cancel_this/

lib.rs

//! This crate provides a user-friendly way to implement cooperative
//! cancellation in Rust based on a wide range of criteria, including
//! *triggers*, *timers*, *OS signals* (Ctrl+C), *memory limit*, or the *Python
//! interpreter linked using PyO3*. It also provides liveness monitoring
//! of "cancellation aware" code.
//!
//! **Why not use `async` instead of cooperative cancellation?** In principle,
//! `async` was designed to solve a different problem, and that's executing IO-bound
//! tasks in a non-blocking fashion. It is not *really* designed for CPU-bound tasks.
//! Consequently, using `async` adds a lot of unnecessary overhead to your project
//! which `cancel_this` does not have (see also the *Performance* section in the project README).
//!
//! **Why not use [`stop-token`](https://crates.io/crates/stop-token),
//! [`CancellationToken`](https://docs.rs/tokio-util/latest/tokio_util/sync/struct.CancellationToken.html)
//! or other cooperative cancellation crates?** So far, all crates I have seen require you
//! to pass the cancellation token around and generally do not make it easy to
//! combine the effects of multiple tokens. In `cancel_this`, the goal was to
//! make cancellation dead simple: You register however many cancellation triggers
//! you want, each trigger is valid within a specific scope (and thread) and can be checked
//! by a macro anywhere in your code.
//!
//! ### Current features
//!
//! - Scoped cancellation using thread-local "cancellation triggers".
//! - Out-of-the-box support for triggers based on atomics and timers.
//! - With feature `ctrlc` enabled, support for cancellation using `SIGINT` signals.
//! - With feature `pyo3` enabled, support for cancellation using `Python::check_signals`.
//! - With feature `memory` enabled, support for cancellation based on memory consumption returned by `memory-stats`.
//! - With feature `liveness` enabled, you can register a per-thread handler invoked
//!   once the thread becomes unresponsive (i.e., cancellation is not checked periodically
//!   within the desired interval).
//! - Practically no overhead in cancellable code when cancellation is not actively used.
//! - Very small overhead for "atomic-based" cancellation triggers and PyO3 cancellation.
//! - All triggers and guards generate [`log`](https://crates.io/crates/log) messages
//!   (`trace` for normal operation, `warn` for issues where panic can be avoided).
//!
//!
//! ### Simple example
//!
//! A simple counter that is eventually canceled by a one-second timeout:
//!
//! ```rust
//! # use std::time::Duration;
//! # use cancel_this::{Cancellable, is_cancelled};
//! # let _ = env_logger::builder().is_test(true).try_init();
//! fn cancellable_counter(count: usize) -> Cancellable<()> {
//!     for _ in 0..count {
//!         is_cancelled!()?;
//!         std::thread::sleep(Duration::from_millis(10));
//!     }
//!     Ok(())
//! }
//!
//! let result: Cancellable<()> = cancel_this::on_timeout(Duration::from_secs(1), || {
//!     cancellable_counter(5)?;
//!     cancellable_counter(10)?;
//!     cancellable_counter(100)?;
//!     Ok(())
//! });
//!
//! assert!(result.is_err());
//! ```
//!
//! ## Complex example
//!
//! This example uses most of the features, including error conversion, never-cancel blocks,
//! and liveness monitoring.
//!
//! ```rust
//! # use std::time::Duration;
//! # use cancel_this::{Cancelled, is_cancelled, LivenessGuard};
//! # let _ = env_logger::builder().is_test(true).try_init();
//! enum ComputeError {
//!     Zero,
//!     Cancelled
//! }
//!
//! impl From<Cancelled> for ComputeError {
//!     fn from(value: Cancelled) -> Self {
//!        ComputeError::Cancelled
//!    }
//! }
//!
//!
//! fn compute(input: u32) -> Result<String, ComputeError> {
//!     if input == 0 {
//!         Err(ComputeError::Zero)
//!     } else {
//!         let mut total: u32 = 0;
//!         for _ in 0..input {
//!             total += input;
//!             is_cancelled!()?;
//!             std::thread::sleep(Duration::from_millis(10));
//!         }
//!         Ok(total.to_string())
//!     }
//! }
//!
//! let guard = LivenessGuard::new(Duration::from_secs(2), |is_alive| {
//!     eprintln!("Thread has not responded in the last two seconds.");
//! });
//!
//! let result: Result<String, ComputeError> = cancel_this::on_timeout(Duration::from_millis(200), || {
//!     let r1 = cancel_this::on_sigint(|| {
//!         // This operation can be canceled using Ctrl+C, but the timeout still applies.
//!         compute(5)
//!     })?;
//!
//!     assert_eq!(r1.as_str(), "25");
//!     // This will be canceled. Instead of using `?`, we check
//!     // that the operation actually got canceled.
//!     let r2 = compute(20);
