Struct Guard

pub struct Guard { /* private fields */ }

Guard allows the user to read AtomicShared and keeps the underlying instance pinned to the thread.

Guard internally prevents the global epoch value from passing through the value announced by the current thread, thus keeping reachable instances in the thread from being garbage collected.

Implementations

impl Guard

pub fn new() -> Self

Creates a new Guard.

Panics

The maximum number of Guard instances in a thread is limited to u32::MAX; a thread panics when the number of Guard instances in the thread exceeds the limit.

Examples

use sdd::Guard;

let guard = Guard::new();

pub fn epoch(&self) -> Epoch

Returns the epoch in which the current thread lives.

This method can be used to check whether a retired memory region is potentially reachable or not. A chunk of memory retired in a witnessed Epoch can be deallocated after the thread has observed three new epochs. For instance, if the witnessed epoch value is 1 in the current thread where the global epoch value is 2, and an instance is retired in the same thread, the instance can be dropped when the thread witnesses 0 which is three epochs away from 1.

In other words, there can be potential readers of the memory chunk until the current thread witnesses the previous epoch. In the above example, the global epoch can be in 2 while the current thread has only witnessed 1, and therefore there can a reader of the memory chunk in another thread in epoch 2. The reader can survive until the global epoch reaches 0, because the thread being in 2 prevents the global epoch from reaching 0.

Examples

use sdd::{Guard, Owned};
use std::sync::atomic::AtomicBool;
use std::sync::atomic::Ordering::Relaxed;

static DROPPED: AtomicBool = AtomicBool::new(false);

struct D(&'static AtomicBool);

impl Drop for D {
    fn drop(&mut self) {
        self.0.store(true, Relaxed);
    }
}

let owned = Owned::new(D(&DROPPED));

let epoch_before = Guard::new().epoch();

drop(owned);
assert!(!DROPPED.load(Relaxed));

while Guard::new().epoch() == epoch_before {
    assert!(!DROPPED.load(Relaxed));
}

while Guard::new().epoch() == epoch_before.next() {
    assert!(!DROPPED.load(Relaxed));
}

while Guard::new().epoch() == epoch_before.next().next() {
    assert!(!DROPPED.load(Relaxed));
}

assert!(DROPPED.load(Relaxed));
assert_eq!(Guard::new().epoch(), epoch_before.prev());

pub fn accelerate(&self)

Forces the Guard to try to start a new epoch when it is dropped.

Examples

use sdd::Guard;

let guard = Guard::new();

let epoch = guard.epoch();
guard.accelerate();

drop(guard);

assert_ne!(epoch, Guard::new().epoch());

pub fn defer_execute<F: 'static + FnOnce()>(&self, f: F)

Executes the supplied closure at a later point of time.

It is guaranteed that the closure will be executed after every Guard at the moment when the method was invoked is dropped, however it is totally non-deterministic when exactly the closure will be executed.

Examples

use sdd::Guard;

let guard = Guard::new();
guard.defer_execute(|| println!("deferred"));

Trait Implementations

impl Default for Guard

fn default() -> Self

Returns the “default value” for a type.

impl Drop for Guard

fn drop(&mut self)

Executes the destructor for this type.

impl UnwindSafe for Guard