Wire bounded output collection into the code mode runtime (#57850)

## Why

Bounding collected output also requires limiting the raw event backlog between
V8 and the cell actor. Previously, runtime events and cell output accumulated
in an unbounded channel and vector.

## What changed

- Select the event transport and `OutputCollector` from the cell's
  `output_truncation` mode. Server mode uses an ordered, bounded channel with
  backpressure; client mode retains untruncated output.
- Transfer the collector with each observation and restore it on failed
  delivery, preserving retained media and omission metadata.
- Include server-mode script errors in the observation's output budget. Build
  completion output only after completion wins against termination, so a
  rejected error cannot evict output returned by termination.
- Keep the public service on client-side truncation.

## Testing

Add regression coverage for channel backpressure, event ordering, receiver
closure, failed observation delivery, termination during completion commit,
and oversized errors across UTF-8 and omission boundaries.

GitOrigin-RevId: 7604b60416e33c636ca41138be65dc0fd452774a
This commit is contained in:
Sean Huang
2026-10-28 22:33:13 +00:00
committed by copyberry
parent 6d97d6ecb4
commit 92e802fb61
21 changed files with 686 additions and 355 deletions
@@ -40,7 +40,7 @@ impl CellHost for PanickingCallbackHost {
async fn commit_completion(
&self,
_stored_value_writes: HashMap<String, Arc<JsonValue>>,
_event: CellEvent,
_event: impl FnOnce() -> CellEvent + Send,
_cell_state: Arc<CellState>,
) -> CompletionCommit {
panic!("unexpected completion commit");
@@ -1,13 +1,9 @@
use codex_code_mode_protocol::CodeModeToolKind;
use codex_code_mode_protocol::ExecuteRequest;
use codex_code_mode_protocol::FunctionCallOutputContentItem;
use codex_code_mode_protocol::ImageDetail;
use codex_code_mode_protocol::ToolDefinition;
use codex_protocol::ToolName;
use crate::session_runtime::CreateCellRequest as CellRequest;
use crate::session_runtime::ImageDetail as CellImageDetail;
use crate::session_runtime::OutputItem as CellOutputItem;
use crate::session_runtime::ToolKind as CellToolKind;
pub(super) fn runtime_request(request: CellRequest) -> ExecuteRequest {
@@ -44,21 +40,3 @@ pub(super) fn cell_tool_kind(kind: CodeModeToolKind) -> CellToolKind {
CodeModeToolKind::Freeform => CellToolKind::Freeform,
}
}
pub(super) fn output_item(item: FunctionCallOutputContentItem) -> CellOutputItem {
match item {
FunctionCallOutputContentItem::InputText { text } => CellOutputItem::Text { text },
FunctionCallOutputContentItem::InputImage { image_url, detail } => CellOutputItem::Image {
image_url,
detail: detail.map(|detail| match detail {
ImageDetail::Auto => CellImageDetail::Auto,
ImageDetail::Low => CellImageDetail::Low,
ImageDetail::High => CellImageDetail::High,
ImageDetail::Original => CellImageDetail::Original,
}),
},
FunctionCallOutputContentItem::InputAudio { audio_url } => {
CellOutputItem::Audio { audio_url }
}
}
}
+84 -129
View File
@@ -18,7 +18,6 @@ use self::callbacks::report_task_result;
use self::callbacks::spawn_notification;
use self::callbacks::spawn_tool;
use self::conversions::cell_tool_kind;
use self::conversions::output_item;
use self::conversions::runtime_request;
use self::types::CellCommand;
pub(crate) use self::types::CellError;
@@ -31,16 +30,18 @@ pub(crate) use self::types::CompletionCommit;
use self::types::CompletionDelivery;
use self::types::ObservationDelivery;
use crate::TaskFailureHandler;
use crate::output::OutputCollector;
use crate::runtime::PendingRuntimeMode;
use crate::runtime::RuntimeCommand;
use crate::runtime::RuntimeControlCommand;
use crate::runtime::RuntimeEvent;
use crate::runtime::RuntimeEventReceiver;
use crate::runtime::runtime_event_channel;
use crate::runtime::spawn_runtime;
use crate::session_runtime::CellEvent;
use crate::session_runtime::CreateCellRequest as CellRequest;
use crate::session_runtime::Observation;
use crate::session_runtime::ObserveMode;
use crate::session_runtime::OutputItem;
use crate::session_runtime::ToolName as CellToolName;
pub(crate) struct CellActor;
@@ -61,7 +62,8 @@ impl CellActor {
),
String,
> {
let (event_tx, event_rx) = mpsc::unbounded_channel();
let (event_tx, event_rx) = runtime_event_channel(request.output_truncation);
let output = OutputCollector::new(request.output_truncation);
let (command_tx, command_rx) = mpsc::unbounded_channel();
let (initial_response_tx, initial_response_rx) = oneshot::channel();
let (runtime_tx, runtime_control_tx, runtime_terminate_handle) = spawn_runtime(
@@ -79,6 +81,7 @@ impl CellActor {
runtime_control_tx,
runtime_terminate_handle,
cell_state,
output,
},
event_rx,
command_rx,
@@ -99,6 +102,7 @@ struct CellContext {
runtime_control_tx: std::sync::mpsc::Sender<RuntimeControlCommand>,
runtime_terminate_handle: v8::IsolateHandle,
cell_state: Arc<CellState>,
output: OutputCollector,
}
struct Observer {
@@ -109,7 +113,7 @@ struct Observer {
async fn run_cell<H: CellHost>(
host: Arc<H>,
context: CellContext,
mut event_rx: mpsc::UnboundedReceiver<RuntimeEvent>,
mut event_rx: RuntimeEventReceiver,
command_rx: mpsc::UnboundedReceiver<CellCommand>,
initial_observer: Observer,
task_failure_handler: Option<TaskFailureHandler>,
@@ -119,10 +123,10 @@ async fn run_cell<H: CellHost>(
runtime_control_tx,
runtime_terminate_handle,
cell_state,
mut output,
} = context;
let cancellation_token = cell_state.cancellation_token();
let callback_cancellation_token = cancellation_token.child_token();
let mut content_items = Vec::new();
let mut pending_tool_call_ids = Vec::new();
let mut pending_frontier_ready = false;
let mut yield_signal = Some(initial_observer.observation.yield_signal.clone());
@@ -164,7 +168,7 @@ async fn run_cell<H: CellHost>(
&cell_state,
observer.take().map(|observer| observer.response_tx),
CellEvent::Terminated {
content_items: std::mem::take(&mut content_items),
content_items: output.take(),
},
);
break;
@@ -202,27 +206,8 @@ async fn run_cell<H: CellHost>(
continue;
}
if matches!(mode, ObserveMode::PendingFrontier) && pending_frontier_ready {
pending_frontier_ready = false;
match send_cell_event(
response_tx,
CellEvent::Pending {
content_items: std::mem::take(&mut content_items),
pending_tool_call_ids: std::mem::take(&mut pending_tool_call_ids),
},
) {
Ok(()) => {}
Err(CellEvent::Pending {
content_items: undelivered_items,
pending_tool_call_ids: undelivered_tool_call_ids,
}) => {
content_items = undelivered_items;
pending_tool_call_ids = undelivered_tool_call_ids;
pending_frontier_ready = true;
}
Err(event) => {
panic!("pending delivery returned an unexpected event: {event:?}")
}
}
pending_frontier_ready =
!send_pending(&mut output, response_tx, &mut pending_tool_call_ids);
continue;
}
yield_signal = Some(observation.yield_signal.clone());
@@ -259,19 +244,13 @@ async fn run_cell<H: CellHost>(
} => {
yield_timer = None;
yield_signal = None;
restore_undelivered_yield(
send_observer_event(
observer.take(),
CellEvent::Yielded {
content_items: std::mem::take(&mut content_items),
},
),
&mut content_items,
);
if let Some(observer) = observer.take() {
