ftui_runtime/

retry.rs

// SPDX-License-Identifier: Apache-2.0
//! Retry policies and timeout-enforced task helpers.
//!
//! Provides [`RetryPolicy`] for configurable retry-with-backoff and
//! [`task_with_timeout`] / [`task_with_retry`] constructors that wrap
//! [`Cmd::Task`](crate::Cmd) with deterministic lifecycle guarantees.
//!
//! # Migration rationale
//!
//! Source frameworks often have retry/timeout baked into effect middleware.
//! These helpers give the migration code emitter explicit, testable primitives
//! to target instead of ad-hoc retry loops.
//!
//! # Determinism
//!
//! Backoff delays use fixed formulas (no jitter/randomness) so that
//! replay-based determinism tests can reproduce exact timing sequences.
//!
//! # Example
//!
//! ```
//! use ftui_runtime::retry::{RetryPolicy, BackoffStrategy};
//! use std::time::Duration;
//!
//! let policy = RetryPolicy::new(3, BackoffStrategy::Exponential {
//!     base_ms: 100,
//!     max_ms: 5000,
//! });
//!
//! assert_eq!(policy.delay(0), Duration::from_millis(100));
//! assert_eq!(policy.delay(1), Duration::from_millis(200));
//! assert_eq!(policy.delay(2), Duration::from_millis(400));
//! ```

#![forbid(unsafe_code)]

use crate::cancellation::{CancellationSource, CancellationToken};
use crate::program::{Cmd, TaskSpec};
use web_time::Duration;

const TASK_THREAD_JOIN_TIMEOUT: Duration = Duration::from_millis(250);
const TASK_THREAD_JOIN_POLL: Duration = Duration::from_millis(5);

fn join_task_thread(handle: std::thread::JoinHandle<()>) {
    let _ = handle.join();
}

fn join_task_thread_bounded(handle: std::thread::JoinHandle<()>, task_name: &'static str) {
    let start = web_time::Instant::now();
    while !handle.is_finished() {
        if start.elapsed() >= TASK_THREAD_JOIN_TIMEOUT {
            tracing::warn!(
                task = task_name,
                timeout_ms = TASK_THREAD_JOIN_TIMEOUT.as_millis() as u64,
                "Timed-out worker thread did not exit within the cancellation join timeout; detaching"
            );
            return;
        }
        std::thread::sleep(TASK_THREAD_JOIN_POLL);
    }
    join_task_thread(handle);
}

/// Backoff strategy for retry delays.
#[derive(Debug, Clone, PartialEq)]
#[cfg_attr(
    feature = "state-persistence",
    derive(serde::Serialize, serde::Deserialize)
)]
pub enum BackoffStrategy {
    /// Fixed delay between retries.
    Fixed {
        /// Delay in milliseconds.
        delay_ms: u64,
    },
    /// Exponential backoff: `base_ms * 2^attempt`, capped at `max_ms`.
    Exponential {
        /// Base delay in milliseconds.
        base_ms: u64,
        /// Maximum delay cap in milliseconds.
        max_ms: u64,
    },
    /// Linear backoff: `base_ms * (attempt + 1)`, capped at `max_ms`.
    Linear {
        /// Base delay in milliseconds.
        base_ms: u64,
        /// Maximum delay cap in milliseconds.
        max_ms: u64,
    },
}

/// A retry policy with configurable attempts and backoff.
#[derive(Debug, Clone, PartialEq)]
#[cfg_attr(
    feature = "state-persistence",
    derive(serde::Serialize, serde::Deserialize)
)]
pub struct RetryPolicy {
    /// Maximum number of retry attempts (0 = no retries, just the initial attempt).
    pub max_retries: u32,
    /// Backoff strategy between retries.
    pub backoff: BackoffStrategy,
}

impl RetryPolicy {
    /// Create a new retry policy.
    pub fn new(max_retries: u32, backoff: BackoffStrategy) -> Self {
        Self {
            max_retries,
            backoff,
        }
    }

    /// No retries — execute once.
    pub fn no_retry() -> Self {
        Self {
            max_retries: 0,
            backoff: BackoffStrategy::Fixed { delay_ms: 0 },
        }
    }

