taktora-executor 0.1.2

//! Zero-allocation dispatch — verification for REQ_0060.
//!
//! Uses `taktora_bounded_alloc::CountingAllocator` as the test
//! binary's `#[global_allocator]` so every thread's allocations
//! (WaitSet thread + pool workers) are counted. A differential
//! measurement (`run_n(big) - run_n(small)`) isolates per-iteration
//! steady-state allocations from the one-time setup that happens
//! at the top of `dispatch_loop` (WaitSet construction, trigger
//! attachment, iceoryx2 lazy init).
//!
//! All cases live inside a single `#[test]` function so cargo's
//! parallel test runner cannot interleave another test's allocations
//! into the measurement window. The `CountingAllocator` is
//! process-wide; a per-test Mutex would not protect against the
//! harness's pre-body buffer allocations on a sibling worker thread.

#![allow(missing_docs)]
#![allow(clippy::doc_markdown, clippy::cast_possible_wrap)]

use core::time::Duration;
use taktora_bounded_alloc::CountingAllocator;
use taktora_executor::{ControlFlow, Executor, item, item_with_triggers};

#[global_allocator]
static ALLOC: CountingAllocator = CountingAllocator::new();

fn count_allocs<R>(f: impl FnOnce() -> R) -> (usize, R) {
    ALLOC.reset();
    ALLOC.set_tracking(true);
    let r = f();
    ALLOC.set_tracking(false);
    (ALLOC.alloc_count(), r)
}

// ── Trivial chain that performs no per-iteration work ──────────────────────

fn trivial_chain() -> Vec<Box<dyn taktora_executor::ExecutableItem>> {
    let head = item_with_triggers(
        |d| {
            d.interval(Duration::from_millis(1));
            Ok(())
        },
        |_| Ok(ControlFlow::Continue),
    );
    let mid = item(|_| Ok(ControlFlow::Continue));
    let tail = item(|_| Ok(ControlFlow::Continue));
    vec![Box::new(head), Box::new(mid), Box::new(tail)]
}

// ── Zero-allocation assertions ─────────────────────────────────────────────
//
// REQ_0060 prohibits heap allocations during **steady-state execution** —
// i.e. per-iteration of the dispatch loop. One-time setup performed by
// `dispatch_loop` (WaitSet construction, trigger attachment) is *not*
// steady-state, so we measure per-iteration allocation via a differential:
//
//   run_n(M) - run_n(N) = (M - N) * per_iter_alloc + 0
//
// for M > N and N large enough to absorb first-call lazy initialisation.

const ITERS_BIG: usize = 100;
const ITERS_SMALL: usize = 10;

/// Returns the average steady-state allocations per dispatch iteration.
fn per_iter_allocs(exec: &mut Executor) -> i64 {
    // Warm up to absorb any one-shot init that happens on first dispatch.
    exec.run_n(ITERS_SMALL).unwrap();
    let (a_small, ()) = count_allocs(|| exec.run_n(ITERS_SMALL).unwrap());
    let (a_big, ()) = count_allocs(|| exec.run_n(ITERS_BIG).unwrap());
    let diff = a_big as i64 - a_small as i64;
    let iters = (ITERS_BIG - ITERS_SMALL) as i64;
    // Round up so any fractional alloc per iter is detected.
    (diff + iters - 1) / iters
}

#[test]
fn dispatch_is_zero_allocation() {
    // Case 1: single-threaded chain.
    {
        let mut exec = Executor::builder().worker_threads(0).build().unwrap();
        exec.add_chain(trivial_chain()).unwrap();
        let per_iter = per_iter_allocs(&mut exec);
        assert_eq!(
            per_iter, 0,
            "REQ_0060 violated: ~{per_iter} steady-state allocations per iteration (single-threaded chain)"
        );
    }

    // Case 2: two-worker chain.
    {
        let mut exec = Executor::builder().worker_threads(2).build().unwrap();
        exec.add_chain(trivial_chain()).unwrap();
        let per_iter = per_iter_allocs(&mut exec);
        assert_eq!(
            per_iter, 0,
            "REQ_0060 violated: ~{per_iter} steady-state allocations per iteration (2 worker threads, chain)"
        );
    }

    // Case 3: diamond graph with two workers.
    {
        let mut exec = Executor::builder().worker_threads(2).build().unwrap();
        let mut g = exec.add_graph();
        let r = g.vertex(item_with_triggers(
            |d| {
                d.interval(Duration::from_millis(1));
                Ok(())
            },
            |_| Ok(ControlFlow::Continue),
        ));
        let l = g.vertex(item(|_| Ok(ControlFlow::Continue)));
        let rt = g.vertex(item(|_| Ok(ControlFlow::Continue)));
        let m = g.vertex(item(|_| Ok(ControlFlow::Continue)));
        g.edge(r, l).edge(r, rt).edge(l, m).edge(rt, m).root(r);
        g.build().unwrap();
        let per_iter = per_iter_allocs(&mut exec);
        assert_eq!(
            per_iter, 0,
            "REQ_0060 violated: ~{per_iter} steady-state allocations per iteration (graph diamond, 2 workers)"
        );
    }

    // Case 4: single-threaded single item.
    {
        let mut exec = Executor::builder().worker_threads(0).build().unwrap();
        let it = item_with_triggers(
            |d| {
                d.interval(Duration::from_millis(1));
                Ok(())
            },
            |_| Ok(ControlFlow::Continue),
        );
        exec.add(it).unwrap();
        let per_iter = per_iter_allocs(&mut exec);
        assert_eq!(
            per_iter, 0,
            "REQ_0060 violated: ~{per_iter} steady-state allocations per iteration (single-threaded, Single task)"
        );
    }

    // Case 5 — negative: harness must catch a deliberate per-iteration
    // alloc. If this case stops firing, the counting allocator has lost
    // visibility into worker-thread allocations and the other four
    // cases are meaningless.
    {
        let mut exec = Executor::builder().worker_threads(0).build().unwrap();
        let head = item_with_triggers(
            |d| {
                d.interval(Duration::from_millis(1));
                Ok(())
            },
            |_| {
                let v: Vec<u8> = vec![1, 2, 3];
                core::hint::black_box(&v);
                Ok(ControlFlow::Continue)
            },
        );
        exec.add(head).unwrap();
        exec.run_n(1).unwrap();
        let (allocs, ()) = count_allocs(|| exec.run_n(10).unwrap());
        assert!(
            allocs >= 10,
            "harness regression: counting allocator did not catch deliberate vec! allocations (saw {allocs})"
        );
    }
}