//!     assert!(matches!(r2, Err(ComputeError::Cancelled)));
//!     // Even though the execution is now canceled, we can still execute code in
//!     // the "cancel-never" blocks.
//!     let r3 = cancel_this::never(|| compute(10))?;
//!     assert_eq!(r3.as_str(), "100");
//!     compute(10) // This should get immediately canceled.
//! });
//!
//! // The liveness monitoring is active while `guard` is in scope. Once `guard` is dropped here,
//! // the liveness monitoring is turned off as well.
//! ```
//!
//! ## Multi-threaded example
//!
//! Virtually all triggers and guards provided by `cancel_this` only apply to the current
//! thread. However, since triggers can be safely shared across threads, it is possible to
//! transfer them from one thread to another. Note that the transferred triggers also inherently
//! update the liveness guard of the original thread.
//!
//! ```rust
//! # use std::thread::JoinHandle;
//! # use std::time::Duration;
//! # use cancel_this::{is_cancelled, Cancellable, LivenessGuard};
//! # let _ = env_logger::builder().is_test(true).try_init();
//! let guard = LivenessGuard::new(Duration::from_millis(10), |is_alive| {
//!     // In this test, the liveness guard should never trigger, even though the original
//!     // thread goes to sleep for a long time, waiting to join with the spawned thread.
//!     assert!(is_alive);
//! });
//!
//! let result: Cancellable<u32> = cancel_this::on_timeout(Duration::from_millis(100), || {
//!     let active = cancel_this::active_triggers();
//!     let t1: JoinHandle<Cancellable<u32>> = std::thread::spawn(|| {
//!         cancel_this::on_trigger(active, || {
//!             let mut result = 0u32;
//!             // This cycle is eventually going to get canceled
//!             // by the timer which is "transferred" from the spawning thread.
//!             for i in 0..50 {
//!                 result += 1;
//!                 is_cancelled!()?;
//!                 std::thread::sleep(Duration::from_millis(5));
//!             }
//!             Ok(result)
//!         })
//!     });
//!     // Put the spawning thread to sleep until `t1` finishes.
//!     t1.join().unwrap()
//! });
//!
//! assert!(result.is_err());
//! ```
//!
//! Doing the same without transferring cancellation triggers will cause the spawning
//! thread to be registered as unresponsive and the compute thread to never actually get
//! canceled:
//!
//! ```rust
//! # use std::thread::JoinHandle;
//! # use std::time::Duration;
//! # use cancel_this::{is_cancelled, Cancellable, LivenessGuard};
//! # let _ = env_logger::builder().is_test(true).try_init();
//! let guard = LivenessGuard::new(Duration::from_millis(10), |is_alive| {
//!     // In this test, the liveness guard should never trigger, even though the original
//!     // thread goes to sleep for a long time, waiting to join with the spawned thread.
//!     assert!(!is_alive);
//! });
//!
//! let result: Cancellable<u32> = cancel_this::on_timeout(Duration::from_millis(100), || {
//!     let t1: JoinHandle<Cancellable<u32>> = std::thread::spawn(|| {
//!         let mut result = 0u32;
//!         // This cycle is never going to get canceled because we didn't transfer
//!         // the timeout trigger from the original thread.
//!         for i in 0..50 {
//!             result += 1;
//!             is_cancelled!()?;
//!             std::thread::sleep(Duration::from_millis(5));
//!         }
//!         Ok(result)
//!     });
//!     // Put the spawning thread to sleep until `t1` finishes.
//!     t1.join().unwrap()
//! });
//!
//! assert!(result.is_ok());
//! ```
//!
//! ## Cached triggers example
//!
//! If you need the absolute lowest overhead, you might want to sacrifice some of the
//! ergonomics provided by `cancel_this`. To reduce overhead, you can create a local copy
//! of the thread-local triggers the same way as in the multithreaded example and use it
//! directly with `is_cancelled`. This significantly reduces the overhead of each
//! cancellation check.
//!
//! ```rust
//! # use std::time::Duration;
//! # use cancel_this::{is_cancelled, Cancellable};
//! # let _ = env_logger::builder().is_test(true).try_init();
//! let result: Cancellable<u32> = cancel_this::on_timeout(Duration::from_millis(100), || {
//!     let cache = cancel_this::active_triggers();
//!     let mut result = 0u32;
//!     while true {
//!         // The overhead of this `is_cancelled` call is reduced because triggers
//!         // are cached in a local variable. However, the cache is only valid within
//!         // the scope where it was obtained.
//!         is_cancelled!(cache)?;
//!         result += 1;
//!     }
//!     Ok(result)
//! });
//! assert!(result.is_err());
//! ```
//!