    /// Compute the delay before the given attempt (0-indexed).
    pub fn delay(&self, attempt: u32) -> Duration {
        match &self.backoff {
            BackoffStrategy::Fixed { delay_ms } => Duration::from_millis(*delay_ms),
            BackoffStrategy::Exponential { base_ms, max_ms } => {
                let multiplier = 1u64.checked_shl(attempt).unwrap_or(u64::MAX);
                let delay = base_ms.saturating_mul(multiplier);
                Duration::from_millis(delay.min(*max_ms))
            }
            BackoffStrategy::Linear { base_ms, max_ms } => {
                let delay = base_ms.saturating_mul(u64::from(attempt) + 1);
                Duration::from_millis(delay.min(*max_ms))
            }
        }
    }

    /// Total maximum delay across all retries (for timeout budgeting).
    pub fn total_max_delay(&self) -> Duration {
        let mut total = Duration::ZERO;
        for i in 0..self.max_retries {
            total += self.delay(i);
        }
        total
    }
}

/// Create a [`Cmd::Task`] that enforces a cooperative timeout.
///
/// The worker closure receives a [`CancellationToken`] and must honor it for
/// timely timeout teardown. On timeout, the runtime requests cancellation and
/// returns `on_timeout`; any late worker result is discarded.
pub fn task_with_timeout<M, F>(timeout: Duration, f: F, on_timeout: M) -> Cmd<M>
where
    M: Send + 'static,
    F: FnOnce(CancellationToken) -> M + Send + 'static,
{
    Cmd::task(move || {
        let source = CancellationSource::new();
        let token = source.token();
        let (tx, rx) = std::sync::mpsc::channel();
        let handle = std::thread::spawn(move || {
            let result = f(token);
            let _ = tx.send(result);
        });
        match rx.recv_timeout(timeout) {
            Ok(msg) => {
                join_task_thread(handle);
                msg
            }
            Err(_) => {
                source.cancel();
                join_task_thread_bounded(handle, "task_with_timeout");
                on_timeout
            }
        }
    })
}

/// Create a [`Cmd::Task`] with a named spec and cooperative timeout.
pub fn task_with_timeout_named<M, F>(
    name: impl Into<String>,
    timeout: Duration,
    f: F,
    on_timeout: M,
) -> Cmd<M>
where
    M: Send + 'static,
    F: FnOnce(CancellationToken) -> M + Send + 'static,
{
    Cmd::task_with_spec(TaskSpec::default().with_name(name), move || {
        let source = CancellationSource::new();
        let token = source.token();
        let (tx, rx) = std::sync::mpsc::channel();
        let handle = std::thread::spawn(move || {
            let result = f(token);
            let _ = tx.send(result);
        });
        match rx.recv_timeout(timeout) {
            Ok(msg) => {
                join_task_thread(handle);
                msg
            }
            Err(_) => {
                source.cancel();
                join_task_thread_bounded(handle, "task_with_timeout_named");
                on_timeout
            }
        }
    })
}

/// Create a [`Cmd::Task`] that retries on failure with the given policy.
///
/// The `f` closure returns `Result<M, String>`. On `Ok`, the message is
/// returned immediately. On `Err`, the task retries according to the policy,
/// sleeping between attempts. After all retries are exhausted, `on_exhaust`
/// is called with the last error to produce a fallback message.
pub fn task_with_retry<M, F>(policy: RetryPolicy, f: F, on_exhaust: fn(String) -> M) -> Cmd<M>
where
    M: Send + 'static,
    F: Fn() -> Result<M, String> + Send + 'static,
{
    Cmd::task(move || {
        let mut last_err = String::new();
        for attempt in 0..=policy.max_retries {
            match f() {
                Ok(msg) => return msg,
                Err(e) => {
                    last_err = e;
                    if attempt < policy.max_retries {
                        std::thread::sleep(policy.delay(attempt));
                    }
                }
            }
        }
        on_exhaust(last_err)
    })
}

