commonware_runtime/

mocks.rs

//! Mock implementations of runtime primitives for testing.

use crate::{
    Blob, BlobVersion, BufMut, Clock, Error, Handle, IoBufs, IoBufsMut, Metrics, Name, ReadOptions,
    Spawner, Supervisor, WriteOptions,
    signal::Signal,
    telemetry::metrics::{Metric, Registered},
};
use bytes::{Bytes, BytesMut};
use commonware_macros::boxed;
use commonware_utils::{
    channel::{fallible::OneshotExt, oneshot},
    sync::Mutex,
};
use governor::clock::{Clock as GovernorClock, ReasonablyRealtime};
use rand::{TryCryptoRng, TryRng};
use std::{
    future::{Future, poll_fn},
    mem,
    sync::Arc,
    task::Poll,
};

cfg_if::cfg_if! {
    if #[cfg(any(test, feature = "test-utils"))] {
        use crate::{IoBufMut, utils::reschedule};
        use futures::poll;
        use std::sync::atomic::{AtomicUsize, Ordering};
    }
}

/// In-memory storage with exclusive logical opens and durable snapshot inspection.
#[cfg(any(test, feature = "test-utils"))]
#[derive(Clone, Default)]
pub struct Storage {
    inner: crate::storage::memory::Storage,
    opens: Arc<crate::storage::memory::open::Opens>,
}

#[cfg(any(test, feature = "test-utils"))]
impl Storage {
    /// Create an empty memory storage backend.
    pub fn new() -> Self {
        Self::default()
    }

    /// Compute a CRC32C digest of all durable blob contents.
    pub fn audit(&self) -> [u8; 4] {
        self.inner.audit()
    }

    /// Return a copy of a blob's durable raw contents without interpreting its container header.
    pub fn raw_blob(&self, partition: &str, name: &[u8]) -> Option<Vec<u8>> {
        self.inner.raw_blob(partition, name)
    }

    /// Return a copy of a blob's durable logical contents, or `None` when the blob is missing or
    /// its container header does not resolve.
    pub fn logical_blob(&self, partition: &str, name: &[u8]) -> Option<Vec<u8>> {
        self.inner.logical_blob(partition, name)
    }

    /// Install durable raw contents without validating the blob's container header.
    ///
    /// This retires the prior incarnation and permits a new open while old handles remain alive.
    pub fn set_raw_blob(&self, partition: &str, name: &[u8], content: Vec<u8>) {
        self.opens
            .replace(partition, Some(name), || {
                self.inner.set_raw_blob(partition, name, content);
                Ok(())
            })
            .expect("installing a raw memory image cannot fail");
    }
}

#[cfg(any(test, feature = "test-utils"))]
impl crate::Storage for Storage {
    type Blob = crate::storage::memory::open::Blob<crate::storage::memory::Blob>;

    async fn open_versioned(
        &self,
        partition: &str,
        name: &[u8],
        versions: std::ops::RangeInclusive<BlobVersion>,
    ) -> Result<(Self::Blob, u64, BlobVersion), Error> {
        let opened = self.opens.open(
            partition,
            name,
            self.inner.open_versioned(partition, name, versions),
        )?;
        Ok(opened.finish())
    }

    async fn remove(&self, partition: &str, name: Option<&[u8]>) -> Result<(), Error> {
        self.opens
            .remove(partition, name, self.inner.remove(partition, name))
    }

    async fn scan(&self, partition: &str) -> Result<Vec<Vec<u8>>, Error> {
        self.inner.scan(partition).await
    }
}

/// Default buffer size (64 KB). Controls both how much data the stream
/// pulls per recv and the backpressure threshold for send.
const DEFAULT_BUFFER_SIZE: usize = 64 * 1024;

/// A mock channel struct that is used internally by Sink and Stream.
pub struct Channel {
    /// Stores the bytes sent by the sink that are not yet read by the stream.
    buffer: BytesMut,

    /// If the stream is waiting to read bytes, the waiter stores the number of
    /// bytes that the stream is waiting for, as well as the oneshot sender that
    /// the sink uses to send the bytes to the stream directly.
    waiter: Option<(usize, oneshot::Sender<Bytes>)>,

    /// Target buffer size, used to bound both the stream's local buffer
    /// and the shared buffer (backpressure threshold).
    buffer_size: usize,

    /// If the sink is blocked waiting for the buffer to drain, this holds
    /// the oneshot sender that the stream uses to wake the sink.
    drain_waiter: Option<oneshot::Sender<()>>,

    /// Tracks whether the sink is still alive and able to send messages.
    sink_alive: bool,

    /// Tracks whether the stream is still alive and able to receive messages.
    stream_alive: bool,
}

impl Channel {
    /// Returns an async-safe Sink/Stream pair with default buffer size.
    pub fn init() -> (Sink, Stream) {
        Self::init_with_buffer_size(DEFAULT_BUFFER_SIZE)
    }

    /// Returns an async-safe Sink/Stream pair with the specified buffer size.
    pub fn init_with_buffer_size(buffer_size: usize) -> (Sink, Stream) {
        let channel = Arc::new(Mutex::new(Self {
            buffer: BytesMut::new(),
            waiter: None,
            buffer_size,
            drain_waiter: None,
            sink_alive: true,
            stream_alive: true,
        }));
        (
            Sink {
                channel: channel.clone(),
                state: SinkState::Open,
            },
            Stream {
                channel,
                buffer: BytesMut::new(),
                poisoned: false,
            },
        )
    }

    /// Restores bytes that were detached from the front of the shared buffer.
    fn restore_front(&mut self, data: Bytes) {
        if data.is_empty() {
            return;
        }

        let mut restored = BytesMut::with_capacity(data.len() + self.buffer.len());
        restored.extend_from_slice(&data);
        restored.extend_from_slice(&self.buffer);
        self.buffer = restored;
    }

    /// Marks the sink as closed and wakes any waiter.
    fn close_sink(&mut self) {
        self.sink_alive = false;

        // If there is a waiter, resolve it by dropping the oneshot sender.
        self.waiter.take();
    }
}

struct RecvWaiterGuard {
    channel: Arc<Mutex<Channel>>,
    active: bool,
}

impl RecvWaiterGuard {
    const fn new(channel: Arc<Mutex<Channel>>) -> Self {
        Self {
            channel,
            active: true,
        }
    }

    const fn disarm(&mut self) {
        self.active = false;
    }
}

impl Drop for RecvWaiterGuard {
    fn drop(&mut self) {
        if !self.active {
            return;
        }

        self.channel.lock().waiter.take();
    }
}

/// A mock sink that implements the Sink trait.
pub struct Sink {
    channel: Arc<Mutex<Channel>>,
    state: SinkState,
}