send_yield(&mut output, observer.response_tx);
}
}
maybe_event = async {
if runtime_closed {
std::future::pending::<Option<RuntimeEvent>>().await
std::future::pending().await
} else {
event_rx.recv().await
}
@@ -292,7 +271,7 @@ async fn run_cell<H: CellHost>(
&cell_state,
observer.take().map(|observer| observer.response_tx),
CellEvent::Terminated {
content_items: std::mem::take(&mut content_items),
content_items: output.take(),
},
);
break;
@@ -313,22 +292,17 @@ async fn run_cell<H: CellHost>(
task_failure_handler.as_ref(),
)
.await;
let event = CellEvent::Completed {
content_items: std::mem::take(&mut content_items),
result_buffered_for: std::time::Duration::ZERO,
error_text: Some("exec runtime ended unexpectedly".to_string()),
};
let rejected_event = match host
host
.commit_completion(
HashMap::new(),
event,
|| {
take_completed_event(&mut output, Some(
"exec runtime ended unexpectedly".to_string(),
))
},
Arc::clone(&cell_state),
)
.await
{
CompletionCommit::Committed => None,
CompletionCommit::Rejected(event) => Some(event),
};
.await;
match cell_state.deliver_completion(
observer.take().map(|observer| observer.response_tx),
) {
@@ -339,7 +313,7 @@ async fn run_cell<H: CellHost>(
&cell_state,
response_tx,
CellEvent::Terminated {
content_items: rejected_completion_content(rejected_event),
content_items: output.take(),
},
);
break;
@@ -356,59 +330,29 @@ async fn run_cell<H: CellHost>(
}
RuntimeEvent::Pending => {
runtime_paused = true;
if matches!(
observer.as_ref().map(|observer| observer.observation.mode),
Some(ObserveMode::PendingFrontier)
) {
if let Some(observer) = observer.take_if(|observer| {
matches!(observer.observation.mode, ObserveMode::PendingFrontier)
}) {
yield_timer = None;
yield_signal = None;
pending_frontier_ready = false;
match send_observer_event(
observer.take(),
CellEvent::Pending {
content_items: std::mem::take(&mut content_items),
pending_tool_call_ids: std::mem::take(
&mut pending_tool_call_ids,
),
},
) {
Ok(()) => {}
Err(CellEvent::Pending {
content_items: undelivered_items,
pending_tool_call_ids: undelivered_tool_call_ids,
}) => {
content_items = undelivered_items;
pending_tool_call_ids = undelivered_tool_call_ids;
pending_frontier_ready = true;
}
Err(event) => {
panic!("pending delivery returned an unexpected event: {event:?}")
}
}
pending_frontier_ready =
!send_pending(&mut output, observer.response_tx, &mut pending_tool_call_ids);
} else {
pending_tool_call_ids.clear();
let _ = runtime_control_tx.send(RuntimeControlCommand::Continue);
runtime_paused = false;
}
}
RuntimeEvent::ContentItem(item) => content_items.push(output_item(item)),
RuntimeEvent::ContentItem(item) => {
output.push(item);
}
RuntimeEvent::YieldRequested => {
let yield_observer = matches!(
observer.as_ref().map(|observer| observer.observation.mode),
Some(ObserveMode::YieldAfter(_))
);
if yield_observer {
if let Some(observer) = observer.take_if(|observer| {
matches!(observer.observation.mode, ObserveMode::YieldAfter(_))
}) {
yield_timer = None;
yield_signal = None;
restore_undelivered_yield(
send_observer_event(
observer.take(),
CellEvent::Yielded {
content_items: std::mem::take(&mut content_items),
},
),
&mut content_items,
);
send_yield(&mut output, observer.response_tx);
}
}
RuntimeEvent::Notify { call_id, text } => {
@@ -456,7 +400,7 @@ async fn run_cell<H: CellHost>(
&cell_state,
observer.take().map(|observer| observer.response_tx),
CellEvent::Terminated {
content_items: std::mem::take(&mut content_items),
content_items: output.take(),
},
);
break;
@@ -469,22 +413,13 @@ async fn run_cell<H: CellHost>(
task_failure_handler.as_ref(),
)
.await;
let event = CellEvent::Completed {
content_items: std::mem::take(&mut content_items),
result_buffered_for: std::time::Duration::ZERO,
error_text,
};
let rejected_event = match host
host
.commit_completion(
stored_value_writes,
event,
|| take_completed_event(&mut output, error_text),
Arc::clone(&cell_state),
)
.await
{
CompletionCommit::Committed => None,
CompletionCommit::Rejected(event) => Some(event),
};
.await;
match cell_state.deliver_completion(
observer.take().map(|observer| observer.response_tx),
) {
@@ -495,7 +430,7 @@ async fn run_cell<H: CellHost>(
&cell_state,
response_tx,
CellEvent::Terminated {
content_items: rejected_completion_content(rejected_event),
content_items: output.take(),
},
);
break;
@@ -539,11 +474,14 @@ async fn run_cell<H: CellHost>(
host.closed().await;
}
fn send_observer_event(observer: Option<Observer>, event: CellEvent) -> Result<(), CellEvent> {
let Some(observer) = observer else {
return Err(event);
};
send_cell_event(observer.response_tx, event)
/// Construct completion only after it has won the terminal decision.
fn take_completed_event(output: &mut OutputCollector, error_text: Option<String>) -> CellEvent {
let error_text = error_text.map(|error_text| output.collect_error(error_text));
CellEvent::Completed {
content_items: output.take(),
error_text,
result_buffered_for: std::time::Duration::ZERO,
}
}
fn send_cell_event(
@@ -557,25 +495,42 @@ fn send_cell_event(
}
}
fn restore_undelivered_yield(delivery: Result<(), CellEvent>, content_items: &mut Vec<OutputItem>) {
match delivery {
Ok(()) => {}
Err(CellEvent::Yielded {
content_items: mut undelivered_items,
}) => {
undelivered_items.append(content_items);
*content_items = undelivered_items;
}
Err(event) => panic!("yield delivery returned an unexpected event: {event:?}"),
}
fn send_yield(
output: &mut OutputCollector,
response_tx: oneshot::Sender<Result<CellEvent, CellError>>,
) {
output.try_deliver(|content_items| {
send_cell_event(response_tx, CellEvent::Yielded { content_items }).map_err(|event| {
let CellEvent::Yielded { content_items } = event else {
unreachable!("yield delivery returned a different event");
};
content_items
})
});
}
fn rejected_completion_content(event: Option<CellEvent>) -> Vec<OutputItem> {
match event {
Some(CellEvent::Completed { content_items, .. }) => content_items,
None => Vec::new(),
Some(event) => panic!("completion commit rejected an unexpected event: {event:?}"),
}
fn send_pending(
output: &mut OutputCollector,
response_tx: oneshot::Sender<Result<CellEvent, CellError>>,
pending_tool_call_ids: &mut Vec<String>,
) -> bool {
output.try_deliver(|content_items| {
let pending = CellEvent::Pending {
content_items,
pending_tool_call_ids: std::mem::take(pending_tool_call_ids),
};
send_cell_event(response_tx, pending).map_err(|event| {
let CellEvent::Pending {
content_items,
pending_tool_call_ids: returned_ids,
} = event
else {
unreachable!("pending delivery returned a different event");
};
*pending_tool_call_ids = returned_ids;
content_items
})
})
}
fn finish_termination(
@@ -5,8 +5,9 @@ use std::sync::atomic::Ordering;
use std::sync::mpsc as std_mpsc;
use std::time::Duration;
use codex_code_mode_protocol::CodeModeOutputTruncation;
use codex_code_mode_protocol::ExecuteRequest;
use codex_code_mode_protocol::FunctionCallOutputContentItem;
use codex_code_mode_protocol::host::BOUNDED_OUTPUT_MAX_IMAGE_BYTES;
use pretty_assertions::assert_eq;
use serde_json::Value as JsonValue;
use tokio::sync::mpsc;
