3 //! ## The threading model
5 //! An executing Rust program consists of a collection of native OS threads,
6 //! each with their own stack and local state. Threads can be named, and
7 //! provide some built-in support for low-level synchronization.
9 //! Communication between threads can be done through
10 //! [channels], Rust's message-passing types, along with [other forms of thread
11 //! synchronization](../../std/sync/index.html) and shared-memory data
12 //! structures. In particular, types that are guaranteed to be
13 //! threadsafe are easily shared between threads using the
14 //! atomically-reference-counted container, [`Arc`].
16 //! Fatal logic errors in Rust cause *thread panic*, during which
17 //! a thread will unwind the stack, running destructors and freeing
18 //! owned resources. While not meant as a 'try/catch' mechanism, panics
19 //! in Rust can nonetheless be caught (unless compiling with `panic=abort`) with
20 //! [`catch_unwind`](../../std/panic/fn.catch_unwind.html) and recovered
21 //! from, or alternatively be resumed with
22 //! [`resume_unwind`](../../std/panic/fn.resume_unwind.html). If the panic
23 //! is not caught the thread will exit, but the panic may optionally be
24 //! detected from a different thread with [`join`]. If the main thread panics
25 //! without the panic being caught, the application will exit with a
26 //! non-zero exit code.
28 //! When the main thread of a Rust program terminates, the entire program shuts
29 //! down, even if other threads are still running. However, this module provides
30 //! convenient facilities for automatically waiting for the termination of a
31 //! thread (i.e., join).
33 //! ## Spawning a thread
35 //! A new thread can be spawned using the [`thread::spawn`][`spawn`] function:
40 //! thread::spawn(move || {
45 //! In this example, the spawned thread is "detached," which means that there is
46 //! no way for the program to learn when the spawned thread completes or otherwise
49 //! To learn when a thread completes, it is necessary to capture the [`JoinHandle`]
50 //! object that is returned by the call to [`spawn`], which provides
51 //! a `join` method that allows the caller to wait for the completion of the
57 //! let thread_join_handle = thread::spawn(move || {
61 //! let res = thread_join_handle.join();
64 //! The [`join`] method returns a [`thread::Result`] containing [`Ok`] of the final
65 //! value produced by the spawned thread, or [`Err`] of the value given to
66 //! a call to [`panic!`] if the thread panicked.
68 //! Note that there is no parent/child relationship between a thread that spawns a
69 //! new thread and the thread being spawned. In particular, the spawned thread may or
70 //! may not outlive the spawning thread, unless the spawning thread is the main thread.
72 //! ## Configuring threads
74 //! A new thread can be configured before it is spawned via the [`Builder`] type,
75 //! which currently allows you to set the name and stack size for the thread:
78 //! # #![allow(unused_must_use)]
81 //! thread::Builder::new().name("thread1".to_string()).spawn(move || {
82 //! println!("Hello, world!");
86 //! ## The `Thread` type
88 //! Threads are represented via the [`Thread`] type, which you can get in one of
91 //! * By spawning a new thread, e.g., using the [`thread::spawn`][`spawn`]
92 //! function, and calling [`thread`][`JoinHandle::thread`] on the [`JoinHandle`].
93 //! * By requesting the current thread, using the [`thread::current`] function.
95 //! The [`thread::current`] function is available even for threads not spawned
96 //! by the APIs of this module.
98 //! ## Thread-local storage
100 //! This module also provides an implementation of thread-local storage for Rust
101 //! programs. Thread-local storage is a method of storing data into a global
102 //! variable that each thread in the program will have its own copy of.
103 //! Threads do not share this data, so accesses do not need to be synchronized.
105 //! A thread-local key owns the value it contains and will destroy the value when the
106 //! thread exits. It is created with the [`thread_local!`] macro and can contain any
107 //! value that is `'static` (no borrowed pointers). It provides an accessor function,
108 //! [`with`], that yields a shared reference to the value to the specified
109 //! closure. Thread-local keys allow only shared access to values, as there would be no
110 //! way to guarantee uniqueness if mutable borrows were allowed. Most values
111 //! will want to make use of some form of **interior mutability** through the
112 //! [`Cell`] or [`RefCell`] types.
114 //! ## Naming threads
116 //! Threads are able to have associated names for identification purposes. By default, spawned
117 //! threads are unnamed. To specify a name for a thread, build the thread with [`Builder`] and pass
118 //! the desired thread name to [`Builder::name`]. To retrieve the thread name from within the
119 //! thread, use [`Thread::name`]. A couple of examples where the name of a thread gets used:
121 //! * If a panic occurs in a named thread, the thread name will be printed in the panic message.
122 //! * The thread name is provided to the OS where applicable (e.g., `pthread_setname_np` in
123 //! unix-like platforms).
127 //! The default stack size for spawned threads is 2 MiB, though this particular stack size is
128 //! subject to change in the future. There are two ways to manually specify the stack size for
131 //! * Build the thread with [`Builder`] and pass the desired stack size to [`Builder::stack_size`].
132 //! * Set the `RUST_MIN_STACK` environment variable to an integer representing the desired stack
133 //! size (in bytes). Note that setting [`Builder::stack_size`] will override this.
135 //! Note that the stack size of the main thread is *not* determined by Rust.
137 //! [channels]: crate::sync::mpsc
138 //! [`join`]: JoinHandle::join
139 //! [`Result`]: crate::result::Result
140 //! [`Ok`]: crate::result::Result::Ok
141 //! [`Err`]: crate::result::Result::Err
142 //! [`thread::current`]: current
143 //! [`thread::Result`]: Result
144 //! [`unpark`]: Thread::unpark
145 //! [`thread::park_timeout`]: park_timeout
146 //! [`Cell`]: crate::cell::Cell
147 //! [`RefCell`]: crate::cell::RefCell
148 //! [`with`]: LocalKey::with
149 //! [`thread_local!`]: crate::thread_local
151 #![stable(feature = "rust1", since = "1.0.0")]
152 #![deny(unsafe_op_in_unsafe_fn)]
153 // Under `test`, `__FastLocalKeyInner` seems unused.
154 #![cfg_attr(test, allow(dead_code))]
156 #[cfg(all(test, not(target_os = "emscripten")))]
160 use crate::cell::UnsafeCell;
161 use crate::ffi::{CStr, CString};
164 use crate::marker::PhantomData;
165 use crate::mem::{self, forget};
166 use crate::num::NonZeroU64;
167 use crate::num::NonZeroUsize;
169 use crate::panicking;
171 use crate::ptr::addr_of_mut;
173 use crate::sync::Arc;
174 use crate::sys::thread as imp;
175 use crate::sys_common::thread;
176 use crate::sys_common::thread_info;
177 use crate::sys_common::thread_parker::Parker;
178 use crate::sys_common::{AsInner, IntoInner};
179 use crate::time::Duration;
181 ////////////////////////////////////////////////////////////////////////////////
182 // Thread-local storage
183 ////////////////////////////////////////////////////////////////////////////////
188 #[stable(feature = "scoped_threads", since = "1.63.0")]
191 #[stable(feature = "scoped_threads", since = "1.63.0")]
192 pub use scoped::{scope, Scope, ScopedJoinHandle};
194 #[stable(feature = "rust1", since = "1.0.0")]
195 pub use self::local::{AccessError, LocalKey};
197 // Provide the type used by the thread_local! macro to access TLS keys. This
198 // needs to be kept in sync with the macro itself (in `local.rs`).