/// Lifecycle state for the mock sink half.
enum SinkState {
    /// Sends may be attempted.
    Open,
    /// A send is currently in progress.
    Sending,
    /// The sink has been closed.
    Closed,
}

impl Sink {
    fn close(&mut self) {
        if matches!(self.state, SinkState::Closed) {
            return;
        }
        self.channel.lock().close_sink();
        self.state = SinkState::Closed;
    }
}

impl crate::Sink for Sink {
    async fn send(&mut self, bufs: impl Into<IoBufs> + Send) -> Result<(), Error> {
        match self.state {
            SinkState::Open => {}
            SinkState::Sending => {
                self.close();
                return Err(Error::Closed);
            }
            SinkState::Closed => return Err(Error::Closed),
        }

        let drain_recv = {
            let mut channel = self.channel.lock();

            // If the receiver is dead, we cannot send any more messages.
            if !channel.stream_alive {
                channel.close_sink();
                self.state = SinkState::Closed;
                return Err(Error::SendFailed);
            }

            channel.buffer.put(bufs.into());

            // If there is a waiter and the buffer is large enough,
            // resolve the waiter (while clearing the waiter field).
            if channel
                .waiter
                .as_ref()
                .is_some_and(|(requested, _)| *requested <= channel.buffer.len())
            {
                // Send up to buffer_size bytes (but at least requested amount)
                let (requested, os_send) = channel.waiter.take().unwrap();
                let send_amount = channel.buffer.len().min(requested.max(channel.buffer_size));
                let data = channel.buffer.split_to(send_amount).freeze();

                // A canceled recv should behave like a buffered transport:
                // preserve the bytes and allow a subsequent recv to consume them.
                if let Err(data) = os_send.send(data) {
                    channel.restore_front(data);
                    if !channel.stream_alive {
                        channel.close_sink();
                        self.state = SinkState::Closed;
                        return Err(Error::SendFailed);
                    }
                }
            }

            // If the buffer exceeds the write limit, block until the
            // receiver drains enough data.
            if channel.buffer.len() > channel.buffer_size {
                assert!(channel.drain_waiter.is_none());
                let (os_send, os_recv) = oneshot::channel();
                channel.drain_waiter = Some(os_send);
                os_recv
            } else {
                return Ok(());
            }
        };

        // Mark the sink as sending before awaiting so cancellation can be
        // detected by the next send.
        self.state = SinkState::Sending;

        // Wait for the receiver to drain the buffer.
        match drain_recv.await {
            Ok(()) => {
                self.state = SinkState::Open;
                Ok(())
            }
            Err(_) => {
                self.close();
                Err(Error::SendFailed)
            }
        }
    }
}

impl Drop for Sink {
    fn drop(&mut self) {
        self.close();
    }
}

/// A mock stream that implements the Stream trait.
pub struct Stream {
    channel: Arc<Mutex<Channel>>,
    /// Local buffer for data that has been received but not yet consumed.
    buffer: BytesMut,
    poisoned: bool,
}

impl crate::Stream for Stream {
    async fn recv(&mut self, len: usize) -> Result<IoBufs, Error> {
        if self.poisoned {
            return Err(Error::Closed);
        }

        let os_recv = {
            let mut channel = self.channel.lock();

            // Pull data from channel buffer into local buffer.
            let target = len.max(channel.buffer_size);
            let pull_amount = channel
                .buffer
                .len()
                .min(target.saturating_sub(self.buffer.len()));
            if pull_amount > 0 {
                let data = channel.buffer.split_to(pull_amount);
                self.buffer.extend_from_slice(&data);

                // Wake a blocked sender if the buffer drained below the limit.
                if channel.buffer.len() <= channel.buffer_size
                    && let Some(sender) = channel.drain_waiter.take()
                {
                    sender.send_lossy(());
                }
            }

            // If we have enough, return immediately.
            if self.buffer.len() >= len {
                return Ok(IoBufs::from(self.buffer.split_to(len).freeze()));
            }

            // If the sink is dead, we cannot receive any more messages.
            if !channel.sink_alive {
                self.poisoned = true;
                return Err(Error::RecvFailed);
            }

            // Set up waiter for remaining amount.
            let remaining = len - self.buffer.len();
            assert!(channel.waiter.is_none());
            let (os_send, os_recv) = oneshot::channel();
            channel.waiter = Some((remaining, os_send));
            os_recv
        };

        let mut waiter_guard = RecvWaiterGuard::new(self.channel.clone());

        // Pre-poison so that cancellation  leaves the stream permanently closed.
        self.poisoned = true;

        // Wait for the waiter to be resolved.
        let data = match os_recv.await {
            Ok(data) => {
                waiter_guard.disarm();
                self.poisoned = false;
                data
            }
            Err(_) => {
                waiter_guard.disarm();
                return Err(Error::RecvFailed);
            }
        };
        self.buffer.extend_from_slice(&data);

        assert!(self.buffer.len() >= len);
        Ok(IoBufs::from(self.buffer.split_to(len).freeze()))
    }

    fn peek(&self, max_len: usize) -> &[u8] {
        let len = max_len.min(self.buffer.len());
        &self.buffer[..len]
    }
}

impl Drop for Stream {
    fn drop(&mut self) {
        let mut channel = self.channel.lock();
        channel.stream_alive = false;

        // Wake a blocked sender so it can observe the closed stream.
        channel.drain_waiter.take();
    }
}

/// A sync deferred by a [DelayedSyncBlob], held open until explicitly completed.
pub struct DeferredSync {
    /// Completes the sync with the provided result (success runs the inner blob's sync).
    pub release: oneshot::Sender<Result<(), Error>>,

    /// Resolves once the deferred sync's handle begins waiting on `release`.
    pub blocked: oneshot::Receiver<()>,
}

/// Coordinates durability operations for a [DelayedSyncContext] or [DelayedSyncBlob].
///
/// Every started sync parks in a deferred queue (in start order) until a test
/// releases it or [Self::unblock] runs. [Self::arm] additionally installs a one-shot gate that blocks
/// the next durability operation and counts operations from that point on
/// ([Self::calls]). The gate is pushed onto the deferred queue when [Self::arm]
/// is called, before any operation reaches it.
#[derive(Clone, Default)]
pub struct PendingSyncs {
    state: Arc<Mutex<State>>,
}