@@ -14,9 +15,14 @@ use tokio::sync::oneshot;
use tokio_util::sync::CancellationToken;
use super::*;
use crate::runtime::RuntimeEventSender;
use crate::session_runtime::OutputItem;
struct TestHost;
#[derive(Default)]
struct TestHost {
commit_started: CancellationToken,
resume_commit: Option<CancellationToken>,
}
#[derive(Default)]
struct RecordingHost {
@@ -44,9 +50,13 @@ impl CellHost for TestHost {
async fn commit_completion(
&self,
_stored_value_writes: HashMap<String, Arc<JsonValue>>,
event: CellEvent,
event: impl FnOnce() -> CellEvent + Send,
cell_state: Arc<CellState>,
) -> CompletionCommit {
self.commit_started.cancel();
if let Some(resume) = &self.resume_commit {
resume.cancelled().await;
}
cell_state.commit_completion(event, || {})
}
@@ -75,7 +85,7 @@ impl CellHost for RecordingHost {
async fn commit_completion(
&self,
_stored_value_writes: HashMap<String, Arc<JsonValue>>,
event: CellEvent,
event: impl FnOnce() -> CellEvent + Send,
cell_state: Arc<CellState>,
) -> CompletionCommit {
cell_state.commit_completion(event, || {})
@@ -85,7 +95,7 @@ impl CellHost for RecordingHost {
}
struct CellActorHarness {
event_tx: mpsc::UnboundedSender<RuntimeEvent>,
event_tx: RuntimeEventSender,
handle: CellHandle,
yield_signal: CancellationToken,
initial_event_rx: oneshot::Receiver<Result<CellEvent, CellError>>,
@@ -95,7 +105,7 @@ struct CellActorHarness {
}
fn spawn_cell_actor_harness(initial_observe_mode: ObserveMode) -> CellActorHarness {
spawn_cell_actor_harness_with_host(initial_observe_mode, Arc::new(TestHost))
spawn_cell_actor_harness_with_host(initial_observe_mode, Arc::new(TestHost::default()))
}
fn spawn_cell_actor_harness_with_host<H: CellHost>(
@@ -105,6 +115,7 @@ fn spawn_cell_actor_harness_with_host<H: CellHost>(
spawn_cell_actor_harness_with_host_and_failure_handler(
initial_observe_mode,
host,
CodeModeOutputTruncation::Client,
/*task_failure_handler*/ None,
)
}
@@ -112,9 +123,11 @@ fn spawn_cell_actor_harness_with_host<H: CellHost>(
fn spawn_cell_actor_harness_with_host_and_failure_handler<H: CellHost>(
initial_observe_mode: ObserveMode,
host: Arc<H>,
output_truncation: CodeModeOutputTruncation,
task_failure_handler: Option<TaskFailureHandler>,
) -> CellActorHarness {
let (event_tx, event_rx) = mpsc::unbounded_channel();
let (event_tx, event_rx) = runtime_event_channel(output_truncation);
let output = OutputCollector::new(output_truncation);
let (command_tx, command_rx) = mpsc::unbounded_channel();
let (initial_event_tx, initial_event_rx) = oneshot::channel();
let (runtime_event_tx, runtime_event_rx) = mpsc::unbounded_channel();
@@ -143,6 +156,7 @@ fn spawn_cell_actor_harness_with_host_and_failure_handler<H: CellHost>(
runtime_control_tx,
runtime_terminate_handle,
cell_state,
output,
},
event_rx,
command_rx,
@@ -169,7 +183,8 @@ async fn unexpected_runtime_thread_exit_is_reported_to_the_session_owner() {
let (failure_tx, mut failure_rx) = mpsc::unbounded_channel();
let harness = spawn_cell_actor_harness_with_host_and_failure_handler(
ObserveMode::YieldAfter(Duration::from_secs(60)),
Arc::new(TestHost),
Arc::new(TestHost::default()),
CodeModeOutputTruncation::Client,
Some(Arc::new(move |reason| {
let _ = failure_tx.send(reason);
})),
@@ -193,19 +208,74 @@ async fn unexpected_runtime_thread_exit_is_reported_to_the_session_owner() {
harness.task.await.expect("cell task");
}
#[tokio::test]
async fn termination_during_completion_commit_preserves_output_and_discards_losing_error() {
for mode in [
CodeModeOutputTruncation::Client,
CodeModeOutputTruncation::Server,
] {
// The actor has received Result but cannot commit until the host releases
// the session store. Termination must decide which output is published.
let resume_commit = CancellationToken::new();
let host = Arc::new(TestHost {
resume_commit: Some(resume_commit.clone()),
..TestHost::default()
});
let harness = spawn_cell_actor_harness_with_host_and_failure_handler(
ObserveMode::YieldAfter(Duration::from_secs(60)),
Arc::clone(&host),
mode,
/*task_failure_handler*/ None,
);
// Simulate V8's dedicated thread, where server-mode sends can block
// while the event queue is full.
let event_tx = harness.event_tx.clone();
tokio::task::spawn_blocking(move || {
event_tx
.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText {
text: "output that must survive termination".to_string(),
}))
.expect("output");
event_tx
.blocking_send(RuntimeEvent::Result {
stored_value_writes: HashMap::new(),
error_text: Some("uncommitted error".repeat(128 * 1024)),
})
.expect("completion");
})
.await
.expect("runtime producer");
tokio::time::timeout(Duration::from_secs(5), host.commit_started.cancelled())
.await
.expect("commit reached");
let termination = harness.handle.terminate();
resume_commit.cancel();
let mut expected_output = OutputCollector::new(mode);
expected_output.push(OutputItem::InputText {
text: "output that must survive termination".to_string(),
});
let expected = Ok(CellEvent::Terminated {
content_items: expected_output,
});
assert_eq!(termination.await, expected, "{mode:?}");
assert_eq!(harness.initial_event_rx.await.expect("observer"), expected);
harness.task.await.expect("cell task");
}
}
#[tokio::test]
async fn runtime_thread_panic_remains_a_cell_error_without_owner_supervision() {
let harness = spawn_cell_actor_harness(ObserveMode::YieldAfter(Duration::from_secs(60)));
harness
.event_tx
.send(RuntimeEvent::ThreadPanicked)
.blocking_send(RuntimeEvent::ThreadPanicked)
.expect("runtime panic event");
drop(harness.event_tx);
assert_eq!(
harness.initial_event_rx.await.expect("initial event"),
Ok(CellEvent::Completed {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
result_buffered_for: Duration::ZERO,
error_text: Some("exec runtime ended unexpectedly".to_string()),
})
@@ -226,20 +296,21 @@ async fn wait_for_notification(host: &RecordingHost) {
#[tokio::test]
async fn yield_timer_preempts_buffered_runtime_output() {
let harness = spawn_cell_actor_harness(ObserveMode::YieldAfter(Duration::ZERO));
harness.event_tx.send(RuntimeEvent::Started).unwrap();
harness
.event_tx
.send(RuntimeEvent::ContentItem(
FunctionCallOutputContentItem::InputText {
text: "queued output".to_string(),
},
))
.blocking_send(RuntimeEvent::Started)
.unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText {
text: "queued output".to_string(),
}))
.unwrap();
assert_eq!(
harness.initial_event_rx.await.unwrap(),
Ok(CellEvent::Yielded {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
})
);
@@ -248,9 +319,9 @@ async fn yield_timer_preempts_buffered_runtime_output() {
assert_eq!(
termination.await,
Ok(CellEvent::Terminated {
content_items: vec![OutputItem::Text {
content_items: OutputCollector::Client(vec![OutputItem::InputText {
text: "queued output".to_string(),
}],
}]),
})
);
harness.task.await.unwrap();
@@ -259,19 +330,23 @@ async fn yield_timer_preempts_buffered_runtime_output() {
#[tokio::test]
async fn queued_termination_preempts_unobserved_runtime_completion() {
let harness = spawn_cell_actor_harness(ObserveMode::YieldAfter(Duration::ZERO));
harness.event_tx.send(RuntimeEvent::Started).unwrap();
harness.event_tx.send(RuntimeEvent::YieldRequested).unwrap();