199 // There are three types: "static", "fast", "OS". The "OS" thread local key
200 // type is accessed via platform-specific API calls and is slow, while the "fast"
201 // key type is accessed via code generated via LLVM, where TLS keys are set up
202 // by the elf linker. "static" is for single-threaded platforms where a global
203 // static is sufficient.
205 #[unstable(feature = "libstd_thread_internals", issue = "none")]
209 not(all(target_family = "wasm", not(target_feature = "atomics"))),
212 pub use self::local::fast::Key as __FastLocalKeyInner;
214 // when building for tests, use real std's type
215 #[unstable(feature = "libstd_thread_internals", issue = "none")]
219 not(all(target_family = "wasm", not(target_feature = "atomics"))),
221 pub use realstd::thread::__FastLocalKeyInner;
223 #[unstable(feature = "libstd_thread_internals", issue = "none")]
225 not(target_thread_local),
226 not(all(target_family = "wasm", not(target_feature = "atomics"))),
229 pub use self::local::os::Key as __OsLocalKeyInner;
231 #[unstable(feature = "libstd_thread_internals", issue = "none")]
232 #[cfg(all(target_family = "wasm", not(target_feature = "atomics")))]
234 pub use self::local::statik::Key as __StaticLocalKeyInner;
236 ////////////////////////////////////////////////////////////////////////////////
238 ////////////////////////////////////////////////////////////////////////////////
240 /// Thread factory, which can be used in order to configure the properties of
243 /// Methods can be chained on it in order to configure it.
245 /// The two configurations available are:
247 /// - [`name`]: specifies an [associated name for the thread][naming-threads]
248 /// - [`stack_size`]: specifies the [desired stack size for the thread][stack-size]
250 /// The [`spawn`] method will take ownership of the builder and create an
251 /// [`io::Result`] to the thread handle with the given configuration.
253 /// The [`thread::spawn`] free function uses a `Builder` with default
254 /// configuration and [`unwrap`]s its return value.
256 /// You may want to use [`spawn`] instead of [`thread::spawn`], when you want
257 /// to recover from a failure to launch a thread, indeed the free function will
258 /// panic where the `Builder` method will return a [`io::Result`].
265 /// let builder = thread::Builder::new();
267 /// let handler = builder.spawn(|| {
271 /// handler.join().unwrap();
274 /// [`stack_size`]: Builder::stack_size
275 /// [`name`]: Builder::name
276 /// [`spawn`]: Builder::spawn
277 /// [`thread::spawn`]: spawn
278 /// [`io::Result`]: crate::io::Result
279 /// [`unwrap`]: crate::result::Result::unwrap
280 /// [naming-threads]: ./index.html#naming-threads
281 /// [stack-size]: ./index.html#stack-size
282 #[must_use = "must eventually spawn the thread"]
283 #[stable(feature = "rust1", since = "1.0.0")]
286 // A name for the thread-to-be, for identification in panic messages
287 name: Option<String>,
288 // The size of the stack for the spawned thread in bytes
289 stack_size: Option<usize>,
293 /// Generates the base configuration for spawning a thread, from which
294 /// configuration methods can be chained.
301 /// let builder = thread::Builder::new()
302 /// .name("foo".into())
303 /// .stack_size(32 * 1024);
305 /// let handler = builder.spawn(|| {
309 /// handler.join().unwrap();
311 #[stable(feature = "rust1", since = "1.0.0")]
312 pub fn new() -> Builder {
313 Builder { name: None, stack_size: None }
316 /// Names the thread-to-be. Currently the name is used for identification
317 /// only in panic messages.
319 /// The name must not contain null bytes (`\0`).
321 /// For more information about named threads, see
322 /// [this module-level documentation][naming-threads].
329 /// let builder = thread::Builder::new()
330 /// .name("foo".into());
332 /// let handler = builder.spawn(|| {
333 /// assert_eq!(thread::current().name(), Some("foo"))
336 /// handler.join().unwrap();
339 /// [naming-threads]: ./index.html#naming-threads
340 #[stable(feature = "rust1", since = "1.0.0")]
341 pub fn name(mut self, name: String) -> Builder {
342 self.name = Some(name);
346 /// Sets the size of the stack (in bytes) for the new thread.
348 /// The actual stack size may be greater than this value if
349 /// the platform specifies a minimal stack size.
351 /// For more information about the stack size for threads, see
352 /// [this module-level documentation][stack-size].
359 /// let builder = thread::Builder::new().stack_size(32 * 1024);
362 /// [stack-size]: ./index.html#stack-size
363 #[stable(feature = "rust1", since = "1.0.0")]
364 pub fn stack_size(mut self, size: usize) -> Builder {
365 self.stack_size = Some(size);
369 /// Spawns a new thread by taking ownership of the `Builder`, and returns an
370 /// [`io::Result`] to its [`JoinHandle`].
372 /// The spawned thread may outlive the caller (unless the caller thread
373 /// is the main thread; the whole process is terminated when the main
374 /// thread finishes). The join handle can be used to block on
375 /// termination of the spawned thread, including recovering its panics.
377 /// For a more complete documentation see [`thread::spawn`][`spawn`].
381 /// Unlike the [`spawn`] free function, this method yields an
382 /// [`io::Result`] to capture any failure to create the thread at
385 /// [`io::Result`]: crate::io::Result
389 /// Panics if a thread name was set and it contained null bytes.
396 /// let builder = thread::Builder::new();
398 /// let handler = builder.spawn(|| {
402 /// handler.join().unwrap();
404 #[stable(feature = "rust1", since = "1.0.0")]
405 pub fn spawn<F, T>(self, f: F) -> io::Result<JoinHandle<T>>
411 unsafe { self.spawn_unchecked(f) }
414 /// Spawns a new thread without any lifetime restrictions by taking ownership
415 /// of the `Builder`, and returns an [`io::Result`] to its [`JoinHandle`].
417 /// The spawned thread may outlive the caller (unless the caller thread
418 /// is the main thread; the whole process is terminated when the main
419 /// thread finishes). The join handle can be used to block on
420 /// termination of the spawned thread, including recovering its panics.
422 /// This method is identical to [`thread::Builder::spawn`][`Builder::spawn`],
423 /// except for the relaxed lifetime bounds, which render it unsafe.
424 /// For a more complete documentation see [`thread::spawn`][`spawn`].