/// State shared by all clones of a [PendingSyncs].
#[derive(Default)]
struct State {
    /// Deferred syncs in start order.
    syncs: Vec<DeferredSync>,
    /// One-shot gate blocking the next durability operation (see [PendingSyncs::arm]).
    gate: SyncGateState,
    /// Sticky: stop parking started syncs (see [PendingSyncs::unblock]).
    unblocked: bool,
    /// Sticky: syncs resolve to an injected error (see [PendingSyncs::arm_fail]).
    fail: bool,
    /// Started syncs issued.
    starts: usize,
    /// Started syncs whose completion futures have begun executing.
    entered: usize,
    /// Started syncs that completed durably.
    completions: usize,
}

impl State {
    /// Creates a waiter parked in the deferred queue.
    fn defer(&mut self) -> SyncWaiter {
        let (release, release_rx) = oneshot::channel();
        let (entered, blocked) = oneshot::channel();
        self.syncs.push(DeferredSync { release, blocked });
        SyncWaiter {
            entered,
            release: release_rx,
        }
    }

    /// Records a durability operation if the gate is armed, returning the
    /// one-shot gate waiter if it has not been consumed yet.
    const fn observe(&mut self) -> Option<SyncWaiter> {
        if !self.gate.tracking {
            return None;
        }
        self.gate.calls += 1;
        self.gate.waiter.take()
    }

    /// Parks a new deferred sync, unless [PendingSyncs::unblock] already ran.
    fn park(&mut self) -> Option<SyncWaiter> {
        if self.unblocked {
            return None;
        }
        Some(self.defer())
    }
}

/// Forwards [Supervisor], [Clock], [GovernorClock], [ReasonablyRealtime],
/// [Metrics], [TryRng], and [TryCryptoRng] to the wrapped context for test
/// context wrappers with one extra field (named by the second argument).
macro_rules! forward_context {
    ($wrapper:ident, $field:ident) => {
        impl<E: Supervisor> Supervisor for $wrapper<E> {
            fn name(&self) -> Name {
                self.inner.name()
            }

            fn child(&self, label: &'static str) -> Self {
                Self {
                    inner: self.inner.child(label),
                    $field: self.$field.clone(),
                }
            }

            fn with_attribute(self, key: &'static str, value: impl std::fmt::Display) -> Self {
                Self {
                    inner: self.inner.with_attribute(key, value),
                    $field: self.$field,
                }
            }
        }

        impl<E: Clock> Clock for $wrapper<E> {
            fn current(&self) -> std::time::SystemTime {
                self.inner.current()
            }

            fn sleep(
                &self,
                duration: std::time::Duration,
            ) -> impl Future<Output = ()> + Send + 'static + use<E> {
                self.inner.sleep(duration)
            }

            fn sleep_until(
                &self,
                deadline: std::time::SystemTime,
            ) -> impl Future<Output = ()> + Send + 'static + use<E> {
                self.inner.sleep_until(deadline)
            }
        }

        impl<E: Clock> GovernorClock for $wrapper<E> {
            type Instant = std::time::SystemTime;

            fn now(&self) -> Self::Instant {
                self.current()
            }
        }

        impl<E: Clock> ReasonablyRealtime for $wrapper<E> {}

        impl<E: Metrics> Metrics for $wrapper<E> {
            fn register<N: Into<String>, H: Into<String>, M: Metric>(
                &self,
                name: N,
                help: H,
                metric: M,
            ) -> Registered<M> {
                self.inner.register(name, help, metric)
            }

            fn encode(&self) -> String {
                self.inner.encode()
            }
        }

        impl<E: TryRng> TryRng for $wrapper<E> {
            type Error = E::Error;

            fn try_next_u32(&mut self) -> Result<u32, Self::Error> {
                self.inner.try_next_u32()
            }

            fn try_next_u64(&mut self) -> Result<u64, Self::Error> {
                self.inner.try_next_u64()
            }

            fn try_fill_bytes(&mut self, dest: &mut [u8]) -> Result<(), Self::Error> {
                self.inner.try_fill_bytes(dest)
            }
        }

        impl<E: TryCryptoRng> TryCryptoRng for $wrapper<E> {}
    };
}

/// Snapshot of the options observed by a [RecordingContext] or [RecordingBlob].
#[cfg(any(test, feature = "test-utils"))]
#[derive(Clone, Debug, Default, Eq, PartialEq)]
pub struct RecordingSnapshot {
    /// Options supplied to read operations, in call order, with one entry per range of a batched
    /// read.
    pub reads: Vec<ReadOptions>,
    /// Range count of each batched read, in call order.
    pub batches: Vec<usize>,
    /// Options supplied to write operations, in call order.
    pub writes: Vec<WriteOptions>,
}

/// Shared observations produced by recording storage wrappers.
#[cfg(any(test, feature = "test-utils"))]
#[derive(Clone, Default)]
pub struct Recordings {
    state: Arc<Mutex<RecordingSnapshot>>,
}

#[cfg(any(test, feature = "test-utils"))]
impl Recordings {
    /// Return a snapshot of all observations recorded so far.
    pub fn snapshot(&self) -> RecordingSnapshot {
        self.state.lock().clone()
    }

    /// Remove all recorded observations.
    pub fn clear(&self) {
        *self.state.lock() = RecordingSnapshot::default();
    }

    fn read(&self, options: ReadOptions) {
        self.state.lock().reads.push(options);
    }

    fn batch(&self, ranges: usize, options: ReadOptions) {
        let mut state = self.state.lock();
        state.reads.extend(std::iter::repeat_n(options, ranges));
        state.batches.push(ranges);
    }

    fn write(&self, options: WriteOptions) {
        self.state.lock().writes.push(options);
    }
}

/// Context wrapper that records options supplied to every opened blob.
#[cfg(any(test, feature = "test-utils"))]
#[derive(Clone)]
pub struct RecordingContext<E> {
    /// Wrapped context.
    pub inner: E,
    /// Observations shared by this context and all blobs opened through it.
    pub recordings: Recordings,
}

#[cfg(any(test, feature = "test-utils"))]
impl<E> RecordingContext<E> {
    /// Wrap `inner` and return both the context and its shared observations.
    pub fn new(inner: E) -> (Self, Recordings) {
        let recordings = Recordings::default();
        (
            Self {
                inner,
                recordings: recordings.clone(),
            },
            recordings,
        )
    }
}

#[cfg(any(test, feature = "test-utils"))]
forward_context!(RecordingContext, recordings);