harness
.event_tx
.send(RuntimeEvent::ContentItem(
FunctionCallOutputContentItem::InputText {
text: "queued output".to_string(),
},
))
.blocking_send(RuntimeEvent::Started)
.unwrap();
harness
.event_tx
.send(RuntimeEvent::Result {
.blocking_send(RuntimeEvent::YieldRequested)
.unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText {
text: "queued output".to_string(),
}))
.unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::Result {
stored_value_writes: HashMap::new(),
error_text: None,
})
@@ -280,9 +355,9 @@ async fn queued_termination_preempts_unobserved_runtime_completion() {
let termination = harness.handle.terminate();
let terminated = Ok(CellEvent::Terminated {
content_items: vec![OutputItem::Text {
content_items: OutputCollector::Client(vec![OutputItem::InputText {
text: "queued output".to_string(),
}],
}]),
});
assert_eq!(termination.await, terminated.clone());
assert_eq!(harness.initial_event_rx.await.unwrap(), terminated);
@@ -296,7 +371,10 @@ async fn observation_dropped_before_dequeue_does_not_consume_output() {
ObserveMode::YieldAfter(Duration::from_secs(60)),
Arc::clone(&host),
);
harness.event_tx.send(RuntimeEvent::YieldRequested).unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::YieldRequested)
.unwrap();
assert!(harness.initial_event_rx.await.unwrap().is_ok());
drop(
@@ -306,16 +384,17 @@ async fn observation_dropped_before_dequeue_does_not_consume_output() {
);
harness
.event_tx
.send(RuntimeEvent::ContentItem(
FunctionCallOutputContentItem::InputText {
text: "survives pre-dequeue cancellation".to_string(),
},
))
.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText {
text: "survives pre-dequeue cancellation".to_string(),
}))
.unwrap();
harness.event_tx.send(RuntimeEvent::YieldRequested).unwrap();
harness
.event_tx
.send(RuntimeEvent::Notify {
.blocking_send(RuntimeEvent::YieldRequested)
.unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::Notify {
call_id: "after-dropped-command".to_string(),
text: "barrier".to_string(),
})
@@ -328,9 +407,9 @@ async fn observation_dropped_before_dequeue_does_not_consume_output() {
.observe(ObserveMode::YieldAfter(Duration::ZERO))
.await,
Ok(CellEvent::Yielded {
content_items: vec![OutputItem::Text {
content_items: OutputCollector::Client(vec![OutputItem::InputText {
text: "survives pre-dequeue cancellation".to_string(),
}],
}]),
})
);
@@ -339,7 +418,7 @@ async fn observation_dropped_before_dequeue_does_not_consume_output() {
assert_eq!(
termination.await,
Ok(CellEvent::Terminated {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
})
);
harness.task.await.unwrap();
@@ -352,7 +431,10 @@ async fn dropped_yield_observer_preserves_output_for_the_next_observation() {
ObserveMode::YieldAfter(Duration::from_secs(60)),
Arc::clone(&host),
);
harness.event_tx.send(RuntimeEvent::YieldRequested).unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::YieldRequested)
.unwrap();
assert!(harness.initial_event_rx.await.unwrap().is_ok());
let dropped_observation = harness
@@ -368,16 +450,17 @@ async fn dropped_yield_observer_preserves_output_for_the_next_observation() {
drop(dropped_observation);
harness
.event_tx
.send(RuntimeEvent::ContentItem(
FunctionCallOutputContentItem::InputText {
text: "survives active cancellation".to_string(),
},
))
.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText {
text: "survives active cancellation".to_string(),
}))
.unwrap();
harness.event_tx.send(RuntimeEvent::YieldRequested).unwrap();
harness
.event_tx
.send(RuntimeEvent::Notify {
.blocking_send(RuntimeEvent::YieldRequested)
.unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::Notify {
call_id: "after-dropped-observer".to_string(),
text: "barrier".to_string(),
})
@@ -390,9 +473,9 @@ async fn dropped_yield_observer_preserves_output_for_the_next_observation() {
.observe(ObserveMode::YieldAfter(Duration::ZERO))
.await,
Ok(CellEvent::Yielded {
content_items: vec![OutputItem::Text {
content_items: OutputCollector::Client(vec![OutputItem::InputText {
text: "survives active cancellation".to_string(),
}],
}]),
})
);
@@ -401,7 +484,7 @@ async fn dropped_yield_observer_preserves_output_for_the_next_observation() {
assert_eq!(
termination.await,
Ok(CellEvent::Terminated {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
})
);
harness.task.await.unwrap();
@@ -414,7 +497,10 @@ async fn dropped_pending_observer_preserves_the_frontier_for_the_next_observatio
ObserveMode::YieldAfter(Duration::from_secs(60)),
Arc::clone(&host),
);
harness.event_tx.send(RuntimeEvent::YieldRequested).unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::YieldRequested)
.unwrap();
assert!(harness.initial_event_rx.await.unwrap().is_ok());
let dropped_observation = harness.handle.observe(ObserveMode::PendingFrontier);
@@ -423,9 +509,16 @@ async fn dropped_pending_observer_preserves_the_frontier_for_the_next_observatio
Err(CellError::Busy)
);
drop(dropped_observation);
let item = OutputItem::InputText {
text: "before frontier".to_owned(),
};
harness
.event_tx
.send(RuntimeEvent::ToolCall {
.blocking_send(RuntimeEvent::ContentItem(item.clone()))
.unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::ToolCall {
id: "tool-1".to_string(),
name: codex_protocol::ToolName {
name: "echo".to_string(),
@@ -435,10 +528,13 @@ async fn dropped_pending_observer_preserves_the_frontier_for_the_next_observatio
input: Some(serde_json::json!({})),
})
.unwrap();
harness.event_tx.send(RuntimeEvent::Pending).unwrap();
harness
.event_tx
.send(RuntimeEvent::Notify {
.blocking_send(RuntimeEvent::Pending)
.unwrap();
harness
.event_tx
.blocking_send(RuntimeEvent::Notify {
call_id: "after-dropped-pending".to_string(),
text: "barrier".to_string(),
})
@@ -448,7 +544,7 @@ async fn dropped_pending_observer_preserves_the_frontier_for_the_next_observatio
assert_eq!(
harness.handle.observe(ObserveMode::PendingFrontier).await,
Ok(CellEvent::Pending {
content_items: Vec::new(),
content_items: OutputCollector::Client(vec![item]),
pending_tool_call_ids: vec!["tool-1".to_string()],
})
);
@@ -462,7 +558,7 @@ async fn dropped_pending_observer_preserves_the_frontier_for_the_next_observatio
assert_eq!(
termination.await,
Ok(CellEvent::Terminated {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
})
);
harness.task.await.unwrap();
@@ -472,12 +568,12 @@ async fn dropped_pending_observer_preserves_the_frontier_for_the_next_observatio
async fn only_the_first_termination_claims_a_buffered_completion() {
let cell_state = CellState::new(CancellationToken::new());
let completion = CellEvent::Completed {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
result_buffered_for: Duration::ZERO,
error_text: None,
};
assert_eq!(
cell_state.commit_completion(completion.clone(), || {}),
cell_state.commit_completion(|| completion.clone(), || {}),
CompletionCommit::Committed
);
assert!(matches!(
@@ -493,7 +589,7 @@ async fn only_the_first_termination_claims_a_buffered_completion() {
assert_eq!(first_termination.await, Ok(completion.clone()));
assert_eq!(
cell_state.finish_termination(CellEvent::Terminated {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
}),
Some(completion)
);
