Relax SeqCst ordering in standard library.
Every single SeqCst in the standard library is unnecessary. In all cases, Relaxed or Release+Acquire was sufficient.
As I [wrote](https://marabos.nl/atomics/memory-ordering.html#common-misconceptions) in my book on atomics:
> [..] when reading code, SeqCst basically tells the reader: "this operation depends on the total order of every single SeqCst operation in the program," which is an incredibly far-reaching claim. The same code would likely be easier to review and verify if it used weaker memory ordering instead, if possible. For example, Release effectively tells the reader: "this relates to an acquire operation on the same variable," which involves far fewer considerations when forming an understanding of the code.
>
> It is advisable to see SeqCst as a warning sign. Seeing it in the wild often means that either something complicated is going on, or simply that the author did not take the time to analyze their memory ordering related assumptions, both of which are reasons for extra scrutiny.
r? ````@Amanieu```` ````@joboet````
Use `UnsafeCell` for fast constant thread locals
This uses `UnsafeCell` instead of `static mut` for fast constant thread locals. This changes the type of the TLS shims to return `&UnsafeCell<T>` instead of `*mut T` which means they are always non-null so LLVM can optimize away the check for `Some` in `LocalKey::with` if `T` has no destructor.
LLVM is currently unable to do this optimization as we lose the fact that `__getit` always returns `Some` as it gets optimized to just returning the value of the TLS shim.
Bump windows-bindgen to 0.55.0
windows-bindgen is the crate used to generate std's Windows API bindings.
Not many changes for us, it's mostly just simplifying the generate code (e.g. no more `-> ()`). The one substantial change is some structs now use `i8` byte arrays instead of `u8`. However, this only impacts one test.
change std::process to drop supplementary groups based on CAP_SETGID
A trivial rebase of #95982
Should fix#39186 (from what I can tell)
Original description:
> Fixes#88716
>
> * Before this change, when a process was given a uid via `std::os::unix::process::CommandExt.uid`, there would be a `setgroups` call (when the process runs) to clear supplementary groups for the child **if the parent was root** (to remove potentially unwanted permissions).
> * After this change, supplementary groups are cleared if we have permission to do so, that is, if we have the CAP_SETGID capability.
>
> This new behavior was agreed upon in #88716 but there was a bit of uncertainty from `@Amanieu` here: [#88716 (comment)](https://github.com/rust-lang/rust/issues/88716#issuecomment-973366600)
>
> > I agree with this change, but is it really necessary to ignore an EPERM from setgroups? If you have permissions to change UID then you should also have permissions to change groups. I would feel more comfortable if we documented set_uid as requiring both UID and GID changing permissions.
>
> The way I've currently written it, we ignore an EPERM as that's what #88716 originally suggested. I'm not at all an expert in any of this so I'd appreciate feedback on whether that was the right way to go.
unix time module now return result
First try to fix#108277 without break anything.
if anyone who read this know tips to be able to check compilation for different target I could use some help. So far I installed many target with rustup but `./x check --all-targets` doesn't seem to use them.
TODO:
- [x] better error
- [ ] test, how ?
`@rustbot` label -S-waiting-on-author +S-waiting-on-review
std support for wasm32 panic=unwind
Tracking issue: #118168
This adds std support for `-Cpanic=unwind` on wasm, and with it slightly more fleshed out rustc support. Now, the stable default is still panic=abort without exception-handling, but if you `-Zbuild-std` with `RUSTFLAGS=-Cpanic=unwind`, you get wasm exception-handling try/catch blocks in the binary:
```rust
#[no_mangle]
pub fn foo_bar(x: bool) -> *mut u8 {
let s = Box::<str>::from("hello");
maybe_panic(x);
Box::into_raw(s).cast()
}
#[inline(never)]
#[no_mangle]
fn maybe_panic(x: bool) {
if x {
panic!("AAAAA");
}
}
```
```wat
;; snip...
(try $label$5
(do
(call $maybe_panic
(local.get $0)
)
(br $label$1)
)
(catch_all
(global.set $__stack_pointer
(local.get $1)
)
(call $__rust_dealloc
(local.get $2)
(i32.const 5)
(i32.const 1)
)
(rethrow $label$5)
)
)
;; snip...
```
std::threads: revisit stack address calculation on netbsd.
like older linux glibc versions, we need to get the guard size
and increasing the stack's bottom address accordingly.
Win10: Use `GetSystemTimePreciseAsFileTime` directly
On Windows 10 we can use `GetSystemTimePreciseAsFileTime` directly instead of lazy loading it (with a fallback).