428 /// Unlike the [`spawn`] free function, this method yields an
429 /// [`io::Result`] to capture any failure to create the thread at
434 /// Panics if a thread name was set and it contained null bytes.
438 /// The caller has to ensure that the spawned thread does not outlive any
439 /// references in the supplied thread closure and its return type.
440 /// This can be guaranteed in two ways:
442 /// - ensure that [`join`][`JoinHandle::join`] is called before any referenced
444 /// - use only types with `'static` lifetime bounds, i.e., those with no or only
445 /// `'static` references (both [`thread::Builder::spawn`][`Builder::spawn`]
446 /// and [`thread::spawn`][`spawn`] enforce this property statically)
451 /// #![feature(thread_spawn_unchecked)]
454 /// let builder = thread::Builder::new();
457 /// let thread_x = &x;
459 /// let handler = unsafe {
460 /// builder.spawn_unchecked(move || {
461 /// println!("x = {}", *thread_x);
465 /// // caller has to ensure `join()` is called, otherwise
466 /// // it is possible to access freed memory if `x` gets
467 /// // dropped before the thread closure is executed!
468 /// handler.join().unwrap();
471 /// [`io::Result`]: crate::io::Result
472 #[unstable(feature = "thread_spawn_unchecked", issue = "55132")]
473 pub unsafe fn spawn_unchecked<'a, F, T>(self, f: F) -> io::Result<JoinHandle<T>>
479 Ok(JoinHandle(unsafe { self.spawn_unchecked_(f, None) }?))
482 unsafe fn spawn_unchecked_<'a, 'scope, F, T>(
485 scope_data: Option<Arc<scoped::ScopeData>>,
486 ) -> io::Result<JoinInner<'scope, T>>
493 let Builder { name, stack_size } = self;
495 let stack_size = stack_size.unwrap_or_else(thread::min_stack);
497 let my_thread = Thread::new(name.map(|name| {
498 CString::new(name).expect("thread name may not contain interior null bytes")
500 let their_thread = my_thread.clone();
502 let my_packet: Arc<Packet<'scope, T>> = Arc::new(Packet {
504 result: UnsafeCell::new(None),
505 _marker: PhantomData,
507 let their_packet = my_packet.clone();
509 let output_capture = crate::io::set_output_capture(None);
510 crate::io::set_output_capture(output_capture.clone());
512 // Pass `f` in `MaybeUninit` because actually that closure might *run longer than the lifetime of `F`*.
513 // See <https://github.com/rust-lang/rust/issues/101983> for more details.
514 // To prevent leaks we use a wrapper that drops its contents.
516 struct MaybeDangling<T>(mem::MaybeUninit<T>);
517 impl<T> MaybeDangling<T> {
518 fn new(x: T) -> Self {
519 MaybeDangling(mem::MaybeUninit::new(x))
521 fn into_inner(self) -> T {
522 // SAFETY: we are always initiailized.
523 let ret = unsafe { self.0.assume_init_read() };
524 // Make sure we don't drop.
529 impl<T> Drop for MaybeDangling<T> {
531 // SAFETY: we are always initiailized.
532 unsafe { self.0.assume_init_drop() };
536 let f = MaybeDangling::new(f);
538 if let Some(name) = their_thread.cname() {
539 imp::Thread::set_name(name);
542 crate::io::set_output_capture(output_capture);
544 // SAFETY: we constructed `f` initialized.
545 let f = f.into_inner();
546 // SAFETY: the stack guard passed is the one for the current thread.
547 // This means the current thread's stack and the new thread's stack
548 // are properly set and protected from each other.
549 thread_info::set(unsafe { imp::guard::current() }, their_thread);
550 let try_result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
551 crate::sys_common::backtrace::__rust_begin_short_backtrace(f)
553 // SAFETY: `their_packet` as been built just above and moved by the
554 // closure (it is an Arc<...>) and `my_packet` will be stored in the
555 // same `JoinInner` as this closure meaning the mutation will be
556 // safe (not modify it and affect a value far away).
557 unsafe { *their_packet.result.get() = Some(try_result) };
558 // Here `their_packet` gets dropped, and if this is the last `Arc` for that packet that
559 // will call `decrement_num_running_threads` and therefore signal that this thread is
562 // Here, the lifetime `'a` and even `'scope` can end. `main` keeps running for a bit
563 // after that before returning itself.
566 if let Some(scope_data) = &my_packet.scope {
567 scope_data.increment_num_running_threads();
573 // `imp::Thread::new` takes a closure with a `'static` lifetime, since it's passed
574 // through FFI or otherwise used with low-level threading primitives that have no
575 // notion of or way to enforce lifetimes.
577 // As mentioned in the `Safety` section of this function's documentation, the caller of
578 // this function needs to guarantee that the passed-in lifetime is sufficiently long
579 // for the lifetime of the thread.
581 // Similarly, the `sys` implementation must guarantee that no references to the closure
582 // exist after the thread has terminated, which is signaled by `Thread::join`
587 mem::transmute::<Box<dyn FnOnce() + 'a>, Box<dyn FnOnce() + 'static>>(
598 ////////////////////////////////////////////////////////////////////////////////
600 ////////////////////////////////////////////////////////////////////////////////
602 /// Spawns a new thread, returning a [`JoinHandle`] for it.
604 /// The join handle provides a [`join`] method that can be used to join the spawned
605 /// thread. If the spawned thread panics, [`join`] will return an [`Err`] containing
606 /// the argument given to [`panic!`].
608 /// If the join handle is dropped, the spawned thread will implicitly be *detached*.
609 /// In this case, the spawned thread may no longer be joined.
610 /// (It is the responsibility of the program to either eventually join threads it
611 /// creates or detach them; otherwise, a resource leak will result.)
613 /// This call will create a thread using default parameters of [`Builder`], if you
614 /// want to specify the stack size or the name of the thread, use this API
617 /// As you can see in the signature of `spawn` there are two constraints on
618 /// both the closure given to `spawn` and its return value, let's explain them:
620 /// - The `'static` constraint means that the closure and its return value
621 /// must have a lifetime of the whole program execution. The reason for this
622 /// is that threads can outlive the lifetime they have been created in.
624 /// Indeed if the thread, and by extension its return value, can outlive their
625 /// caller, we need to make sure that they will be valid afterwards, and since
626 /// we *can't* know when it will return we need to have them valid as long as
627 /// possible, that is until the end of the program, hence the `'static`
629 /// - The [`Send`] constraint is because the closure will need to be passed
630 /// *by value* from the thread where it is spawned to the new thread. Its
631 /// return value will need to be passed from the new thread to the thread
632 /// where it is `join`ed.
633 /// As a reminder, the [`Send`] marker trait expresses that it is safe to be
634 /// passed from thread to thread. [`Sync`] expresses that it is safe to have a
635 /// reference be passed from thread to thread.
639 /// Panics if the OS fails to create a thread; use [`Builder::spawn`]
640 /// to recover from such errors.