#[cfg(any(test, feature = "test-utils"))]
impl<E: Spawner> Spawner for RecordingContext<E> {
    fn shared(mut self, blocking: bool) -> Self {
        self.inner = self.inner.shared(blocking);
        self
    }

    fn dedicated(mut self) -> Self {
        self.inner = self.inner.dedicated();
        self
    }

    fn spawn<F, Fut, T>(self, f: F) -> Handle<T>
    where
        F: FnOnce(Self) -> Fut + Send + 'static,
        Fut: Future<Output = T> + Send + 'static,
        T: Send + 'static,
    {
        let recordings = self.recordings;
        self.inner.spawn(move |inner| f(Self { inner, recordings }))
    }

    async fn stop(self, value: i32, timeout: Option<std::time::Duration>) -> Result<(), Error> {
        self.inner.stop(value, timeout).await
    }

    fn stopped(&self) -> Signal {
        self.inner.stopped()
    }
}

#[cfg(any(test, feature = "test-utils"))]
impl<E: crate::Storage> crate::Storage for RecordingContext<E> {
    type Blob = RecordingBlob<E::Blob>;

    async fn open_versioned(
        &self,
        partition: &str,
        name: &[u8],
        versions: std::ops::RangeInclusive<BlobVersion>,
    ) -> Result<(Self::Blob, u64, BlobVersion), Error> {
        let (inner, len, version) = self.inner.open_versioned(partition, name, versions).await?;
        Ok((
            RecordingBlob {
                inner,
                recordings: self.recordings.clone(),
            },
            len,
            version,
        ))
    }

    async fn remove(&self, partition: &str, name: Option<&[u8]>) -> Result<(), Error> {
        self.inner.remove(partition, name).await
    }

    async fn scan(&self, partition: &str) -> Result<Vec<Vec<u8>>, Error> {
        self.inner.scan(partition).await
    }
}

/// Blob wrapper that records read and write options and the range count of each batched read,
/// then delegates each operation.
#[cfg(any(test, feature = "test-utils"))]
pub struct RecordingBlob<B> {
    inner: B,
    recordings: Recordings,
}

#[cfg(any(test, feature = "test-utils"))]
impl<B: Blob> Blob for RecordingBlob<B> {
    async fn read_at_buf(
        &self,
        offset: u64,
        len: usize,
        bufs: impl Into<IoBufsMut> + Send,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.recordings.read(options);
        self.inner.read_at_buf(offset, len, bufs, options).await
    }

    async fn read_at(
        &self,
        offset: u64,
        len: usize,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.recordings.read(options);
        self.inner.read_at(offset, len, options).await
    }

    fn read_many(
        &self,
        ranges: &[(u64, usize)],
        options: ReadOptions,
    ) -> impl futures::Stream<Item = Result<(usize, IoBufsMut), Error>> + Send {
        self.recordings.batch(ranges.len(), options);
        self.inner.read_many(ranges, options)
    }

    async fn write_at(
        &self,
        offset: u64,
        bufs: impl Into<IoBufs> + Send,
        options: WriteOptions,
    ) -> Result<(), Error> {
        self.recordings.write(options);
        self.inner.write_at(offset, bufs, options).await
    }

    async fn resize(&self, len: u64) -> Result<(), Error> {
        self.inner.resize(len).await
    }

    async fn sync(&self) -> Result<(), Error> {
        self.inner.sync().await
    }

    async fn start_sync(&self) -> Handle<()> {
        self.inner.start_sync().await
    }
}

/// Blob wrapper that yields after the first backend poll of each read, allowing a test to
/// hand the unresolved read future to another task even when the backend completes immediately.
#[cfg(any(test, feature = "test-utils"))]
pub struct MigratingReadBlob<B> {
    /// Wrapped blob.
    inner: B,
    /// Whether each read must remain pending after its first backend poll.
    require_pending: bool,
    /// Reads started through this blob.
    reads: AtomicUsize,
}

#[cfg(any(test, feature = "test-utils"))]
impl<B> MigratingReadBlob<B> {
    /// Wrap `inner`, optionally requiring the first backend poll of each read to return pending.
    /// Set `require_pending` when a test must exercise registered backend I/O.
    pub const fn new(inner: B, require_pending: bool) -> Self {
        Self {
            inner,
            require_pending,
            reads: AtomicUsize::new(0),
        }
    }

    /// Number of reads started through this blob.
    pub fn reads(&self) -> usize {
        self.reads.load(Ordering::Relaxed)
    }
}

#[cfg(any(test, feature = "test-utils"))]
impl<B: Blob> Blob for MigratingReadBlob<B> {
    async fn read_at(
        &self,
        offset: u64,
        len: usize,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.read_at_buf(offset, len, IoBufMut::with_capacity(len), options)
            .await
    }

    async fn read_at_buf(
        &self,
        offset: u64,
        len: usize,
        bufs: impl Into<IoBufsMut> + Send,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.reads.fetch_add(1, Ordering::Relaxed);
        let mut read = Box::pin(self.inner.read_at_buf(offset, len, bufs, options));

        // Capture exactly one backend poll, then yield before exposing its result. Requiring a
        // pending result lets a test establish that a task handoff carries unresolved backend I/O.
        let first_poll = poll!(&mut read);
        if self.require_pending {
            assert!(
                first_poll.is_pending(),
                "blob read completed before registering pending I/O"
            );
        }
        reschedule().await;

        match first_poll {
            Poll::Ready(result) => result,
            Poll::Pending => read.await,
        }
    }

    async fn write_at(
        &self,
        offset: u64,
        bufs: impl Into<IoBufs> + Send,
        options: WriteOptions,
    ) -> Result<(), Error> {
        self.inner.write_at(offset, bufs, options).await
    }

    async fn resize(&self, len: u64) -> Result<(), Error> {
        self.inner.resize(len).await
    }

    async fn sync(&self) -> Result<(), Error> {
        self.inner.sync().await
    }

    async fn start_sync(&self) -> Handle<()> {
        self.inner.start_sync().await
    }
}

/// Context wrapper whose blobs defer [Blob::start_sync] and can gate blocking syncs in tests.
#[derive(Clone)]
pub struct DelayedSyncContext<E> {
    pub inner: E,
    pub pending: PendingSyncs,
}

forward_context!(DelayedSyncContext, pending);

impl<E: Spawner> Spawner for DelayedSyncContext<E> {
    fn shared(mut self, blocking: bool) -> Self {
        self.inner = self.inner.shared(blocking);
        self
    }

    fn dedicated(mut self) -> Self {
        self.inner = self.inner.dedicated();
        self
    }

    fn spawn<F, Fut, T>(self, f: F) -> Handle<T>
    where
        F: FnOnce(Self) -> Fut + Send + 'static,
        Fut: Future<Output = T> + Send + 'static,
        T: Send + 'static,
    {
        let pending = self.pending;
        self.inner.spawn(move |inner| f(Self { inner, pending }))
    }

    async fn stop(self, value: i32, timeout: Option<std::time::Duration>) -> Result<(), Error> {
        self.inner.stop(value, timeout).await
    }

    fn stopped(&self) -> Signal {
        self.inner.stopped()
    }
}

impl<E: crate::Storage> crate::Storage for DelayedSyncContext<E> {
    type Blob = DelayedSyncBlob<E::Blob>;

    async fn open_versioned(
        &self,
        partition: &str,
        name: &[u8],
        versions: std::ops::RangeInclusive<BlobVersion>,
    ) -> Result<(Self::Blob, u64, BlobVersion), Error> {
        let (inner, len, version) = self.inner.open_versioned(partition, name, versions).await?;
        Ok((
            DelayedSyncBlob {
                inner: Arc::new(inner),
                pending: self.pending.clone(),
            },
            len,
            version,
        ))
    }

    async fn remove(&self, partition: &str, name: Option<&[u8]>) -> Result<(), Error> {
        self.inner.remove(partition, name).await
    }

    async fn scan(&self, partition: &str) -> Result<Vec<Vec<u8>>, Error> {
        self.inner.scan(partition).await
    }
}