/// Create a [`Cmd::Task`] with both retry and timeout.
///
/// Each individual attempt is bounded by `per_attempt_timeout`. The worker
/// receives a [`CancellationToken`] and must honor it for timely timeout
/// teardown. The total number of attempts is governed by the retry policy.
pub fn task_with_retry_and_timeout<M, F>(
    policy: RetryPolicy,
    per_attempt_timeout: Duration,
    f: F,
    on_exhaust: fn(String) -> M,
) -> Cmd<M>
where
    M: Send + 'static,
    F: Fn(CancellationToken) -> Result<M, String> + Send + Sync + 'static,
{
    Cmd::task(move || {
        let f = std::sync::Arc::new(f);
        let mut last_err = String::new();
        for attempt in 0..=policy.max_retries {
            let source = CancellationSource::new();
            let token = source.token();
            let (tx, rx) = std::sync::mpsc::channel();
            let f_clone = std::sync::Arc::clone(&f);
            let handle = std::thread::spawn(move || {
                let result = f_clone(token);
                let _ = tx.send(result);
            });
            match rx.recv_timeout(per_attempt_timeout) {
                Ok(Ok(msg)) => {
                    join_task_thread(handle);
                    return msg;
                }
                Ok(Err(e)) => {
                    join_task_thread(handle);
                    last_err = e;
                }
                Err(_) => {
                    source.cancel();
                    join_task_thread_bounded(handle, "task_with_retry_and_timeout");
                    last_err = "timeout".into();
                }
            }
            if attempt < policy.max_retries {
                std::thread::sleep(policy.delay(attempt));
            }
        }
        on_exhaust(last_err)
    })
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn fixed_backoff_constant_delay() {
        let policy = RetryPolicy::new(3, BackoffStrategy::Fixed { delay_ms: 100 });
        assert_eq!(policy.delay(0), Duration::from_millis(100));
        assert_eq!(policy.delay(1), Duration::from_millis(100));
        assert_eq!(policy.delay(2), Duration::from_millis(100));
    }

    #[test]
    fn exponential_backoff_doubles() {
        let policy = RetryPolicy::new(
            5,
            BackoffStrategy::Exponential {
                base_ms: 100,
                max_ms: 5000,
            },
        );
        assert_eq!(policy.delay(0), Duration::from_millis(100));
        assert_eq!(policy.delay(1), Duration::from_millis(200));
        assert_eq!(policy.delay(2), Duration::from_millis(400));
        assert_eq!(policy.delay(3), Duration::from_millis(800));
    }

    #[test]
    fn exponential_backoff_caps_at_max() {
        let policy = RetryPolicy::new(
            5,
            BackoffStrategy::Exponential {
                base_ms: 1000,
                max_ms: 3000,
            },
        );
        assert_eq!(policy.delay(0), Duration::from_millis(1000));
        assert_eq!(policy.delay(1), Duration::from_millis(2000));
        assert_eq!(policy.delay(2), Duration::from_millis(3000)); // capped
        assert_eq!(policy.delay(3), Duration::from_millis(3000)); // capped
    }

    #[test]
    fn linear_backoff_increments() {
        let policy = RetryPolicy::new(
            4,
            BackoffStrategy::Linear {
                base_ms: 100,
                max_ms: 500,
            },
        );
        assert_eq!(policy.delay(0), Duration::from_millis(100));
        assert_eq!(policy.delay(1), Duration::from_millis(200));
        assert_eq!(policy.delay(2), Duration::from_millis(300));
        assert_eq!(policy.delay(3), Duration::from_millis(400));
        assert_eq!(policy.delay(4), Duration::from_millis(500)); // capped
    }

    #[test]
    fn linear_backoff_caps_at_max() {
        let policy = RetryPolicy::new(
            4,
            BackoffStrategy::Linear {
                base_ms: 200,
                max_ms: 500,
            },
        );
        assert_eq!(policy.delay(2), Duration::from_millis(500)); // 200*3 = 600, capped at 500
    }

    #[test]
    fn no_retry_policy() {
        let policy = RetryPolicy::no_retry();
        assert_eq!(policy.max_retries, 0);
    }

    #[test]
    fn total_max_delay_fixed() {
        let policy = RetryPolicy::new(3, BackoffStrategy::Fixed { delay_ms: 100 });
        assert_eq!(policy.total_max_delay(), Duration::from_millis(300));
    }

    #[test]
    fn total_max_delay_exponential() {
        let policy = RetryPolicy::new(
            3,
            BackoffStrategy::Exponential {
                base_ms: 100,
                max_ms: 10000,
            },
        );
        // Delays: 100 + 200 + 400 = 700
        assert_eq!(policy.total_max_delay(), Duration::from_millis(700));
    }

    #[test]
    fn total_max_delay_zero_retries() {
        let policy = RetryPolicy::no_retry();
        assert_eq!(policy.total_max_delay(), Duration::ZERO);
    }

    #[test]
    fn exponential_backoff_overflow_saturates() {
        let policy = RetryPolicy::new(
            1,
            BackoffStrategy::Exponential {
                base_ms: u64::MAX / 2,
                max_ms: u64::MAX,
            },
        );
        // Should not panic on overflow
        let _ = policy.delay(30);
    }

    #[test]
    fn linear_backoff_overflow_saturates() {
        let policy = RetryPolicy::new(
            1,
            BackoffStrategy::Linear {
                base_ms: u64::MAX / 2,
                max_ms: u64::MAX,
            },
        );
        let _ = policy.delay(30);
    }