644 /// Creating a thread.
649 /// let handler = thread::spawn(|| {
653 /// handler.join().unwrap();
656 /// As mentioned in the module documentation, threads are usually made to
657 /// communicate using [`channels`], here is how it usually looks.
659 /// This example also shows how to use `move`, in order to give ownership
660 /// of values to a thread.
664 /// use std::sync::mpsc::channel;
666 /// let (tx, rx) = channel();
668 /// let sender = thread::spawn(move || {
669 /// tx.send("Hello, thread".to_owned())
670 /// .expect("Unable to send on channel");
673 /// let receiver = thread::spawn(move || {
674 /// let value = rx.recv().expect("Unable to receive from channel");
675 /// println!("{value}");
678 /// sender.join().expect("The sender thread has panicked");
679 /// receiver.join().expect("The receiver thread has panicked");
682 /// A thread can also return a value through its [`JoinHandle`], you can use
683 /// this to make asynchronous computations (futures might be more appropriate
689 /// let computation = thread::spawn(|| {
690 /// // Some expensive computation.
694 /// let result = computation.join().unwrap();
695 /// println!("{result}");
698 /// [`channels`]: crate::sync::mpsc
699 /// [`join`]: JoinHandle::join
700 /// [`Err`]: crate::result::Result::Err
701 #[stable(feature = "rust1", since = "1.0.0")]
702 pub fn spawn<F, T>(f: F) -> JoinHandle<T>
708 Builder::new().spawn(f).expect("failed to spawn thread")
711 /// Gets a handle to the thread that invokes it.
715 /// Getting a handle to the current thread with `thread::current()`:
720 /// let handler = thread::Builder::new()
721 /// .name("named thread".into())
723 /// let handle = thread::current();
724 /// assert_eq!(handle.name(), Some("named thread"));
728 /// handler.join().unwrap();
731 #[stable(feature = "rust1", since = "1.0.0")]
732 pub fn current() -> Thread {
733 thread_info::current_thread().expect(
734 "use of std::thread::current() is not possible \
735 after the thread's local data has been destroyed",
739 /// Cooperatively gives up a timeslice to the OS scheduler.
741 /// This calls the underlying OS scheduler's yield primitive, signaling
742 /// that the calling thread is willing to give up its remaining timeslice
743 /// so that the OS may schedule other threads on the CPU.
745 /// A drawback of yielding in a loop is that if the OS does not have any
746 /// other ready threads to run on the current CPU, the thread will effectively
747 /// busy-wait, which wastes CPU time and energy.
749 /// Therefore, when waiting for events of interest, a programmer's first
750 /// choice should be to use synchronization devices such as [`channel`]s,
751 /// [`Condvar`]s, [`Mutex`]es or [`join`] since these primitives are
752 /// implemented in a blocking manner, giving up the CPU until the event
753 /// of interest has occurred which avoids repeated yielding.
755 /// `yield_now` should thus be used only rarely, mostly in situations where
756 /// repeated polling is required because there is no other suitable way to
757 /// learn when an event of interest has occurred.
764 /// thread::yield_now();
767 /// [`channel`]: crate::sync::mpsc
768 /// [`join`]: JoinHandle::join
769 /// [`Condvar`]: crate::sync::Condvar
770 /// [`Mutex`]: crate::sync::Mutex
771 #[stable(feature = "rust1", since = "1.0.0")]
773 imp::Thread::yield_now()
776 /// Determines whether the current thread is unwinding because of panic.
778 /// A common use of this feature is to poison shared resources when writing
779 /// unsafe code, by checking `panicking` when the `drop` is called.
781 /// This is usually not needed when writing safe code, as [`Mutex`es][Mutex]
782 /// already poison themselves when a thread panics while holding the lock.
784 /// This can also be used in multithreaded applications, in order to send a
785 /// message to other threads warning that a thread has panicked (e.g., for
786 /// monitoring purposes).
793 /// struct SomeStruct;
795 /// impl Drop for SomeStruct {
796 /// fn drop(&mut self) {
797 /// if thread::panicking() {
798 /// println!("dropped while unwinding");
800 /// println!("dropped while not unwinding");
807 /// let a = SomeStruct;
812 /// let b = SomeStruct;
817 /// [Mutex]: crate::sync::Mutex
820 #[stable(feature = "rust1", since = "1.0.0")]
821 pub fn panicking() -> bool {
822 panicking::panicking()
827 /// Puts the current thread to sleep for at least the specified amount of time.
829 /// The thread may sleep longer than the duration specified due to scheduling
830 /// specifics or platform-dependent functionality. It will never sleep less.
832 /// This function is blocking, and should not be used in `async` functions.
834 /// # Platform-specific behavior
836 /// On Unix platforms, the underlying syscall may be interrupted by a
837 /// spurious wakeup or signal handler. To ensure the sleep occurs for at least
838 /// the specified duration, this function may invoke that system call multiple
846 /// // Let's sleep for 2 seconds:
847 /// thread::sleep_ms(2000);
849 #[stable(feature = "rust1", since = "1.0.0")]
850 #[deprecated(since = "1.6.0", note = "replaced by `std::thread::sleep`")]
851 pub fn sleep_ms(ms: u32) {
852 sleep(Duration::from_millis(ms as u64))
855 /// Puts the current thread to sleep for at least the specified amount of time.
857 /// The thread may sleep longer than the duration specified due to scheduling
858 /// specifics or platform-dependent functionality. It will never sleep less.
860 /// This function is blocking, and should not be used in `async` functions.
862 /// # Platform-specific behavior
864 /// On Unix platforms, the underlying syscall may be interrupted by a
865 /// spurious wakeup or signal handler. To ensure the sleep occurs for at least
866 /// the specified duration, this function may invoke that system call multiple
868 /// Platforms which do not support nanosecond precision for sleeping will
869 /// have `dur` rounded up to the nearest granularity of time they can sleep for.
871 /// Currently, specifying a zero duration on Unix platforms returns immediately
872 /// without invoking the underlying [`nanosleep`] syscall, whereas on Windows
873 /// platforms the underlying [`Sleep`] syscall is always invoked.
874 /// If the intention is to yield the current time-slice you may want to use
875 /// [`yield_now`] instead.
877 /// [`nanosleep`]: https://linux.die.net/man/2/nanosleep
878 /// [`Sleep`]: https://docs.microsoft.com/en-us/windows/win32/api/synchapi/nf-synchapi-sleep
883 /// use std::{thread, time};
885 /// let ten_millis = time::Duration::from_millis(10);
886 /// let now = time::Instant::now();
888 /// thread::sleep(ten_millis);
890 /// assert!(now.elapsed() >= ten_millis);
892 #[stable(feature = "thread_sleep", since = "1.4.0")]
893 pub fn sleep(dur: Duration) {
894 imp::Thread::sleep(dur)
897 /// Used to ensure that `park` and `park_timeout` do not unwind, as that can
898 /// cause undefined behaviour if not handled correctly (see #102398 for context).
