thread::scope: document how join interacts with TLS destructors
Fixes https://github.com/rust-lang/rust/issues/116237 by documenting the current behavior regarding thread-local destructors as intended. (I'm not stoked about this, but documenting it is better than leaving it unclear.)
This also adds documentation for explicit `join` calls (both for scoped and regular threads), saying that those *will* wait for TLS destructors. That reflects my understanding of the current implementation, which calls `join` on the native thread handle. Are we okay with guaranteeing that? I think we should, so people have at least some chance of implementing "wait for all destructors" manually. This fixes https://github.com/rust-lang/rust/issues/127571.
Cc @rust-lang/libs-api
Remove outdated part of comment claiming thread_local re-enters global allocator
Fix typo in doc comment
Add comments for guarantees given and footnote that System may still be called
Revise mention of using the global allocator
Allow for the possibility that the global allocator is the system allocator.
Co-authored-by: Mark Rousskov <mark.simulacrum@gmail.com>
Replace "use this API instead" with a link to Builder::spawn. Edit
the paragraph to make it slightly clearer.
The Scope::spawn method already included a. Make the docs for the two
nearly the same.
Support `#[rustc_align_static]` inside `thread_local!`
Tracking issue: rust-lang/rust#146177
```rust
thread_local! {
#[rustc_align_static(64)]
static SO_ALIGNED: u64 = const { 0 };
}
```
This increases the amount of recursion the macro performs (once per attribute in addition to the previous once per item), making it easier to hit the recursion limit. I’ve added workarounds to limit the impact in the case of long doc comments, but this still needs a crater run just in case.
r? libs
``@rustbot`` label A-attributes A-macros A-thread-locals F-static_align T-libs
Accessing the thread id is often used in profiling and debugging, as
well as some approaches for sound single-threaded access to data.
Currently the only way to access the thread id is by first obtaining a
handle to the current thread. While this is not exactly slow, it does
require an atomic inc-ref and dec-ref operation, as well as the
injection of `Thread`'s drop code into the caller.
This publicly exposes the existing fast-path for accessing the current
thread id.
Migrate the standard library from using the external `cfg_if` crate to
using the now-built-in `cfg_select` macro.
This does not yet eliminate the dependency from
`library/std/Cargo.toml`, because while the standard library itself no
longer uses `cfg_if`, it also incorporates the `backtrace` crate, which
does.
Migration assisted by the following vim command (after selecting the
full `cfg_if!` invocation):
```
'<,'>s/\(cfg_if::\)\?cfg_if/cfg_select/ | '<,'>s/^\( *\)} else {/\1}\r\1_ => {/c | '<,'>s/^\( *\)} else if #\[cfg(\(.*\))\] /\1}\r\1\2 => /e | '<,'>s/if #\[cfg(\(.*\))\] {/\1 => {/e
```
This is imperfect, but substantially accelerated the process. This
prompts for confirmation on the `} else {` since that can also appear
inside one of the arms. This also requires manual intervention to handle
any multi-line conditions.
`panic!` does not print any identifying information for threads that are
unnamed. However, in many cases, the thread ID can be determined.
This changes the panic message from something like this:
thread '<unnamed>' panicked at src/main.rs:3:5:
explicit panic
To something like this:
thread '<unnamed>' (0xff9bf) panicked at src/main.rs:3:5:
explicit panic
Stack overflow messages are updated as well.
This change applies to both named and unnamed threads. The ID printed is
the OS integer thread ID rather than the Rust thread ID, which should
also be what debuggers print.
thread name in stack overflow message
Fixesrust-lang/rust#144481, which is caused by the thread name not being initialised yet when setting up the stack overflow information. Unfortunately, the stack overflow UI test did not test for the correct thread name being present, and testing this separately didn't occur to me when writing https://github.com/rust-lang/rust/pull/140628.
This PR contains the smallest possible fix I could think of: passing the thread name explicitly to the platform thread creation function. In the future I'd very much like to explore some possibilities around merging the thread packet and thread handle into one structure and using that in the platform code instead – but that's best left for another PR.
This PR also amends the stack overflow test to check for thread names, so we don't run into this again.
``@rustbot`` label +beta-nominated
Guarantee 8 bytes of alignment in Thread::into_raw
When using `AtomicPtr` for synchronization it's incredibly useful when you've got a couple bits you can stuff metadata in. By guaranteeing that `Thread`'s `Inner` struct is aligned to 8 bytes everyone can use the bottom 3 bits to signal other things, such as a critical section, etc.
This guarantee is thus very useful and costs us nothing.
Using clock nanosleep leads to more accurate sleep times on platforms
where it is supported.
To enable using clock_nanosleep this makes `sleep_until` platform
specific. That unfortunatly requires identical placeholder
implementations for the other platforms (windows/mac/wasm etc).
we will land platform specific implementations for those later. See the
`sleep_until` tracking issue.
This requires an accessors for the Instant type. As that accessor is only
used on the platforms that have clock_nanosleep it is marked as allow_unused.
32bit time_t targets do not use clock_nanosleep atm, they instead rely
on the same placeholder as the other platforms. We could make them
use clock_nanosleep too in the future using `__clock_nanosleep_time64`.
__clock_nanosleep_time64 is documented at:
https://www.gnu.org/software/libc/manual/html_node/64_002dbit-time-symbol-handling.html
Jules Bertholet