/// Cancellation error type.
mod error;

/// Various types of triggers, including corresponding `when_*` helper functions.
mod triggers;

/// Implements a "liveness guard", which monitors the frequency with which cancellations are
/// checked, making sure the process
#[cfg(feature = "liveness")]
mod liveness;
#[cfg(feature = "liveness")]
pub use liveness::*;

#[cfg(not(feature = "liveness"))]
mod liveness {
    #[derive(Clone, Default)]
    pub(crate) struct LivenessInterceptor<R: crate::CancellationTrigger + Clone>(R);

    impl<R: crate::CancellationTrigger + Clone> LivenessInterceptor<R> {
        pub fn as_inner_mut(&mut self) -> &mut R {
            &mut self.0
        }

        pub fn as_inner(&self) -> &R {
            &self.0
        }
    }

    impl LivenessInterceptor<crate::CancelChain> {
        pub fn clone_and_flatten(&self) -> crate::triggers::DynamicCancellationTrigger {
            // If liveness monitoring is off, we can just use normal flattening.
            self.as_inner().clone_and_flatten()
        }
    }

    impl<R: crate::CancellationTrigger + Clone> crate::CancellationTrigger for LivenessInterceptor<R> {
        fn is_cancelled(&self) -> bool {
            // If liveness monitoring is off, we do nothing.
            self.0.is_cancelled()
        }

        fn type_name(&self) -> &'static str {
            self.0.type_name()
        }
    }
}

pub use error::*;
use liveness::LivenessInterceptor;
use std::cell::RefCell;
pub use triggers::*;

/// The "default" [`Cancelled`] cause, reported when the trigger type is unknown.
pub const UNKNOWN_CAUSE: &str = "UnknownCancellationTrigger";

thread_local! {
    /// The correct usage of this value lies in the fact that references to the `CancelChain`
    /// will never leak out of this crate (we can hand out copies to the downstream users).
    /// Within the crate, the triggers are either read when checking cancellation status or
    /// written when entering/leaving the scope. However, these two actions are never performed
    /// simultaneously.
    static TRIGGER: RefCell<LivenessInterceptor<CancelChain>> = RefCell::new(LivenessInterceptor::default());
}

/// Call this macro every time your code wants to check for cancellation. It returns
/// `Result<(), Cancelled>`, which can typically be propagated using the `?` operator.
#[macro_export]
macro_rules! is_cancelled {
    () => {
        $crate::check_local_cancellation()
    };
    ($handler:ident) => {
        $crate::check_cancellation(&$handler)
    };
}

/// Returns [`Cancelled`] if [`CancellationTrigger::is_cancelled`] of the given
/// `trigger` is true. In typical situations, you don't use this method directly,
/// but instead use the [`is_cancelled`] macro.
pub fn check_cancellation<TCancel: CancellationTrigger>(
    trigger: &TCancel,
) -> Result<(), Cancelled> {
    if trigger.is_cancelled() {
        Err(Cancelled::new(trigger.type_name()))
    } else {
        Ok(())
    }
}

/// Check if the current thread-local cancellation trigger is canceled. In typical situations,
/// you don't use this method directly, but instead use the [`is_cancelled`] macro.
///
/// To avoid a repeated borrow of the thread-local value in performance-sensitive applications,
/// you can use [`active_triggers`] to cache the value in a local variable.
pub fn check_local_cancellation() -> Result<(), Cancelled> {
    TRIGGER.with_borrow(check_cancellation)
}

/// Get a snapshot of the current thread-local cancellation trigger.
///
/// This value can be either used to initialize triggers in a new thread using [`on_trigger`],
/// or used directly as an argument to the [`is_cancelled`] macro to speed up cancellation checks.
pub fn active_triggers() -> DynamicCancellationTrigger {
    TRIGGER.with_borrow(|trigger| trigger.clone_and_flatten())
}

/// Run the `action` in a context where a cancellation can be signaled using the given `trigger`.
///
/// Once the action is completed, the trigger is deregistered and does not apply
/// to further code execution.
pub fn on_trigger<TResult, TError, TCancel, TAction>(
    trigger: TCancel,
    action: TAction,
) -> Result<TResult, TError>
where
    TCancel: CancellationTrigger + 'static,
    TAction: FnOnce() -> Result<TResult, TError>,
    TError: From<Cancelled>,
{
    TRIGGER.with_borrow_mut(|thread_trigger| thread_trigger.as_inner_mut().push(trigger));
    let result = action();
    TRIGGER.with_borrow_mut(|thread_trigger| thread_trigger.as_inner_mut().pop());
    result
}