901 impl Drop for PanicGuard {
903 rtabort!("an irrecoverable error occurred while synchronizing threads")
907 /// Blocks unless or until the current thread's token is made available.
909 /// A call to `park` does not guarantee that the thread will remain parked
910 /// forever, and callers should be prepared for this possibility. However,
911 /// it is guaranteed that this function will not panic (it may abort the
912 /// process if the implementation encounters some rare errors).
914 /// # park and unpark
916 /// Every thread is equipped with some basic low-level blocking support, via the
917 /// [`thread::park`][`park`] function and [`thread::Thread::unpark`][`unpark`]
918 /// method. [`park`] blocks the current thread, which can then be resumed from
919 /// another thread by calling the [`unpark`] method on the blocked thread's
922 /// Conceptually, each [`Thread`] handle has an associated token, which is
923 /// initially not present:
925 /// * The [`thread::park`][`park`] function blocks the current thread unless or
926 /// until the token is available for its thread handle, at which point it
927 /// atomically consumes the token. It may also return *spuriously*, without
928 /// consuming the token. [`thread::park_timeout`] does the same, but allows
929 /// specifying a maximum time to block the thread for.
931 /// * The [`unpark`] method on a [`Thread`] atomically makes the token available
932 /// if it wasn't already. Because the token is initially absent, [`unpark`]
933 /// followed by [`park`] will result in the second call returning immediately.
935 /// In other words, each [`Thread`] acts a bit like a spinlock that can be
936 /// locked and unlocked using `park` and `unpark`.
938 /// Notice that being unblocked does not imply any synchronization with someone
939 /// that unparked this thread, it could also be spurious.
940 /// For example, it would be a valid, but inefficient, implementation to make both [`park`] and
941 /// [`unpark`] return immediately without doing anything.
943 /// The API is typically used by acquiring a handle to the current thread,
944 /// placing that handle in a shared data structure so that other threads can
945 /// find it, and then `park`ing in a loop. When some desired condition is met, another
946 /// thread calls [`unpark`] on the handle.
948 /// The motivation for this design is twofold:
950 /// * It avoids the need to allocate mutexes and condvars when building new
951 /// synchronization primitives; the threads already provide basic
952 /// blocking/signaling.
954 /// * It can be implemented very efficiently on many platforms.
960 /// use std::sync::{Arc, atomic::{Ordering, AtomicBool}};
961 /// use std::time::Duration;
963 /// let flag = Arc::new(AtomicBool::new(false));
964 /// let flag2 = Arc::clone(&flag);
966 /// let parked_thread = thread::spawn(move || {
967 /// // We want to wait until the flag is set. We *could* just spin, but using
968 /// // park/unpark is more efficient.
969 /// while !flag2.load(Ordering::Acquire) {
970 /// println!("Parking thread");
972 /// // We *could* get here spuriously, i.e., way before the 10ms below are over!
973 /// // But that is no problem, we are in a loop until the flag is set anyway.
974 /// println!("Thread unparked");
976 /// println!("Flag received");
979 /// // Let some time pass for the thread to be spawned.
980 /// thread::sleep(Duration::from_millis(10));
982 /// // Set the flag, and let the thread wake up.
983 /// // There is no race condition here, if `unpark`
984 /// // happens first, `park` will return immediately.
985 /// // Hence there is no risk of a deadlock.
986 /// flag.store(true, Ordering::Release);
987 /// println!("Unpark the thread");
988 /// parked_thread.thread().unpark();
990 /// parked_thread.join().unwrap();
993 /// [`unpark`]: Thread::unpark
994 /// [`thread::park_timeout`]: park_timeout
995 #[stable(feature = "rust1", since = "1.0.0")]
997 let guard = PanicGuard;
998 // SAFETY: park_timeout is called on the parker owned by this thread.
1000 current().inner.as_ref().parker().park();
1002 // No panic occurred, do not abort.
1006 /// Use [`park_timeout`].
1008 /// Blocks unless or until the current thread's token is made available or
1009 /// the specified duration has been reached (may wake spuriously).
1011 /// The semantics of this function are equivalent to [`park`] except
1012 /// that the thread will be blocked for roughly no longer than `dur`. This
1013 /// method should not be used for precise timing due to anomalies such as
1014 /// preemption or platform differences that might not cause the maximum
1015 /// amount of time waited to be precisely `ms` long.
1017 /// See the [park documentation][`park`] for more detail.
1018 #[stable(feature = "rust1", since = "1.0.0")]
1019 #[deprecated(since = "1.6.0", note = "replaced by `std::thread::park_timeout`")]
1020 pub fn park_timeout_ms(ms: u32) {
1021 park_timeout(Duration::from_millis(ms as u64))
1024 /// Blocks unless or until the current thread's token is made available or
1025 /// the specified duration has been reached (may wake spuriously).
1027 /// The semantics of this function are equivalent to [`park`][park] except
1028 /// that the thread will be blocked for roughly no longer than `dur`. This
1029 /// method should not be used for precise timing due to anomalies such as
1030 /// preemption or platform differences that might not cause the maximum
1031 /// amount of time waited to be precisely `dur` long.
1033 /// See the [park documentation][park] for more details.
1035 /// # Platform-specific behavior
1037 /// Platforms which do not support nanosecond precision for sleeping will have
1038 /// `dur` rounded up to the nearest granularity of time they can sleep for.
1042 /// Waiting for the complete expiration of the timeout:
1045 /// use std::thread::park_timeout;
1046 /// use std::time::{Instant, Duration};
1048 /// let timeout = Duration::from_secs(2);
1049 /// let beginning_park = Instant::now();
1051 /// let mut timeout_remaining = timeout;
1053 /// park_timeout(timeout_remaining);
1054 /// let elapsed = beginning_park.elapsed();
1055 /// if elapsed >= timeout {
1058 /// println!("restarting park_timeout after {elapsed:?}");
1059 /// timeout_remaining = timeout - elapsed;
1062 #[stable(feature = "park_timeout", since = "1.4.0")]
1063 pub fn park_timeout(dur: Duration) {
1064 let guard = PanicGuard;
1065 // SAFETY: park_timeout is called on the parker owned by this thread.
1067 current().inner.as_ref().parker().park_timeout(dur);
1069 // No panic occurred, do not abort.
1073 ////////////////////////////////////////////////////////////////////////////////
1075 ////////////////////////////////////////////////////////////////////////////////
1077 /// A unique identifier for a running thread.
1079 /// A `ThreadId` is an opaque object that uniquely identifies each thread
1080 /// created during the lifetime of a process. `ThreadId`s are guaranteed not to
1081 /// be reused, even when a thread terminates. `ThreadId`s are under the control
1082 /// of Rust's standard library and there may not be any relationship between
1083 /// `ThreadId` and the underlying platform's notion of a thread identifier --
1084 /// the two concepts cannot, therefore, be used interchangeably. A `ThreadId`
1085 /// can be retrieved from the [`id`] method on a [`Thread`].