@@ -504,20 +600,17 @@ async fn termination_claim_prevents_stored_value_commit() {
let cell_state = CellState::new(CancellationToken::new());
let termination = cell_state.request_termination();
let mut commit_ran = false;
let completion = CellEvent::Completed {
content_items: Vec::new(),
result_buffered_for: Duration::ZERO,
error_text: None,
};
assert_eq!(
cell_state.commit_completion(completion.clone(), || commit_ran = true),
CompletionCommit::Rejected(completion)
cell_state.commit_completion(
|| panic!("terminated cell must not construct completion output"),
|| commit_ran = true
),
CompletionCommit::Rejected
);
assert!(!commit_ran);
let terminated = CellEvent::Terminated {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
};
assert_eq!(
cell_state.finish_termination(terminated.clone()),
@@ -530,14 +623,14 @@ async fn termination_claim_prevents_stored_value_commit() {
async fn failed_completion_delivery_rebuffers_the_event() {
let cell_state = CellState::new(CancellationToken::new());
let completion = |result_buffered_for| CellEvent::Completed {
content_items: vec![OutputItem::Text {
content_items: OutputCollector::Client(vec![OutputItem::InputText {
text: "done".to_string(),
}],
}]),
result_buffered_for,
error_text: None,
};
assert_eq!(
cell_state.commit_completion(completion(Duration::ZERO), || {}),
cell_state.commit_completion(|| completion(Duration::ZERO), || {}),
CompletionCommit::Committed
);
let (response_tx, response_rx) = oneshot::channel();
@@ -565,3 +658,49 @@ async fn failed_completion_delivery_rebuffers_the_event() {
));
assert_eq!(response_rx.try_recv(), Ok(Ok(completion(2 * interval))));
}
#[test]
fn a_failed_yield_restores_image_allocation_and_omission_order() {
let mut output = OutputCollector::new(CodeModeOutputTruncation::Server);
let image_url = "i".repeat(BOUNDED_OUTPUT_MAX_IMAGE_BYTES);
let pointer = image_url.as_ptr();
output.push(OutputItem::InputImage {
image_url,
detail: None,
});
output.push(OutputItem::InputImage {
image_url: "excess".to_string(),
detail: None,
});
let (tx, rx) = oneshot::channel();
drop(rx);
send_yield(&mut output, tx);
output.push(OutputItem::InputText {
text: "after".to_string(),
});
let (tx, mut rx) = oneshot::channel();
send_yield(&mut output, tx);
let CellEvent::Yielded { content_items } = rx.try_recv().unwrap().unwrap() else {
panic!("expected a yield");
};
let items = crate::output::render_output(
content_items.into_collected_output(),
/*max_output_tokens*/ None,
);
let [
OutputItem::InputImage { image_url, .. },
OutputItem::InputText { text: notice },
OutputItem::InputText { text },
] = items.as_slice()
else {
panic!("expected image, omission notice, and subsequent text");
};
assert_eq!(image_url.as_ptr(), pointer);
assert!(notice.contains("1 images"), "{notice}");
assert_eq!(text, "after");
assert_eq!(
output.take(),
OutputCollector::new(CodeModeOutputTruncation::Server)
);
}
@@ -55,7 +55,7 @@ pub(crate) trait CellHost: Send + Sync + 'static {
fn commit_completion(
&self,
stored_value_writes: HashMap<String, Arc<JsonValue>>,
event: CellEvent,
event: impl FnOnce() -> CellEvent + Send,
cell_state: Arc<CellState>,
) -> impl Future<Output = CompletionCommit> + Send;
@@ -136,7 +136,7 @@ pub(crate) enum CompletionDelivery {
#[derive(Debug, PartialEq)]
pub(crate) enum CompletionCommit {
Committed,
Rejected(CellEvent),
Rejected,
}
pub(crate) enum ObservationDelivery {
@@ -200,7 +200,7 @@ impl CellState {
pub(crate) fn commit_completion(
&self,
event: CellEvent,
event: impl FnOnce() -> CellEvent,
commit: impl FnOnce(),
) -> CompletionCommit {
let mut phase = self
@@ -208,11 +208,13 @@ impl CellState {
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner);
if !matches!(*phase, CellPhase::Running) || self.cancellation_token.is_cancelled() {
return CompletionCommit::Rejected(event);
return CompletionCommit::Rejected;
}
commit();
*phase = CellPhase::Completed {
event,
// Construct terminal output only after completion wins. In bounded
// mode, an error may evict user output; termination must not see it.
event: event(),
completed_at: Instant::now(),
};
CompletionCommit::Committed
-5
View File
@@ -1,9 +1,4 @@
mod cell_actor;
#[allow(
dead_code,
unused_imports,
reason = "Wired into the runtime in the next stack layer"
)]
mod output;
mod runtime;
mod service;
@@ -11,6 +11,7 @@ mod output_collection;
mod output_render;
pub(crate) use output_collection::BoundedOutput;
#[allow(unused_imports, reason = "Renderer is wired in the next stack layer")]
pub(crate) use output_render::render_output;
/// The actor transfers the whole collector at each observation, so a failed
@@ -38,6 +39,20 @@ impl OutputCollector {
}
}
/// Server errors share the observation's budget; preserve their presence
/// for completion status. Client errors retain their original formatting.
pub(crate) fn collect_error(&mut self, error_text: String) -> String {
match self {
Self::Client(_) => error_text,
Self::Server(output) => {
output
.get_or_insert_with(Box::default)
.push_error(&error_text);
String::new()
}
}
}
pub(crate) fn take(&mut self) -> Self {
match self {
Self::Client(items) => Self::Client(std::mem::take(items)),
@@ -435,6 +435,34 @@ impl HeadTailOutput {
}
impl BoundedOutput {
/// A completed error is one logical text item, including its label. Do not
/// allocate an arbitrarily large formatted copy before selecting its output.
pub(crate) fn push_error(&mut self, error: &str) {
const LABEL: &str = "Script error:\n";
// No selection can retain more than MAX_OUTPUT_BYTES from either end
// of this item. Keep both possible spans until push() decides how the
// existing head, tail and item pressure affect them. Round outward so
// a UTF-8 boundary cannot pull bytes from the prefix across the gap.
let head_end = error.ceil_char_boundary(error.len().min(MAX_OUTPUT_BYTES - LABEL.len()));
let tail_start =
error.floor_char_boundary(error.len().saturating_sub(MAX_OUTPUT_BYTES).max(head_end));
self.push(OutputItem::InputText {
text: format!("{LABEL}{}{}", &error[..head_end], &error[tail_start..]),
});
let skipped = tail_start - head_end;
if skipped > 0 {
// This item alone exceeds the content budget. Its newest, retained
// suffix owns the gap, even when an image separates it from older loss.
let Retention::HeadTail(output) = &mut self.state else {
unreachable!("an oversized error must enter head/tail retention");
};
let Some(RetainedItem::Partial(text)) = output.tail.back_mut() else {
unreachable!("the tail retains the end of the oversized error");
};
text.omitted = text.omitted.saturating_add(skipped as u64);
}
}
pub(crate) fn push(&mut self, item: OutputItem) {
if let Err(loss) = self.media.admit(&item) {
self.push_retained(RetainedItem::Gap(loss));
@@ -299,3 +299,34 @@ fn repeated_eviction_keeps_recent_writes_and_gaps_on_each_side_of_an_image() {
assert_eq!(after.omitted.text_bytes, (writes * chunk.len()) as u64);
assert_eq!(after.omitted.text_items, (1 + writes - recent) as u64);
}
#[test]
fn terminal_error_matches_ordinary_text_across_loss_boundaries() {
let label = "Script error:\n";
// Rounding inward at the emoji can wrongly pull XYZ into the retained suffix.