    #[test]
    fn retry_policy_clone_eq() {
        let policy = RetryPolicy::new(
            3,
            BackoffStrategy::Exponential {
                base_ms: 100,
                max_ms: 5000,
            },
        );
        let cloned = policy.clone();
        assert_eq!(policy, cloned);
    }

    #[test]
    fn task_with_retry_succeeds_first_try() {
        #[derive(Debug, PartialEq)]
        enum Msg {
            Ok(i32),
            Err(String),
        }

        let policy = RetryPolicy::new(3, BackoffStrategy::Fixed { delay_ms: 1 });
        let cmd = task_with_retry(policy, || Ok(Msg::Ok(42)), Msg::Err);

        // Verify it produces a Task variant
        assert_eq!(cmd.type_name(), "Task");
    }

    #[test]
    fn task_with_timeout_produces_task() {
        #[derive(Debug)]
        #[allow(dead_code)]
        enum Msg {
            Result(i32),
            Timeout,
        }

        let cmd = task_with_timeout(
            Duration::from_secs(1),
            |_token| Msg::Result(42),
            Msg::Timeout,
        );
        assert_eq!(cmd.type_name(), "Task");
    }

    #[test]
    fn task_with_timeout_requests_cancellation_on_timeout() {
        use std::sync::Arc;
        use std::sync::atomic::{AtomicBool, Ordering};

        #[derive(Debug, PartialEq)]
        enum Msg {
            Finished,
            Timeout,
        }

        let cancelled = Arc::new(AtomicBool::new(false));
        let worker_exited = Arc::new(AtomicBool::new(false));
        let cancelled_flag = Arc::clone(&cancelled);
        let exited_flag = Arc::clone(&worker_exited);

        let cmd = task_with_timeout(
            Duration::from_millis(10),
            move |token| {
                cancelled_flag.store(token.wait_timeout(Duration::from_secs(1)), Ordering::SeqCst);
                exited_flag.store(true, Ordering::SeqCst);
                Msg::Finished
            },
            Msg::Timeout,
        );

        let result = match cmd {
            Cmd::Task(_, task) => task(),
            other => panic!("expected Task, got {other:?}"),
        };

        assert_eq!(result, Msg::Timeout);
        std::thread::sleep(Duration::from_millis(50));
        assert!(cancelled.load(Ordering::SeqCst));
        assert!(worker_exited.load(Ordering::SeqCst));
    }

    #[test]
    fn task_with_retry_and_timeout_cancels_each_timed_out_attempt() {
        use std::sync::Arc;
        use std::sync::atomic::{AtomicUsize, Ordering};

        #[derive(Debug, PartialEq)]
        enum Msg {
            Exhausted(String),
        }

        fn on_exhaust(err: String) -> Msg {
            Msg::Exhausted(err)
        }

        let attempts = Arc::new(AtomicUsize::new(0));
        let cancelled = Arc::new(AtomicUsize::new(0));
        let attempts_flag = Arc::clone(&attempts);
        let cancelled_flag = Arc::clone(&cancelled);
        let policy = RetryPolicy::new(1, BackoffStrategy::Fixed { delay_ms: 0 });

        let cmd = task_with_retry_and_timeout(
            policy,
            Duration::from_millis(10),
            move |token| {
                attempts_flag.fetch_add(1, Ordering::SeqCst);
                if token.wait_timeout(Duration::from_secs(1)) {
                    cancelled_flag.fetch_add(1, Ordering::SeqCst);
                }
                Err("cancelled".to_owned())
            },
            on_exhaust,
        );

        let result = match cmd {
            Cmd::Task(_, task) => task(),
            other => panic!("expected Task, got {other:?}"),
        };

        assert_eq!(result, Msg::Exhausted("timeout".to_owned()));
        std::thread::sleep(Duration::from_millis(50));
        assert_eq!(attempts.load(Ordering::SeqCst), 2);
        assert_eq!(cancelled.load(Ordering::SeqCst), 2);
    }

    #[test]
    fn backoff_strategy_variants_debug() {
        let fixed = BackoffStrategy::Fixed { delay_ms: 100 };
        let exp = BackoffStrategy::Exponential {
            base_ms: 100,
            max_ms: 5000,
        };
        let linear = BackoffStrategy::Linear {
            base_ms: 100,
            max_ms: 500,
        };
        // Just verify Debug doesn't panic
        let _ = format!("{fixed:?}");
        let _ = format!("{exp:?}");
        let _ = format!("{linear:?}");
    }
}