1090 /// use std::thread;
1092 /// let other_thread = thread::spawn(|| {
1093 /// thread::current().id()
1096 /// let other_thread_id = other_thread.join().unwrap();
1097 /// assert!(thread::current().id() != other_thread_id);
1100 /// [`id`]: Thread::id
1101 #[stable(feature = "thread_id", since = "1.19.0")]
1102 #[derive(Eq, PartialEq, Clone, Copy, Hash, Debug)]
1103 pub struct ThreadId(NonZeroU64);
1106 // Generate a new unique thread ID.
1107 fn new() -> ThreadId {
1109 fn exhausted() -> ! {
1110 panic!("failed to generate unique thread ID: bitspace exhausted")
1114 if #[cfg(target_has_atomic = "64")] {
1115 use crate::sync::atomic::{AtomicU64, Ordering::Relaxed};
1117 static COUNTER: AtomicU64 = AtomicU64::new(0);
1119 let mut last = COUNTER.load(Relaxed);
1121 let Some(id) = last.checked_add(1) else {
1125 match COUNTER.compare_exchange_weak(last, id, Relaxed, Relaxed) {
1126 Ok(_) => return ThreadId(NonZeroU64::new(id).unwrap()),
1127 Err(id) => last = id,
1131 use crate::sys_common::mutex::StaticMutex;
1133 // It is UB to attempt to acquire this mutex reentrantly!
1134 static GUARD: StaticMutex = StaticMutex::new();
1135 static mut COUNTER: u64 = 0;
1138 let guard = GUARD.lock();
1140 let Some(id) = COUNTER.checked_add(1) else {
1141 drop(guard); // in case the panic handler ends up calling `ThreadId::new()`, avoid reentrant lock acquire.
1147 ThreadId(NonZeroU64::new(id).unwrap())
1153 /// This returns a numeric identifier for the thread identified by this
1156 /// As noted in the documentation for the type itself, it is essentially an
1157 /// opaque ID, but is guaranteed to be unique for each thread. The returned
1158 /// value is entirely opaque -- only equality testing is stable. Note that
1159 /// it is not guaranteed which values new threads will return, and this may
1160 /// change across Rust versions.
1162 #[unstable(feature = "thread_id_value", issue = "67939")]
1163 pub fn as_u64(&self) -> NonZeroU64 {
1168 ////////////////////////////////////////////////////////////////////////////////
1170 ////////////////////////////////////////////////////////////////////////////////
1172 /// The internal representation of a `Thread` handle
1174 name: Option<CString>, // Guaranteed to be UTF-8
1180 fn parker(self: Pin<&Self>) -> Pin<&Parker> {
1181 unsafe { Pin::map_unchecked(self, |inner| &inner.parker) }
1186 #[stable(feature = "rust1", since = "1.0.0")]
1187 /// A handle to a thread.
1189 /// Threads are represented via the `Thread` type, which you can get in one of
1192 /// * By spawning a new thread, e.g., using the [`thread::spawn`][`spawn`]
1193 /// function, and calling [`thread`][`JoinHandle::thread`] on the
1195 /// * By requesting the current thread, using the [`thread::current`] function.
1197 /// The [`thread::current`] function is available even for threads not spawned
1198 /// by the APIs of this module.
1200 /// There is usually no need to create a `Thread` struct yourself, one
1201 /// should instead use a function like `spawn` to create new threads, see the
1202 /// docs of [`Builder`] and [`spawn`] for more details.
1204 /// [`thread::current`]: current
1206 inner: Pin<Arc<Inner>>,
1210 // Used only internally to construct a thread object without spawning
1211 // Panics if the name contains nuls.
1212 pub(crate) fn new(name: Option<CString>) -> Thread {
1213 // We have to use `unsafe` here to construct the `Parker` in-place,
1214 // which is required for the UNIX implementation.
1216 // SAFETY: We pin the Arc immediately after creation, so its address never
1218 let inner = unsafe {
1219 let mut arc = Arc::<Inner>::new_uninit();
1220 let ptr = Arc::get_mut_unchecked(&mut arc).as_mut_ptr();
1221 addr_of_mut!((*ptr).name).write(name);
1222 addr_of_mut!((*ptr).id).write(ThreadId::new());
1223 Parker::new(addr_of_mut!((*ptr).parker));
1224 Pin::new_unchecked(arc.assume_init())
1230 /// Atomically makes the handle's token available if it is not already.
1232 /// Every thread is equipped with some basic low-level blocking support, via
1233 /// the [`park`][park] function and the `unpark()` method. These can be
1234 /// used as a more CPU-efficient implementation of a spinlock.
1236 /// See the [park documentation][park] for more details.
1241 /// use std::thread;
1242 /// use std::time::Duration;
1244 /// let parked_thread = thread::Builder::new()
1246 /// println!("Parking thread");
1248 /// println!("Thread unparked");
1252 /// // Let some time pass for the thread to be spawned.
1253 /// thread::sleep(Duration::from_millis(10));
1255 /// println!("Unpark the thread");
1256 /// parked_thread.thread().unpark();
1258 /// parked_thread.join().unwrap();
1260 #[stable(feature = "rust1", since = "1.0.0")]
1262 pub fn unpark(&self) {
1263 self.inner.as_ref().parker().unpark();
1266 /// Gets the thread's unique identifier.
1271 /// use std::thread;
1273 /// let other_thread = thread::spawn(|| {
1274 /// thread::current().id()
1277 /// let other_thread_id = other_thread.join().unwrap();
1278 /// assert!(thread::current().id() != other_thread_id);
1280 #[stable(feature = "thread_id", since = "1.19.0")]
1282 pub fn id(&self) -> ThreadId {
1286 /// Gets the thread's name.
1288 /// For more information about named threads, see
1289 /// [this module-level documentation][naming-threads].
1293 /// Threads by default have no name specified:
1296 /// use std::thread;
1298 /// let builder = thread::Builder::new();
1300 /// let handler = builder.spawn(|| {
1301 /// assert!(thread::current().name().is_none());
1304 /// handler.join().unwrap();
1307 /// Thread with a specified name:
1310 /// use std::thread;
1312 /// let builder = thread::Builder::new()
1313 /// .name("foo".into());
1315 /// let handler = builder.spawn(|| {
1316 /// assert_eq!(thread::current().name(), Some("foo"))
1319 /// handler.join().unwrap();
1322 /// [naming-threads]: ./index.html#naming-threads
1323 #[stable(feature = "rust1", since = "1.0.0")]
1325 pub fn name(&self) -> Option<&str> {
1326 self.cname().map(|s| unsafe { str::from_utf8_unchecked(s.to_bytes()) })
1329 fn cname(&self) -> Option<&CStr> {
1330 self.inner.name.as_deref()
1334 #[stable(feature = "rust1", since = "1.0.0")]
1335 impl fmt::Debug for Thread {
1336 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1337 f.debug_struct("Thread")
1338 .field("id", &self.id())
1339 .field("name", &self.name())
1340 .finish_non_exhaustive()
1344 ////////////////////////////////////////////////////////////////////////////////
1346 ////////////////////////////////////////////////////////////////////////////////
1348 /// A specialized [`Result`] type for threads.