let error = format!(
"{}XYZ{}💡{}",
"p".repeat(MAX_OUTPUT_BYTES - label.len() - 3),
"m".repeat(22),
"s".repeat(MAX_OUTPUT_BYTES - 3),
);
let half = text(&"p".repeat(MAX_OUTPUT_BYTES / 2));
for (prefix, gap_count) in [
(Vec::new(), 1),
(vec![text(""); MAX_OUTPUT_ITEMS / 2], 1),
(vec![half.clone(), text("lost"), image("i".into()), half], 2),
] {
let mut terminal = BoundedOutput::default();
let mut ordinary = BoundedOutput::default();
for item in prefix {
terminal.push(item.clone());
ordinary.push(item);
}
terminal.push_error(&error);
ordinary.push(text(&format!("{label}{error}")));
assert!(retained_capacity(&terminal) < MAX_OUTPUT_BYTES + TEXT_BLOCK_BYTES);
let actual = terminal.into_collected_output();
assert_eq!(actual.gaps.len(), gap_count);
assert!(actual == ordinary.into_collected_output());
}
}
@@ -30,6 +30,10 @@ const MAX_POSITIONED_NOTICES: usize = 4;
/// budget: Core's warning describes the retained text alone. The shared
/// formatter moves collector boundaries along with their source content.
/// If losses are too scattered for individual notices, report their totals once.
#[cfg_attr(
not(test),
allow(dead_code, reason = "Renderer is wired in the next stack layer")
)]
pub(crate) fn render_output(
output: CollectedOutput,
max_output_tokens: Option<usize>,
@@ -1,4 +1,4 @@
use codex_code_mode_protocol::FunctionCallOutputContentItem;
use crate::session_runtime::OutputItem;
use std::sync::Arc;
use super::RuntimeEvent;
@@ -67,7 +67,7 @@ pub(super) fn tool_callback(
state.next_tool_call_id = state.next_tool_call_id.saturating_add(1);
let event_tx = state.event_tx.clone();
state.pending_tool_calls.insert(id.clone(), resolver);
let _ = event_tx.send(RuntimeEvent::ToolCall {
let _ = event_tx.blocking_send(RuntimeEvent::ToolCall {
id,
name: tool_name,
kind: tool_kind,
@@ -96,9 +96,9 @@ pub(super) fn text_callback(
if let Some(state) = scope.get_slot::<RuntimeState>()
&& !state.exit_requested
{
let _ = state.event_tx.send(RuntimeEvent::ContentItem(
FunctionCallOutputContentItem::InputText { text },
));
let _ = state
.event_tx
.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText { text }));
}
retval.set(v8::undefined(scope).into());
}
@@ -120,7 +120,9 @@ pub(super) fn audio_callback(
if let Some(state) = scope.get_slot::<RuntimeState>()
&& !state.exit_requested
{
let _ = state.event_tx.send(RuntimeEvent::ContentItem(audio_item));
let _ = state
.event_tx
.blocking_send(RuntimeEvent::ContentItem(audio_item));
}
retval.set(v8::undefined(scope).into());
}
@@ -155,7 +157,9 @@ pub(super) fn image_callback(
if let Some(state) = scope.get_slot::<RuntimeState>()
&& !state.exit_requested
{
let _ = state.event_tx.send(RuntimeEvent::ContentItem(image_item));
let _ = state
.event_tx
.blocking_send(RuntimeEvent::ContentItem(image_item));
}
retval.set(v8::undefined(scope).into());
}
@@ -184,11 +188,13 @@ pub(super) fn generated_image_callback(
if let Some(state) = scope.get_slot::<RuntimeState>()
&& !state.exit_requested
{
let _ = state.event_tx.send(RuntimeEvent::ContentItem(image_item));
let _ = state
.event_tx
.blocking_send(RuntimeEvent::ContentItem(image_item));
if let Some(text) = output_hint {
let _ = state.event_tx.send(RuntimeEvent::ContentItem(
FunctionCallOutputContentItem::InputText { text },
));
let _ = state
.event_tx
.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText { text }));
}
}
retval.set(v8::undefined(scope).into());
@@ -303,7 +309,7 @@ pub(super) fn notify_callback(
if let Some(state) = scope.get_slot::<RuntimeState>()
&& !state.exit_requested
{
let _ = state.event_tx.send(RuntimeEvent::Notify {
let _ = state.event_tx.blocking_send(RuntimeEvent::Notify {
call_id: state.tool_call_id.clone(),
text,
});
@@ -349,7 +355,7 @@ pub(super) fn yield_control_callback(
if let Some(state) = scope.get_slot::<RuntimeState>()
&& !state.exit_requested
{
let _ = state.event_tx.send(RuntimeEvent::YieldRequested);
let _ = state.event_tx.blocking_send(RuntimeEvent::YieldRequested);
}
}
@@ -16,7 +16,7 @@ use super::super::RuntimeEvent;
use super::super::RuntimeState;
use super::super::ToolCallbackMetadata;
use super::super::globals;
use crate::FunctionCallOutputContentItem;
use crate::session_runtime::OutputItem;
use crate::v8_init::ensure_v8_initialized;
// Deliberately leave JS running: the test must verify host-side admission even
@@ -109,7 +109,7 @@ clearTimeout(earlyTimer); setTimeout(() => {}, 60000);
)
.expect("install globals");
scope.set_slot(RuntimeState {
event_tx,
event_tx: event_tx.into(),
pending_tool_calls: HashMap::new(),
pending_timeouts: HashMap::new(),
stored_values: HashMap::new(),
@@ -152,7 +152,7 @@ const earlyTimer = setTimeout(() => text("early timer"), 60000);
assert!(
matches!(event_rx.try_recv(), Ok(RuntimeEvent::ContentItem(
FunctionCallOutputContentItem::InputText { text }
OutputItem::InputText { text }
)) if text == "before"),
"{label}: pre-exit output"
);
@@ -0,0 +1,77 @@
//! Ordered runtime event transport with backpressure for bounded collectors.
use tokio::sync::mpsc;
use super::RuntimeEvent;
use codex_code_mode_protocol::CodeModeOutputTruncation;
// Amortize small-write handoffs while keeping the backlog of raw payloads
// ahead of the bounded collector small.
const EVENT_CAPACITY: usize = 4;
#[derive(Clone)]
pub(crate) enum RuntimeEventSender {
Legacy(mpsc::UnboundedSender<RuntimeEvent>),
Bounded(mpsc::Sender<RuntimeEvent>),
}
pub(crate) enum RuntimeEventReceiver {
Legacy(mpsc::UnboundedReceiver<RuntimeEvent>),
Bounded(mpsc::Receiver<RuntimeEvent>),
}
pub(crate) fn runtime_event_channel(
mode: CodeModeOutputTruncation,
) -> (RuntimeEventSender, RuntimeEventReceiver) {
match mode {
CodeModeOutputTruncation::Client => {
let (tx, rx) = mpsc::unbounded_channel();
(
RuntimeEventSender::Legacy(tx),
RuntimeEventReceiver::Legacy(rx),
)
}
CodeModeOutputTruncation::Server => {
let (tx, rx) = mpsc::channel(EVENT_CAPACITY);
(
RuntimeEventSender::Bounded(tx),
RuntimeEventReceiver::Bounded(rx),
)
}
}
}
impl From<mpsc::UnboundedSender<RuntimeEvent>> for RuntimeEventSender {
fn from(value: mpsc::UnboundedSender<RuntimeEvent>) -> Self {
Self::Legacy(value)
}
}
impl RuntimeEventReceiver {
pub(crate) async fn recv(&mut self) -> Option<RuntimeEvent> {
match self {
Self::Legacy(rx) => rx.recv().await,
Self::Bounded(rx) => rx.recv().await,
}
}
}
impl RuntimeEventSender {
/// Bounded sends run only on the dedicated V8 thread, preserving event order
/// without making synchronous JavaScript output helpers return promises.