/// Blob wrapper that parks each started sync and supports one-shot blocking sync tracking.
///
/// A started sync keeps the wrapped blob, and with it the logical open, alive until its handle
/// completes or drops.
pub struct DelayedSyncBlob<B> {
    inner: Arc<B>,
    pending: PendingSyncs,
}

impl<B> DelayedSyncBlob<B> {
    /// Wrap `inner`, returning the blob and the list its deferred syncs are pushed onto.
    pub fn new(inner: B) -> (Self, PendingSyncs) {
        let pending = PendingSyncs::default();
        (
            Self {
                inner: Arc::new(inner),
                pending: pending.clone(),
            },
            pending,
        )
    }
}

impl<B: Blob> Blob for DelayedSyncBlob<B> {
    async fn read_at_buf(
        &self,
        offset: u64,
        len: usize,
        bufs: impl Into<IoBufsMut> + Send,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.inner.read_at_buf(offset, len, bufs, options).await
    }

    async fn read_at(
        &self,
        offset: u64,
        len: usize,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.inner.read_at(offset, len, options).await
    }

    async fn write_at(
        &self,
        offset: u64,
        bufs: impl Into<IoBufs> + Send,
        options: WriteOptions,
    ) -> Result<(), Error> {
        if !options.contains(WriteOptions::SYNC) || !self.pending.tracking() {
            return self.inner.write_at(offset, bufs, options).await;
        }
        self.inner
            .write_at(offset, bufs, options.without(WriteOptions::SYNC))
            .await?;
        self.sync().await
    }

    async fn resize(&self, len: u64) -> Result<(), Error> {
        self.inner.resize(len).await
    }

    async fn sync(&self) -> Result<(), Error> {
        self.pending.wait().await?;
        self.inner.sync().await
    }

    async fn start_sync(&self) -> Handle<()> {
        let pending = self.pending.clone();
        let inner = self.inner.clone();
        let waiter = {
            let mut state = pending.state.lock();
            state.starts += 1;
            // An armed gate takes precedence over parking.
            state.observe().or_else(|| state.park())
        };
        Handle::from_future(async move {
            let fail = {
                let mut state = pending.state.lock();
                state.entered += 1;
                state.fail
            };
            match waiter {
                Some(waiter) => waiter.wait().await?,
                None if fail => return Err(injected_sync_failure()),
                None => {}
            }
            inner.sync().await?;
            pending.state.lock().completions += 1;
            Ok(())
        })
    }
}

/// Take the oldest pending sync, panicking if none was started.
pub fn next_pending_sync(pending: &PendingSyncs) -> DeferredSync {
    let mut pending = pending.lock();
    assert!(!pending.is_empty(), "no pending sync was started");
    pending.remove(0)
}

/// Complete the oldest `count` pending syncs successfully.
pub fn release_next_pending_syncs(pending: &PendingSyncs, count: usize) {
    let syncs = {
        let mut pending = pending.lock();
        assert!(
            pending.len() >= count,
            "not enough pending syncs: have {}, need {count}",
            pending.len()
        );
        pending.drain(..count).collect::<Vec<_>>()
    };
    for sync in syncs {
        let _ = sync.release.send(Ok(()));
    }
}

/// Complete all pending syncs successfully.
pub fn release_pending_syncs(pending: &PendingSyncs) {
    for sync in mem::take(&mut *pending.lock()) {
        let _ = sync.release.send(Ok(()));
    }
}

/// Drive `fut` to completion, releasing any parked syncs each time it stalls.
#[boxed]
pub async fn drive_pending_syncs<T>(pending: &PendingSyncs, fut: impl Future<Output = T>) -> T {
    let mut fut = std::pin::pin!(fut);
    poll_fn(|cx| match fut.as_mut().poll(cx) {
        Poll::Ready(out) => Poll::Ready(out),
        Poll::Pending => {
            // A concurrent task may park a new sync after this release, so
            // self-wake to check again on the next scheduler tick.
            release_pending_syncs(pending);
            cx.waker().wake_by_ref();
            Poll::Pending
        }
    })
    .await
}

/// Fail all pending syncs with an injected I/O error.
pub fn fail_pending_syncs(pending: &PendingSyncs) {
    for sync in mem::take(&mut *pending.lock()) {
        let _ = sync.release.send(Err(injected_sync_failure()));
    }
}

/// The error injected by [fail_pending_syncs] and [PendingSyncs::arm_fail].
fn injected_sync_failure() -> Error {
    Error::Io(std::io::Error::other("injected sync failure").into())
}

struct SyncWaiter {
    entered: oneshot::Sender<()>,
    release: oneshot::Receiver<Result<(), Error>>,
}

impl SyncWaiter {
    async fn wait(self) -> Result<(), Error> {
        self.entered.send_lossy(());
        self.release.await.map_err(|_| Error::Closed)??;
        Ok(())
    }
}

#[derive(Default)]
struct SyncGateState {
    tracking: bool,
    calls: usize,
    waiter: Option<SyncWaiter>,
}

impl PendingSyncs {
    /// Locks the deferred sync queue.
    pub fn lock(&self) -> commonware_utils::sync::MappedMutexGuard<'_, Vec<DeferredSync>> {
        commonware_utils::sync::MutexGuard::map(self.state.lock(), |state| &mut state.syncs)
    }

    /// Begins counting durability operations and blocks the next one behind a
    /// one-shot gate (pushed onto the deferred queue so tests can release it).
    ///
    /// Once the gate is consumed, started syncs park in the deferred queue as
    /// usual while [Self::calls] keeps counting.
    pub fn arm(&self) {
        let mut state = self.state.lock();
        assert!(!state.gate.tracking, "sync gate already armed");
        assert!(
            state.gate.waiter.is_none(),
            "sync gate already has a waiter"
        );
        state.gate.tracking = true;
        state.gate.calls = 0;
        let waiter = state.defer();
        state.gate.waiter = Some(waiter);
    }

