equal hashes and to unnecessary collisions.
kmc-solid: Fix off-by-one error in `SystemTime::now`
Fixes a miscalculation of `SystemTime` on the [`*-kmc-solid_*`](https://doc.rust-lang.org/nightly/rustc/platform-support/kmc-solid.html) Tier 3 targets.
Unlike the identically-named libc counterpart `tm::tm_mon`, `SOLID_RTC_TIME::tm_mon` contains a 1-based month number.
Link `try_exists` docs to `Path::exists`
Links to the existing `Path::exists` method from both `std::Path::try_exists` and `std::fs:try_exists`.
Tracking issue for `path_try_exists`: #83186
pub use std::simd::StdFloat;
Syncs portable-simd up to commit rust-lang/portable-simd@03f6fbb21e,
Diff: 533f0fc81a...03f6fbb21e
This sync requires a little bit more legwork because it also introduces a trait into `std::simd`, so that it is no longer simply a reexport of `core::simd`. Out of simple-minded consistency and to allow more options, I replicated the pattern for the way `core::simd` is integrated in the first place, however this is not necessary if it doesn't acquire any interdependencies inside `std`: it could be a simple crate reexport. I just don't know yet if that will happen or not.
To summarize other misc changes:
- Shifts no longer panic, now wrap on too-large shifts (like `Simd` integers usually do!)
- mask16x32 will now be many i16s, not many i32s... 🙃
- `#[must_use]` is spread around generously
- Adjusts division, float min/max, and `Mask::{from,to}_array` internally to be faster
- Adds the much-requested `Simd::cast::<U>` function (equivalent to `simd.to_array().map(|lane| lane as U)`)
Support configuring whether to capture backtraces at runtime
Tracking issue: https://github.com/rust-lang/rust/issues/93346
This adds a new API to the `std::panic` module which configures whether and how the default panic hook will emit a backtrace when a panic occurs.
After discussion with `@yaahc` on [Zulip](https://rust-lang.zulipchat.com/#narrow/stream/219381-t-libs/topic/backtrace.20lib.20vs.2E.20panic), this PR chooses to avoid adjusting or seeking to provide a similar API for the (currently unstable) std::backtrace API. It seems likely that the users of that API may wish to expose more specific settings rather than just a global one (e.g., emulating the `env_logger`, `tracing` per-module configuration) to avoid the cost of capture in hot code. The API added here could plausibly be copied and/or re-exported directly from std::backtrace relatively easily, but I don't think that's the right call as of now.
```rust
mod panic {
#[derive(Copy, Clone, Debug, PartialEq, Eq)]
#[non_exhaustive]
pub enum BacktraceStyle {
Short,
Full,
Off,
}
fn set_backtrace_style(BacktraceStyle);
fn get_backtrace_style() -> Option<BacktraceStyle>;
}
```
Several unresolved questions:
* Do we need to move to a thread-local or otherwise more customizable strategy for whether to capture backtraces? See [this comment](https://github.com/rust-lang/rust/pull/79085#issuecomment-727845826) for some potential use cases for this.
* Proposed answer: no, leave this for third-party hooks.
* Bikeshed on naming of all the options, as usual.
* Should BacktraceStyle be moved into `std::backtrace`?
* It's already somewhat annoying to import and/or re-type the `std::panic::` prefix necessary to use these APIs, probably adding a second module to the mix isn't worth it.
Note that PR #79085 proposed a much simpler API, but particularly in light of the desire to fully replace setting environment variables via `env::set_var` to control the backtrace API, a more complete API seems preferable. This PR likely subsumes that one.
Fix incorrect panic message in example
The panic message when calling the `connect()` should probably be a message about connection failure, not a message about binding address failure.
Document valid values of the char type
As discussed at #93392, the current documentation on what constitutes a valid char isn't very detailed and is partly on the MAX constant rather than the type itself.
This PR expands on that information, stating the actual numerical range, giving examples of what won't work, and also mentions how a `char` might be a valid USV but still not be a defined character (terminology checked against [Unicode 14.0, table 2-3](https://www.unicode.org/versions/Unicode14.0.0/ch02.pdf#M9.61673.TableTitle.Table.22.Types.of.Code.Points)).
Change Termination::report return type to ExitCode
Related to https://github.com/rust-lang/rust/issues/43301
The goal of this change is to minimize the forward compatibility risks in stabilizing Termination. By using the opaque type `ExitCode` instead of an `i32` we leave room for us to evolve the API over time to provide what cross-platform consistency we can / minimize footguns when working with exit codes, where as stabilizing on `i32` would limit what changes we could make in the future in how we represent and construct exit codes.