1350 /// Indicates the manner in which a thread exited.
1352 /// The value contained in the `Result::Err` variant
1353 /// is the value the thread panicked with;
1354 /// that is, the argument the `panic!` macro was called with.
1355 /// Unlike with normal errors, this value doesn't implement
1356 /// the [`Error`](crate::error::Error) trait.
1358 /// Thus, a sensible way to handle a thread panic is to either:
1360 /// 1. propagate the panic with [`std::panic::resume_unwind`]
1361 /// 2. or in case the thread is intended to be a subsystem boundary
1362 /// that is supposed to isolate system-level failures,
1363 /// match on the `Err` variant and handle the panic in an appropriate way
1365 /// A thread that completes without panicking is considered to exit successfully.
1369 /// Matching on the result of a joined thread:
1372 /// use std::{fs, thread, panic};
1374 /// fn copy_in_thread() -> thread::Result<()> {
1375 /// thread::spawn(|| {
1376 /// fs::copy("foo.txt", "bar.txt").unwrap();
1381 /// match copy_in_thread() {
1382 /// Ok(_) => println!("copy succeeded"),
1383 /// Err(e) => panic::resume_unwind(e),
1388 /// [`Result`]: crate::result::Result
1389 /// [`std::panic::resume_unwind`]: crate::panic::resume_unwind
1390 #[stable(feature = "rust1", since = "1.0.0")]
1391 pub type Result<T> = crate::result::Result<T, Box<dyn Any + Send + 'static>>;
1393 // This packet is used to communicate the return value between the spawned
1394 // thread and the rest of the program. It is shared through an `Arc` and
1395 // there's no need for a mutex here because synchronization happens with `join()`
1396 // (the caller will never read this packet until the thread has exited).
1398 // An Arc to the packet is stored into a `JoinInner` which in turns is placed
1400 struct Packet<'scope, T> {
1401 scope: Option<Arc<scoped::ScopeData>>,
1402 result: UnsafeCell<Option<Result<T>>>,
1403 _marker: PhantomData<Option<&'scope scoped::ScopeData>>,
1406 // Due to the usage of `UnsafeCell` we need to manually implement Sync.
1407 // The type `T` should already always be Send (otherwise the thread could not
1408 // have been created) and the Packet is Sync because all access to the
1409 // `UnsafeCell` synchronized (by the `join()` boundary), and `ScopeData` is Sync.
1410 unsafe impl<'scope, T: Sync> Sync for Packet<'scope, T> {}
1412 impl<'scope, T> Drop for Packet<'scope, T> {
1413 fn drop(&mut self) {
1414 // If this packet was for a thread that ran in a scope, the thread
1415 // panicked, and nobody consumed the panic payload, we make sure
1416 // the scope function will panic.
1417 let unhandled_panic = matches!(self.result.get_mut(), Some(Err(_)));
1418 // Drop the result without causing unwinding.
1419 // This is only relevant for threads that aren't join()ed, as
1420 // join() will take the `result` and set it to None, such that
1421 // there is nothing left to drop here.
1422 // If this panics, we should handle that, because we're outside the
1423 // outermost `catch_unwind` of our thread.
1424 // We just abort in that case, since there's nothing else we can do.
1425 // (And even if we tried to handle it somehow, we'd also need to handle
1426 // the case where the panic payload we get out of it also panics on
1427 // drop, and so on. See issue #86027.)
1428 if let Err(_) = panic::catch_unwind(panic::AssertUnwindSafe(|| {
1429 *self.result.get_mut() = None;
1431 rtabort!("thread result panicked on drop");
1433 // Book-keeping so the scope knows when it's done.
1434 if let Some(scope) = &self.scope {
1435 // Now that there will be no more user code running on this thread
1436 // that can use 'scope, mark the thread as 'finished'.
1437 // It's important we only do this after the `result` has been dropped,
1438 // since dropping it might still use things it borrowed from 'scope.
1439 scope.decrement_num_running_threads(unhandled_panic);
1444 /// Inner representation for JoinHandle
1445 struct JoinInner<'scope, T> {
1446 native: imp::Thread,
1448 packet: Arc<Packet<'scope, T>>,
1451 impl<'scope, T> JoinInner<'scope, T> {
1452 fn join(mut self) -> Result<T> {
1454 Arc::get_mut(&mut self.packet).unwrap().result.get_mut().take().unwrap()
1458 /// An owned permission to join on a thread (block on its termination).
1460 /// A `JoinHandle` *detaches* the associated thread when it is dropped, which
1461 /// means that there is no longer any handle to the thread and no way to `join`
1464 /// Due to platform restrictions, it is not possible to [`Clone`] this
1465 /// handle: the ability to join a thread is a uniquely-owned permission.
1467 /// This `struct` is created by the [`thread::spawn`] function and the
1468 /// [`thread::Builder::spawn`] method.
1472 /// Creation from [`thread::spawn`]:
1475 /// use std::thread;
1477 /// let join_handle: thread::JoinHandle<_> = thread::spawn(|| {
1478 /// // some work here
1482 /// Creation from [`thread::Builder::spawn`]:
1485 /// use std::thread;
1487 /// let builder = thread::Builder::new();
1489 /// let join_handle: thread::JoinHandle<_> = builder.spawn(|| {
1490 /// // some work here
1494 /// A thread being detached and outliving the thread that spawned it:
1497 /// use std::thread;
1498 /// use std::time::Duration;
1500 /// let original_thread = thread::spawn(|| {
1501 /// let _detached_thread = thread::spawn(|| {
1502 /// // Here we sleep to make sure that the first thread returns before.
1503 /// thread::sleep(Duration::from_millis(10));
1504 /// // This will be called, even though the JoinHandle is dropped.
1505 /// println!("♫ Still alive ♫");
1509 /// original_thread.join().expect("The thread being joined has panicked");
1510 /// println!("Original thread is joined.");
1512 /// // We make sure that the new thread has time to run, before the main
1513 /// // thread returns.
1515 /// thread::sleep(Duration::from_millis(1000));
1518 /// [`thread::Builder::spawn`]: Builder::spawn
1519 /// [`thread::spawn`]: spawn
1520 #[stable(feature = "rust1", since = "1.0.0")]
1521 pub struct JoinHandle<T>(JoinInner<'static, T>);
1523 #[stable(feature = "joinhandle_impl_send_sync", since = "1.29.0")]
1524 unsafe impl<T> Send for JoinHandle<T> {}
1525 #[stable(feature = "joinhandle_impl_send_sync", since = "1.29.0")]
1526 unsafe impl<T> Sync for JoinHandle<T> {}
1528 impl<T> JoinHandle<T> {
1529 /// Extracts a handle to the underlying thread.