    /// Returns the number of durability operations observed since [Self::arm].
    pub fn calls(&self) -> usize {
        self.state.lock().gate.calls
    }

    fn tracking(&self) -> bool {
        self.state.lock().gate.tracking
    }

    /// Releases every parked sync and permanently stops parking new ones: future started syncs
    /// proceed immediately (or fail, after [Self::arm_fail]). Unlike [release_pending_syncs],
    /// which drains the queue once, this is sticky. A gate armed via [Self::arm] parks in the
    /// same queue, so this releases it too.
    pub fn unblock(&self) {
        let (drained, fail) = {
            let mut state = self.state.lock();
            state.unblocked = true;
            (mem::take(&mut state.syncs), state.fail)
        };
        for sync in drained {
            let result = if fail {
                Err(injected_sync_failure())
            } else {
                Ok(())
            };
            let _ = sync.release.send(result);
        }
    }

    /// Arms every parked and future started sync to resolve to an injected error once released
    /// via [Self::unblock]. The release helpers send explicit results and ignore this.
    pub fn arm_fail(&self) {
        self.state.lock().fail = true;
    }

    /// Number of started syncs issued.
    pub fn starts(&self) -> usize {
        self.state.lock().starts
    }

    /// Number of started syncs whose completion futures have begun executing, parked or not.
    pub fn entered(&self) -> usize {
        self.state.lock().entered
    }

    /// Number of started syncs that completed durably.
    pub fn completions(&self) -> usize {
        self.state.lock().completions
    }

    async fn wait(&self) -> Result<(), Error> {
        let waiter = self.state.lock().observe();
        match waiter {
            Some(waiter) => waiter.wait().await,
            None => Ok(()),
        }
    }
}

/// Controls a [WriteFaultContext]: while armed, every `write_at` fails with an
/// injected error. Successful writes are counted.
#[derive(Clone, Default)]
pub struct WriteFaults {
    state: Arc<Mutex<WriteFaultState>>,
}

#[derive(Default)]
struct WriteFaultState {
    fail: bool,
    writes: u64,
}

impl WriteFaults {
    /// Start failing writes.
    pub fn arm(&self) {
        self.state.lock().fail = true;
    }

    /// Stop failing writes.
    pub fn disarm(&self) {
        self.state.lock().fail = false;
    }

    /// The number of successful writes so far.
    pub fn writes(&self) -> u64 {
        self.state.lock().writes
    }

    fn check(&self) -> Result<(), Error> {
        if self.state.lock().fail {
            return Err(Error::Io(
                std::io::Error::other("injected write failure").into(),
            ));
        }
        Ok(())
    }

    fn note(&self) {
        self.state.lock().writes += 1;
    }
}

/// Context wrapper whose blobs fail `write_at` while the shared [WriteFaults]
/// is armed, counting successful writes. Unlike [DelayedSyncContext], this injects failures
/// into inline writes issued before any blob sync starts.
#[derive(Clone)]
pub struct WriteFaultContext<E> {
    pub inner: E,
    pub faults: WriteFaults,
}

forward_context!(WriteFaultContext, faults);

impl<E: crate::Storage> crate::Storage for WriteFaultContext<E> {
    type Blob = WriteFaultBlob<E::Blob>;

    async fn open_versioned(
        &self,
        partition: &str,
        name: &[u8],
        versions: std::ops::RangeInclusive<BlobVersion>,
    ) -> Result<(Self::Blob, u64, BlobVersion), Error> {
        let (inner, len, version) = self.inner.open_versioned(partition, name, versions).await?;
        Ok((
            WriteFaultBlob {
                inner,
                faults: self.faults.clone(),
            },
            len,
            version,
        ))
    }

    async fn remove(&self, partition: &str, name: Option<&[u8]>) -> Result<(), Error> {
        self.inner.remove(partition, name).await
    }

    async fn scan(&self, partition: &str) -> Result<Vec<Vec<u8>>, Error> {
        self.inner.scan(partition).await
    }
}

/// Blob wrapper that fails `write_at` while its [WriteFaults] is armed.
pub struct WriteFaultBlob<B> {
    inner: B,
    faults: WriteFaults,
}

impl<B: Blob> Blob for WriteFaultBlob<B> {
    async fn read_at_buf(
        &self,
        offset: u64,
        len: usize,
        bufs: impl Into<IoBufsMut> + Send,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.inner.read_at_buf(offset, len, bufs, options).await
    }

    async fn read_at(
        &self,
        offset: u64,
        len: usize,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.inner.read_at(offset, len, options).await
    }

    async fn write_at(
        &self,
        offset: u64,
        bufs: impl Into<IoBufs> + Send,
        options: WriteOptions,
    ) -> Result<(), Error> {
        self.faults.check()?;
        self.inner.write_at(offset, bufs, options).await?;
        self.faults.note();
        Ok(())
    }

    async fn resize(&self, len: u64) -> Result<(), Error> {
        self.inner.resize(len).await
    }

    async fn sync(&self) -> Result<(), Error> {
        self.inner.sync().await
    }

    async fn start_sync(&self) -> Handle<()> {
        self.inner.start_sync().await
    }
}

/// Context wrapper whose blobs fail `sync` and `start_sync` for a single partition.
#[derive(Clone)]
pub struct SyncFaultContext<E> {
    pub inner: E,
    pub fail_partition: String,
}

forward_context!(SyncFaultContext, fail_partition);

impl<E: crate::Storage> crate::Storage for SyncFaultContext<E> {
    type Blob = SyncFaultBlob<E::Blob>;

    async fn open_versioned(
        &self,
        partition: &str,
        name: &[u8],
        versions: std::ops::RangeInclusive<BlobVersion>,
    ) -> Result<(Self::Blob, u64, BlobVersion), Error> {
        let (inner, len, version) = self.inner.open_versioned(partition, name, versions).await?;
        Ok((
            SyncFaultBlob {
                inner,
                faulty: partition == self.fail_partition,
            },
            len,
            version,
        ))
    }

    async fn remove(&self, partition: &str, name: Option<&[u8]>) -> Result<(), Error> {
        self.inner.remove(partition, name).await
    }

    async fn scan(&self, partition: &str) -> Result<Vec<Vec<u8>>, Error> {
        self.inner.scan(partition).await
    }
}