Convert `Unix{Datagram,Stream}::{set_}passcred()` to per-OS traits
These methods are the pre-stabilized API for obtaining peer credentials from an `AF_UNIX` socket, part of the `unix_socket_ancillary_data` feature.
Their current behavior is to get/set one of the `SO_PASSCRED` (Linux), `LOCAL_CREDS_PERSISTENT` (FreeBSD), or `LOCAL_CREDS` (NetBSD) socket options. On other targets the `{set_}passcred()` methods do not exist.
There are two problems with this approach:
1. Having public methods only exist for certain targets isn't permitted in a stable `std` API.
2. These options have generally similar purposes, but they are non-POSIX and their details can differ in subtle and surprising ways (such as whether they continue to be set after the next call to `recvmsg()`).
Splitting into OS-specific extension traits is the preferred solution to both problems.
Dynamically size sigaltstk in std
On modern Linux with Intel AMX and 1KiB matrices,
Arm SVE with potentially 2KiB vectors,
and RISCV Vectors with up to 16KiB vectors,
we must handle dynamic signal stack sizes.
We can do so unconditionally by using getauxval,
but assuming it may return 0 as an answer,
thus falling back to the old constant if needed.
Fixes https://github.com/rust-lang/rust/issues/107795
Implement junction_point
Implements https://github.com/rust-lang/rust/issues/121709
We already had a private implementation that we use for tests so we could just make that public. Except it was very hacky as it was only ever intended for use in testing. I've made an improved version that at least handles path conversion correctly and has less need for things like the `Align8` hack. There's still room for further improvement though.
On modern Linux with Intel AMX and 1KiB matrices,
Arm SVE with potentially 2KiB vectors,
and RISCV Vectors with up to 16KiB vectors,
we must handle dynamic signal stack sizes.
We can do so unconditionally by using getauxval,
but assuming it may return 0 as an answer,
thus falling back to the old constant if needed.
Cleanup windows `abort_internal`
As the comments on the functions say, we define abort in both in panic_abort and in libstd. This PR makes the two implementation (mostly) the same.
Additionally it:
* uses `options(noreturn)` on the asm instead of using `core::intrinsics::unreachable`.
* removed unnecessary allow lints
* added `FAST_FAIL_FATAL_APP_EXIT` to our generated Windows API bindings instead of defining it manually (std only)
Use the OS thread name by default if `THREAD_INFO` has not been initialized
Currently if `THREAD_INFO` hasn't been initialized then the name will be set to `None`. This PR changes it to use the OS thread name by default. This mostly affects foreign threads at the moment but we could expand this to make more use of the OS thread name in the future.
Note: I've only implemented `Thread::get_name` for windows, linux and macos (and macos adjacent) targets. The rest just return `None`.
Delete architecture-specific memchr code in std::sys
Currently all architecture-specific memchr code is only used in `std::io`. Most of the actual `memchr` capacity exposed to the user through the slice API is instead implemented in `core::slice::memchr`.
Hence this commit deletes `memchr` from `std::sys[_common]` and replace calls to it by calls to `core::slice::memchr` functions. This deletes `(r)memchr` from the list of symbols linked to libc.
The interest of putting architecture specific code back in core is linked to the discussion to be had in #113654
Use volatile access instead of `#[used]` for `on_tls_callback`
The first commit adds a volatile load of `p_thread_callback` when registering a dtor so that the compiler knows if the callback is used or not. I don't believe the added volatile instruction is otherwise significant in the context. In my testing using the volatile load allowed the compiler to correctly reason about whether `on_tls_callback` is used or not, allowing it to be omitted entirely in some cases. Admittedly it usually is used due to `Thread` but that can be avoided (e.g. in DLLs or with custom entry points that avoid the offending APIs). Ideally this would be something the compiler could help a bit more with so we didn't have to use tricks like `#[used]` or volatile. But alas.
I also used this as an opportunity to clean up the `unused` lints which I don't think serve a purpose any more.
The second commit removes the volatile load of `_tls_used` with `#cfg[target_thread_local]` because `#[thread_local]` already implies it. And if it ever didn't then `#[thread_local]` would be broken when there aren't any dtors.
add platform-specific function to get the error number for HermitOS
Extending `std` to get the last error number for HermitOS.
HermitOS is a tier 3 platform and this PR changes only files, wich are related to the tier 3 platform.
Extending `std` to get the last error number for HermitOS.
HermitOS is a tier 3 platform and this PR changes only files,
wich are related to the tier 3 platform.