/// Capacity limits the backlog between V8 and the cell actor.
///
/// # Panics
///
/// In server mode, panics if called from an async execution context.
pub(crate) fn blocking_send(&self, event: RuntimeEvent) -> Result<(), ()> {
match self {
Self::Legacy(tx) => tx.send(event).map_err(|_| ()),
Self::Bounded(tx) => tx.blocking_send(event).map_err(|_| ()),
}
}
}
#[cfg(test)]
#[path = "event_channel_tests.rs"]
mod tests;
@@ -0,0 +1,133 @@
//! Raw emissions and control events share one bounded FIFO before collection.
use pretty_assertions::assert_eq;
use tokio::sync::oneshot;
use super::*;
use crate::session_runtime::OutputItem;
use codex_code_mode_protocol::host::BOUNDED_OUTPUT_MAX_BYTES as MAX_OUTPUT_BYTES;
#[tokio::test]
async fn a_blocked_send_uses_the_released_slot_before_the_receiver_takes_the_item() {
let (tx, mut rx) = runtime_event_channel(CodeModeOutputTruncation::Server);
let RuntimeEventSender::Bounded(raw) = tx.clone() else {
panic!("expected bounded sender");
};
let mut permits = raw.reserve_many(EVENT_CAPACITY).await.unwrap();
let (done_tx, mut done_rx) = oneshot::channel();
let text = "😀".repeat(MAX_OUTPUT_BYTES);
let producer_text = text.clone();
let producer = std::thread::spawn(move || {
let result = tx.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText {
text: producer_text,
}));
let _ = done_tx.send(result);
});
// Queue a competing waiter before releasing the slot. A blocking producer can
// win the slot in FIFO order; a try_send can never find an unreserved slot.
let sent = tokio::time::timeout(std::time::Duration::from_secs(5), async {
let mut slot = permits.next().unwrap();
loop {
let mut next = std::pin::pin!(raw.reserve());
assert!(futures::poll!(next.as_mut()).is_pending());
drop(slot);
tokio::select! {
result = &mut done_rx => break result.unwrap(),
permit = &mut next => slot = permit.unwrap(),
}
tokio::task::yield_now().await;
}
})
.await
.expect("the producer must eventually acquire the released slot");
assert_eq!(sent, Ok(()));
// Close the test's sender so a missing event fails the receive immediately.
drop(permits);
drop(raw);
let event = rx.recv().await.unwrap();
let RuntimeEvent::ContentItem(item) = event else {
panic!("expected raw content");
};
assert_eq!(item, OutputItem::InputText { text });
producer.join().unwrap();
}
#[tokio::test]
async fn content_and_control_are_received_in_send_order() {
let (tx, mut rx) = runtime_event_channel(CodeModeOutputTruncation::Server);
let producer = std::thread::spawn(move || {
tx.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputImage {
image_url: "image".to_owned(),
detail: None,
}))
.unwrap();
tx.blocking_send(RuntimeEvent::YieldRequested).unwrap();
tx.blocking_send(RuntimeEvent::ContentItem(OutputItem::InputText {
text: "after yield".to_owned(),
}))
.unwrap();
});
assert!(matches!(
rx.recv().await,
Some(RuntimeEvent::ContentItem(OutputItem::InputImage { .. }))
));
assert!(matches!(
rx.recv().await,
Some(RuntimeEvent::YieldRequested)
));
assert!(matches!(
rx.recv().await,
Some(RuntimeEvent::ContentItem(OutputItem::InputText { text }))
if text == "after yield"
));
producer.join().unwrap();
}
#[tokio::test]
async fn full_channel_blocks_the_producer_and_receiver_drop_releases_it() {
let (tx, rx) = runtime_event_channel(CodeModeOutputTruncation::Server);
let RuntimeEventSender::Bounded(tx) = tx else {
panic!("expected bounded sender");
};
for _ in 0..EVENT_CAPACITY {
tx.try_send(RuntimeEvent::Started).unwrap();
}
let (ready_tx, ready_rx) = oneshot::channel();
let producer = tokio::spawn(async move {
// Tokio's blocking_send runs this same send future. Polling it first
// proves the full channel has registered this producer as a waiter.
let mut send = std::pin::pin!(tx.send(RuntimeEvent::YieldRequested));
assert!(futures::poll!(send.as_mut()).is_pending());
ready_tx.send(()).unwrap();
send.await
});
// This current-thread runtime cannot resume us until the producer yields.
// Closing must wake that already-waiting task, not just reject a new send.
ready_rx.await.unwrap();
drop(rx);
let result = tokio::time::timeout(std::time::Duration::from_secs(1), producer)
.await
.expect("receiver drop must wake the blocked producer")
.expect("producer task");
assert!(matches!(
result,
Err(mpsc::error::SendError(RuntimeEvent::YieldRequested))
));
}
#[tokio::test]
async fn sending_after_receiver_drop_rejects_content() {
let (tx, rx) = runtime_event_channel(CodeModeOutputTruncation::Server);
drop(rx);
assert_eq!(
std::thread::spawn(move || tx.blocking_send(RuntimeEvent::ContentItem(
OutputItem::InputAudio {
audio_url: "large audio".repeat(MAX_OUTPUT_BYTES)
}
)))
.join()
.unwrap(),
Err(())
);
}
+24 -22
View File
@@ -1,5 +1,6 @@
mod audio;
mod callbacks;
mod event_channel;
mod globals;
mod module_loader;
mod timers;
@@ -12,17 +13,18 @@ use std::sync::Arc;
use std::sync::mpsc as std_mpsc;
use std::thread;
use crate::TaskFailureHandler;
use crate::session_runtime::OutputItem;
use crate::v8_init::ensure_v8_initialized;
use codex_code_mode_protocol::CodeModeToolKind;
use codex_code_mode_protocol::EnabledToolMetadata;
use codex_code_mode_protocol::ExecuteRequest;
use codex_code_mode_protocol::FunctionCallOutputContentItem;
use codex_code_mode_protocol::normalize_code_mode_identifier;
use codex_protocol::ToolName;
pub(crate) use event_channel::RuntimeEventReceiver;
pub(crate) use event_channel::RuntimeEventSender;
pub(crate) use event_channel::runtime_event_channel;
use serde_json::Value as JsonValue;
use tokio::sync::mpsc;
use crate::TaskFailureHandler;
use crate::v8_init::ensure_v8_initialized;
#[derive(Debug)]
pub(crate) enum RuntimeCommand {
@@ -51,7 +53,7 @@ pub(crate) enum RuntimeControlCommand {
pub(crate) enum RuntimeEvent {
Started,
Pending,
ContentItem(FunctionCallOutputContentItem),
ContentItem(OutputItem),
YieldRequested,
ToolCall {
id: String,
@@ -73,7 +75,7 @@ pub(crate) enum RuntimeEvent {
pub(crate) fn spawn_runtime(
stored_values: HashMap<String, Arc<JsonValue>>,
request: ExecuteRequest,
event_tx: mpsc::UnboundedSender<RuntimeEvent>,
event_tx: impl Into<RuntimeEventSender>,
pending_mode: PendingRuntimeMode,
task_failure_handler: Option<TaskFailureHandler>,
) -> Result<
@@ -84,6 +86,7 @@ pub(crate) fn spawn_runtime(