/// Blob wrapper that fails `sync` and `start_sync` when marked faulty.
pub struct SyncFaultBlob<B> {
    inner: B,
    faulty: bool,
}

impl<B: Blob> Blob for SyncFaultBlob<B> {
    async fn read_at_buf(
        &self,
        offset: u64,
        len: usize,
        bufs: impl Into<IoBufsMut> + Send,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.inner.read_at_buf(offset, len, bufs, options).await
    }

    async fn read_at(
        &self,
        offset: u64,
        len: usize,
        options: ReadOptions,
    ) -> Result<IoBufsMut, Error> {
        self.inner.read_at(offset, len, options).await
    }

    async fn write_at(
        &self,
        offset: u64,
        bufs: impl Into<IoBufs> + Send,
        options: WriteOptions,
    ) -> Result<(), Error> {
        self.inner.write_at(offset, bufs, options).await
    }

    async fn resize(&self, len: u64) -> Result<(), Error> {
        self.inner.resize(len).await
    }

    async fn sync(&self) -> Result<(), Error> {
        if self.faulty {
            let err = std::io::Error::other("injected partition sync fault");
            return Err(Error::Io(err.into()));
        }
        self.inner.sync().await
    }

    async fn start_sync(&self) -> Handle<()> {
        if self.faulty {
            return Handle::ready(self.sync().await);
        }
        self.inner.start_sync().await
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::{Clock, IoBufMut, Runner, Sink, Spawner, Storage as _, Stream, deterministic};
    use commonware_macros::select;
    use std::{thread::sleep, time::Duration};

    #[test]
    fn recording_context_preserves_data_and_records_options() {
        deterministic::Runner::default().start(|context| async move {
            let (context, recordings) = RecordingContext::new(context);
            let (blob, _) = context.open("recording", b"blob").await.unwrap();

            blob.write_at(0, b"data", WriteOptions::DONT_CACHE)
                .await
                .unwrap();
            let read = blob.read_at(0, 4, ReadOptions::DONT_CACHE).await.unwrap();
            assert_eq!(read.coalesce(), b"data");

            let read = blob
                .read_at_buf(0, 4, IoBufMut::with_capacity(4), ReadOptions::default())
                .await
                .unwrap();
            assert_eq!(read.coalesce(), b"data");

            assert_eq!(
                recordings.snapshot(),
                RecordingSnapshot {
                    reads: vec![ReadOptions::DONT_CACHE, ReadOptions::default()],
                    batches: Vec::new(),
                    writes: vec![WriteOptions::DONT_CACHE],
                }
            );
            recordings.clear();
            assert_eq!(recordings.snapshot(), RecordingSnapshot::default());
        });
    }

    async fn assert_read_options_forwarded<E: crate::Storage>(
        context: &E,
        recordings: &Recordings,
        partition: &str,
    ) {
        let (blob, _) = context.open(partition, b"blob").await.unwrap();
        blob.write_at(0, b"data", WriteOptions::default())
            .await
            .unwrap();
        recordings.clear();

        let read = blob.read_at(0, 4, ReadOptions::DONT_CACHE).await.unwrap();
        assert_eq!(read.coalesce(), b"data");

        let read = blob
            .read_at_buf(0, 4, IoBufMut::with_capacity(4), ReadOptions::DONT_CACHE)
            .await
            .unwrap();
        assert_eq!(read.coalesce(), b"data");

        assert_eq!(
            recordings.snapshot(),
            RecordingSnapshot {
                reads: vec![ReadOptions::DONT_CACHE, ReadOptions::DONT_CACHE],
                batches: Vec::new(),
                writes: Vec::new(),
            }
        );
    }

    #[test]
    fn delayed_sync_blob_forwards_read_options() {
        deterministic::Runner::default().start(|context| async move {
            let (inner, recordings) = RecordingContext::new(context);
            let context = DelayedSyncContext {
                inner,
                pending: PendingSyncs::default(),
            };

            assert_read_options_forwarded(&context, &recordings, "delayed_sync").await;
        });
    }

    #[test]
    fn write_fault_blob_forwards_read_options() {
        deterministic::Runner::default().start(|context| async move {
            let (inner, recordings) = RecordingContext::new(context);
            let context = WriteFaultContext {
                inner,
                faults: WriteFaults::default(),
            };

            assert_read_options_forwarded(&context, &recordings, "write_fault").await;
        });
    }

    #[test]
    fn sync_fault_blob_forwards_read_options() {
        deterministic::Runner::default().start(|context| async move {
            let (inner, recordings) = RecordingContext::new(context);
            let context = SyncFaultContext {
                inner,
                fail_partition: "sync_fault".to_string(),
            };

            assert_read_options_forwarded(&context, &recordings, "sync_fault").await;
        });
    }

    #[test]
    fn test_send_recv() {
        let (mut sink, mut stream) = Channel::init();
        let data = b"hello world";

        let executor = deterministic::Runner::default();
        executor.start(|_| async move {
            sink.send(data.as_slice()).await.unwrap();
            let received = stream.recv(data.len()).await.unwrap();
            assert_eq!(received.coalesce(), data);
        });
    }

    #[test]
    fn test_send_recv_partial_multiple() {
        let (mut sink, mut stream) = Channel::init();
        let data = b"hello";
        let data2 = b" world";

        let executor = deterministic::Runner::default();
        executor.start(|_| async move {
            sink.send(data.as_slice()).await.unwrap();
            sink.send(data2.as_slice()).await.unwrap();
            let received = stream.recv(5).await.unwrap();
            assert_eq!(received.coalesce(), b"hello");
            let received = stream.recv(5).await.unwrap();
            assert_eq!(received.coalesce(), b" worl");
            let received = stream.recv(1).await.unwrap();
            assert_eq!(received.coalesce(), b"d");
        });
    }

    #[test]
    fn test_send_recv_async() {
        let (mut sink, mut stream) = Channel::init();
        let data = b"hello world";

        let executor = deterministic::Runner::default();
        executor.start(|_| async move {
            let (received, _) = futures::try_join!(stream.recv(data.len()), async {
                sleep(Duration::from_millis(50));
                sink.send(data.as_slice()).await
            })
            .unwrap();
            assert_eq!(received.coalesce(), data);
        });
    }

    #[test]
    fn test_recv_error_sink_dropped_while_waiting() {
        let (sink, mut stream) = Channel::init();

        let executor = deterministic::Runner::default();
        executor.start(|context| async move {
            futures::join!(
                async {
                    let result = stream.recv(5).await;
                    assert!(matches!(result, Err(Error::RecvFailed)));
                    let result = stream.recv(5).await;
                    assert!(matches!(result, Err(Error::Closed)));
                },
                async {
                    // Wait for the stream to start waiting
                    context.sleep(Duration::from_millis(50)).await;
                    drop(sink);
                }
            );
        });
    }

    #[test]
    fn test_recv_error_sink_dropped_before_recv() {
        let (sink, mut stream) = Channel::init();
        drop(sink); // Drop sink immediately

        let executor = deterministic::Runner::default();
        executor.start(|_| async move {
            let result = stream.recv(5).await;
            assert!(matches!(result, Err(Error::RecvFailed)));
            let result = stream.recv(5).await;
            assert!(matches!(result, Err(Error::Closed)));
        });
    }