1534 /// use std::thread;
1536 /// let builder = thread::Builder::new();
1538 /// let join_handle: thread::JoinHandle<_> = builder.spawn(|| {
1539 /// // some work here
1542 /// let thread = join_handle.thread();
1543 /// println!("thread id: {:?}", thread.id());
1545 #[stable(feature = "rust1", since = "1.0.0")]
1547 pub fn thread(&self) -> &Thread {
1551 /// Waits for the associated thread to finish.
1553 /// This function will return immediately if the associated thread has already finished.
1555 /// In terms of [atomic memory orderings], the completion of the associated
1556 /// thread synchronizes with this function returning. In other words, all
1557 /// operations performed by that thread [happen
1558 /// before](https://doc.rust-lang.org/nomicon/atomics.html#data-accesses) all
1559 /// operations that happen after `join` returns.
1561 /// If the associated thread panics, [`Err`] is returned with the parameter given
1564 /// [`Err`]: crate::result::Result::Err
1565 /// [atomic memory orderings]: crate::sync::atomic
1569 /// This function may panic on some platforms if a thread attempts to join
1570 /// itself or otherwise may create a deadlock with joining threads.
1575 /// use std::thread;
1577 /// let builder = thread::Builder::new();
1579 /// let join_handle: thread::JoinHandle<_> = builder.spawn(|| {
1580 /// // some work here
1582 /// join_handle.join().expect("Couldn't join on the associated thread");
1584 #[stable(feature = "rust1", since = "1.0.0")]
1585 pub fn join(self) -> Result<T> {
1589 /// Checks if the associated thread has finished running its main function.
1591 /// `is_finished` supports implementing a non-blocking join operation, by checking
1592 /// `is_finished`, and calling `join` if it returns `true`. This function does not block. To
1593 /// block while waiting on the thread to finish, use [`join`][Self::join].
1595 /// This might return `true` for a brief moment after the thread's main
1596 /// function has returned, but before the thread itself has stopped running.
1597 /// However, once this returns `true`, [`join`][Self::join] can be expected
1598 /// to return quickly, without blocking for any significant amount of time.
1599 #[stable(feature = "thread_is_running", since = "1.61.0")]
1600 pub fn is_finished(&self) -> bool {
1601 Arc::strong_count(&self.0.packet) == 1
1605 impl<T> AsInner<imp::Thread> for JoinHandle<T> {
1606 fn as_inner(&self) -> &imp::Thread {
1611 impl<T> IntoInner<imp::Thread> for JoinHandle<T> {
1612 fn into_inner(self) -> imp::Thread {
1617 #[stable(feature = "std_debug", since = "1.16.0")]
1618 impl<T> fmt::Debug for JoinHandle<T> {
1619 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1620 f.debug_struct("JoinHandle").finish_non_exhaustive()
1624 fn _assert_sync_and_send() {
1625 fn _assert_both<T: Send + Sync>() {}
1626 _assert_both::<JoinHandle<()>>();
1627 _assert_both::<Thread>();
1630 /// Returns an estimate of the default amount of parallelism a program should use.
1632 /// Parallelism is a resource. A given machine provides a certain capacity for
1633 /// parallelism, i.e., a bound on the number of computations it can perform
1634 /// simultaneously. This number often corresponds to the amount of CPUs a
1635 /// computer has, but it may diverge in various cases.
1637 /// Host environments such as VMs or container orchestrators may want to
1638 /// restrict the amount of parallelism made available to programs in them. This
1639 /// is often done to limit the potential impact of (unintentionally)
1640 /// resource-intensive programs on other programs running on the same machine.
1644 /// The purpose of this API is to provide an easy and portable way to query
1645 /// the default amount of parallelism the program should use. Among other things it
1646 /// does not expose information on NUMA regions, does not account for
1647 /// differences in (co)processor capabilities or current system load,
1648 /// and will not modify the program's global state in order to more accurately
1649 /// query the amount of available parallelism.
1651 /// Where both fixed steady-state and burst limits are available the steady-state
1652 /// capacity will be used to ensure more predictable latencies.
1654 /// Resource limits can be changed during the runtime of a program, therefore the value is
1655 /// not cached and instead recomputed every time this function is called. It should not be
1656 /// called from hot code.
1658 /// The value returned by this function should be considered a simplified
1659 /// approximation of the actual amount of parallelism available at any given
1660 /// time. To get a more detailed or precise overview of the amount of
1661 /// parallelism available to the program, you may wish to use
1662 /// platform-specific APIs as well. The following platform limitations currently
1663 /// apply to `available_parallelism`:
1666 /// - It may undercount the amount of parallelism available on systems with more
1667 /// than 64 logical CPUs. However, programs typically need specific support to
1668 /// take advantage of more than 64 logical CPUs, and in the absence of such
1669 /// support, the number returned by this function accurately reflects the
1670 /// number of logical CPUs the program can use by default.
1671 /// - It may overcount the amount of parallelism available on systems limited by
1672 /// process-wide affinity masks, or job object limitations.
1675 /// - It may overcount the amount of parallelism available when limited by a
1676 /// process-wide affinity mask or cgroup quotas and `sched_getaffinity()` or cgroup fs can't be
1677 /// queried, e.g. due to sandboxing.
1678 /// - It may undercount the amount of parallelism if the current thread's affinity mask
1679 /// does not reflect the process' cpuset, e.g. due to pinned threads.
1680 /// - If the process is in a cgroup v1 cpu controller, this may need to
1681 /// scan mountpoints to find the corresponding cgroup v1 controller,
1682 /// which may take time on systems with large numbers of mountpoints.
1683 /// (This does not apply to cgroup v2, or to processes not in a
1687 /// - It may overcount the amount of parallelism available when running in a VM
1688 /// with CPU usage limits (e.g. an overcommitted host).
1692 /// This function will, but is not limited to, return errors in the following
1695 /// - If the amount of parallelism is not known for the target platform.
1696 /// - If the program lacks permission to query the amount of parallelism made
1697 /// available to it.
1702 /// # #![allow(dead_code)]
1703 /// use std::{io, thread};
1705 /// fn main() -> io::Result<()> {
1706 /// let count = thread::available_parallelism()?.get();
1707 /// assert!(count >= 1_usize);
1711 #[doc(alias = "available_concurrency")] // Alias for a previous name we gave this API on unstable.
1712 #[doc(alias = "hardware_concurrency")] // Alias for C++ `std::thread::hardware_concurrency`.
1713 #[doc(alias = "num_cpus")] // Alias for a popular ecosystem crate which provides similar functionality.
1714 #[stable(feature = "available_parallelism", since = "1.59.0")]
1715 pub fn available_parallelism() -> io::Result<NonZeroUsize> {
1716 imp::available_parallelism()