    #[test]
    fn test_send_error_stream_dropped() {
        let (mut sink, mut stream) = Channel::init();

        let executor = deterministic::Runner::default();
        executor.start(|context| async move {
            // Send some bytes
            assert!(sink.send(b"7 bytes".as_slice()).await.is_ok());

            // Spawn a task to initiate recv's where the first one will succeed and then will drop.
            let handle = context.child("recv").spawn(|_| async move {
                let _ = stream.recv(5).await;
                let _ = stream.recv(5).await;
            });

            // Give the async task a moment to start
            context.sleep(Duration::from_millis(50)).await;

            // Drop the stream by aborting the handle
            handle.abort();
            assert!(matches!(handle.await, Err(Error::Closed)));

            // Try to send a message. The stream is dropped, so this should fail.
            let result = sink.send(b"hello world".as_slice()).await;
            assert!(matches!(result, Err(Error::SendFailed)));
            let result = sink.send(b"hello world".as_slice()).await;
            assert!(matches!(result, Err(Error::Closed)));
        });
    }

    #[test]
    fn test_send_error_stream_dropped_before_send() {
        let (mut sink, stream) = Channel::init();
        drop(stream); // Drop stream immediately

        let executor = deterministic::Runner::default();
        executor.start(|_| async move {
            let result = sink.send(b"hello world".as_slice()).await;
            assert!(matches!(result, Err(Error::SendFailed)));
            let result = sink.send(b"hello world".as_slice()).await;
            assert!(matches!(result, Err(Error::Closed)));
        });
    }

    #[test]
    fn test_recv_timeout() {
        let (_sink, mut stream) = Channel::init();

        // If there is no data to read, test that the recv function just blocks.
        // The timeout should return first.
        let executor = deterministic::Runner::default();
        executor.start(|context| async move {
            select! {
                v = stream.recv(5) => {
                    panic!("unexpected value: {v:?}");
                },
                _ = context.sleep(Duration::from_millis(100)) => "timeout",
            };
        });
    }

    #[test]
    fn test_peek_empty() {
        let (_sink, stream) = Channel::init();

        // Peek on a fresh stream should return empty slice
        assert!(stream.peek(10).is_empty());
    }

    #[test]
    fn test_peek_after_partial_recv() {
        let (mut sink, mut stream) = Channel::init();

        let executor = deterministic::Runner::default();
        executor.start(|_| async move {
            // Send more data than we'll consume
            sink.send(b"hello world".as_slice()).await.unwrap();

            // Recv only part of it
            let received = stream.recv(5).await.unwrap();
            assert_eq!(received.coalesce(), b"hello");

            // Peek should show the remaining data
            assert_eq!(stream.peek(100), b" world");

            // Peek with smaller max_len
            assert_eq!(stream.peek(3), b" wo");

            // Peek doesn't consume - can peek again
            assert_eq!(stream.peek(100), b" world");

            // Recv consumes the peeked data
            let received = stream.recv(6).await.unwrap();
            assert_eq!(received.coalesce(), b" world");

            // Peek is now empty
            assert!(stream.peek(100).is_empty());
        });
    }

    #[test]
    fn test_peek_after_recv_wakeup() {
        let (mut sink, mut stream) = Channel::init_with_buffer_size(64);

        let executor = deterministic::Runner::default();
        executor.start(|context| async move {
            // Spawn recv that will block waiting
            let (tx, rx) = oneshot::channel();
            let recv_handle = context.child("recv").spawn(|_| async move {
                let data = stream.recv(3).await.unwrap();
                tx.send(stream).ok();
                data
            });

            // Let recv set up waiter
            context.sleep(Duration::from_millis(10)).await;

            // Send more than requested
            sink.send(b"ABCDEFGHIJ".as_slice()).await.unwrap();

            // Recv gets its 3 bytes
            let received = recv_handle.await.unwrap();
            assert_eq!(received.coalesce(), b"ABC");

            // Get stream back and verify peek sees remaining data
            let stream = rx.await.unwrap();
            assert_eq!(stream.peek(100), b"DEFGHIJ");
        });
    }

    #[test]
    fn test_peek_multiple_sends() {
        let (mut sink, mut stream) = Channel::init();

        let executor = deterministic::Runner::default();
        executor.start(|_| async move {
            // Send multiple chunks
            sink.send(b"aaa".as_slice()).await.unwrap();
            sink.send(b"bbb".as_slice()).await.unwrap();
            sink.send(b"ccc".as_slice()).await.unwrap();

            // Recv less than total
            let received = stream.recv(4).await.unwrap();
            assert_eq!(received.coalesce(), b"aaab");

            // Peek should show remaining
            assert_eq!(stream.peek(100), b"bbccc");
        });
    }

    #[test]
    fn test_buffer_size_limit() {
        // Use a small buffer capacity for testing
        let (mut sink, mut stream) = Channel::init_with_buffer_size(10);

        let executor = deterministic::Runner::default();
        executor.start(|context| async move {
            // Send more than buffer capacity concurrently with recv
            // so the sender can drain via backpressure.
            let send_handle = context.child("sender").spawn(|_| async move {
                sink.send(b"0123456789ABCDEF".as_slice()).await.unwrap();
                sink
            });

            // Recv a small amount - should only pull up to capacity (10 bytes)
            let received = stream.recv(2).await.unwrap();
            assert_eq!(received.coalesce(), b"01");

            // Peek should show remaining buffered data (8 bytes, not 14)
            assert_eq!(stream.peek(100), b"23456789");

            // The rest should still be in the channel buffer
            // Recv more to pull the remaining data
            let received = stream.recv(8).await.unwrap();
            assert_eq!(received.coalesce(), b"23456789");

            // Now peek should show next chunk from channel (up to capacity)
            let received = stream.recv(2).await.unwrap();
            assert_eq!(received.coalesce(), b"AB");

            assert_eq!(stream.peek(100), b"CDEF");

            // Ensure the sender completes
            send_handle.await.unwrap();
        });
    }

    #[test]
    fn test_recv_before_send() {
        // Use a small buffer capacity for testing
        let (mut sink, mut stream) = Channel::init_with_buffer_size(10);

        let executor = deterministic::Runner::default();
        executor.start(|context| async move {
            // Start recv before send (will wait)
            let recv_handle = context
                .child("recv")
                .spawn(|_| async move { stream.recv(3).await.unwrap() });

            // Give recv time to set up waiter
            context.sleep(Duration::from_millis(10)).await;

            // Send more than capacity
            sink.send(b"ABCDEFGHIJKLMNOP".as_slice()).await.unwrap();

            // Recv should get its 3 bytes
            let received = recv_handle.await.unwrap();
            assert_eq!(received.coalesce(), b"ABC");
        });
    }
}