),
String,
> {
let event_tx = event_tx.into();
ensure_v8_initialized()?;
let (command_tx, command_rx) = std_mpsc::channel();
@@ -128,16 +131,17 @@ pub(crate) fn spawn_runtime(
}
fn spawn_supervised_runtime_thread(
event_tx: mpsc::UnboundedSender<RuntimeEvent>,
event_tx: impl Into<RuntimeEventSender>,
task_failure_handler: Option<TaskFailureHandler>,
runtime: impl FnOnce() + Send + 'static,
) {
let event_tx = event_tx.into();
thread::spawn(move || {
if catch_unwind(AssertUnwindSafe(runtime)).is_err() {
if let Some(task_failure_handler) = task_failure_handler {
task_failure_handler("code-mode V8 runtime thread panicked".to_string());
}
let _ = event_tx.send(RuntimeEvent::ThreadPanicked);
let _ = event_tx.blocking_send(RuntimeEvent::ThreadPanicked);
}
});
}
@@ -157,7 +161,7 @@ struct ToolCallbackMetadata {
}
pub(super) struct RuntimeState {
event_tx: mpsc::UnboundedSender<RuntimeEvent>,
event_tx: RuntimeEventSender,
pending_tool_calls: HashMap<String, v8::Global<v8::PromiseResolver>>,
pending_timeouts: HashMap<u64, timers::ScheduledTimeout>,
stored_values: HashMap<String, Arc<JsonValue>>,
@@ -177,7 +181,7 @@ pub(super) enum CompletionState {
fn run_runtime(
config: RuntimeConfig,
event_tx: mpsc::UnboundedSender<RuntimeEvent>,
event_tx: RuntimeEventSender,
command_rx: std_mpsc::Receiver<RuntimeCommand>,
control_rx: std_mpsc::Receiver<RuntimeControlCommand>,
pending_mode: PendingRuntimeMode,
@@ -221,7 +225,7 @@ fn run_runtime(
exit_requested: false,
});
let _ = event_tx.send(RuntimeEvent::Started);
let _ = event_tx.blocking_send(RuntimeEvent::Started);
let pending_promise = match module_loader::evaluate_main_module(scope, &config.source) {
Ok(pending_promise) => pending_promise,
@@ -289,7 +293,7 @@ fn run_runtime(
}
fn next_runtime_command(
event_tx: &mpsc::UnboundedSender<RuntimeEvent>,
event_tx: &RuntimeEventSender,
command_rx: &std_mpsc::Receiver<RuntimeCommand>,
control_rx: &std_mpsc::Receiver<RuntimeControlCommand>,
pending_mode: PendingRuntimeMode,
@@ -301,7 +305,7 @@ fn next_runtime_command(
Err(std_mpsc::TryRecvError::Empty) => {}
}
let _ = event_tx.send(RuntimeEvent::Pending);
let _ = event_tx.blocking_send(RuntimeEvent::Pending);
match pending_mode {
#[cfg(test)]
PendingRuntimeMode::Continue => return command_rx.recv().ok(),
@@ -316,7 +320,7 @@ fn next_runtime_command(
fn send_scope_result(
scope: &mut v8::PinScope<'_, '_>,
event_tx: &mpsc::UnboundedSender<RuntimeEvent>,
event_tx: &RuntimeEventSender,
error_text: Option<String>,
) {
// Error rendering can itself run JS getters, including exit(). Capture
@@ -330,11 +334,11 @@ fn send_scope_result(
}
fn send_result(
event_tx: &mpsc::UnboundedSender<RuntimeEvent>,
event_tx: &RuntimeEventSender,
stored_value_writes: HashMap<String, Arc<JsonValue>>,
error_text: Option<String>,
) {
let _ = event_tx.send(RuntimeEvent::Result {
let _ = event_tx.blocking_send(RuntimeEvent::Result {
stored_value_writes,
error_text,
});
@@ -355,7 +359,7 @@ mod tests {
use super::RuntimeEvent;
use super::spawn_runtime;
use super::spawn_supervised_runtime_thread;
use crate::FunctionCallOutputContentItem;
use crate::session_runtime::OutputItem;
fn execute_request(source: &str) -> ExecuteRequest {
ExecuteRequest {
@@ -530,9 +534,7 @@ await new Promise(() => {});
.await
.unwrap()
.unwrap();
let RuntimeEvent::ContentItem(FunctionCallOutputContentItem::InputText { text }) =
content_event
else {
let RuntimeEvent::ContentItem(OutputItem::InputText { text }) = content_event else {
panic!("expected resumed runtime output");
};
assert_eq!(text, "after");
@@ -602,7 +604,7 @@ setTimeout(() => text("late"), 3_600_000);
}
assert_eq!(
output,
vec![FunctionCallOutputContentItem::InputText {
vec![OutputItem::InputText {
text: "done".to_string()
}]
);
@@ -22,7 +22,7 @@ fn discarded_tool_response_is_collectible_before_cell_ends() {
let (event_tx, _event_rx) = mpsc::unbounded_channel();
let (runtime_command_tx, _runtime_command_rx) = std_mpsc::channel();
scope.set_slot(RuntimeState {
event_tx,
event_tx: event_tx.into(),
pending_tool_calls: HashMap::new(),
pending_timeouts: HashMap::new(),
stored_values: HashMap::new(),
@@ -1,7 +1,7 @@
use serde_json::Value as JsonValue;
use crate::session_runtime::OutputItem;
use codex_code_mode_protocol::DEFAULT_IMAGE_DETAIL;
use codex_code_mode_protocol::FunctionCallOutputContentItem;
use codex_code_mode_protocol::ImageDetail;
use super::RuntimeState;
@@ -48,8 +48,8 @@ pub(super) fn normalize_output_image(
scope: &mut v8::PinScope<'_, '_>,
value: v8::Local<'_, v8::Value>,
detail_override: Option<String>,
) -> Result<FunctionCallOutputContentItem, ()> {
let result = (|| -> Result<FunctionCallOutputContentItem, String> {
) -> Result<OutputItem, ()> {
let result = (|| -> Result<OutputItem, String> {
let (image_url, detail) = if value.is_string() {
(value.to_rust_string_lossy(scope), None)
} else if value.is_object() && !value.is_array() {
@@ -96,7 +96,7 @@ pub(super) fn normalize_output_image(
None => Some(DEFAULT_IMAGE_DETAIL),
};
Ok(FunctionCallOutputContentItem::InputImage { image_url, detail })
Ok(OutputItem::InputImage { image_url, detail })
})();
match result {
@@ -191,8 +191,8 @@ fn parse_image_detail_value<'s>(
pub(super) fn normalize_output_audio(
scope: &mut v8::PinScope<'_, '_>,
value: v8::Local<'_, v8::Value>,
) -> Result<FunctionCallOutputContentItem, ()> {
let result = (|| -> Result<FunctionCallOutputContentItem, String> {
) -> Result<OutputItem, ()> {
let result = (|| -> Result<OutputItem, String> {
let audio_url = if value.is_string() {
value.to_rust_string_lossy(scope)
} else if value.is_object() && !value.is_array() {
@@ -219,13 +219,13 @@ pub(super) fn normalize_output_audio(
// Tiny tool-generated clips cannot be encoded reliably by audio models.
if wav_duration_seconds(&audio_url).is_some_and(|duration| duration < 0.025) {
return Ok(FunctionCallOutputContentItem::InputText {
return Ok(OutputItem::InputText {
text: "Audio output omitted because the clip is shorter than 25 ms; use a longer clip."
.to_string(),
});
}
Ok(FunctionCallOutputContentItem::InputAudio { audio_url })
Ok(OutputItem::InputAudio { audio_url })
})();
match result {
+15 -25
View File
@@ -3,6 +3,7 @@ use std::time::Duration;
use codex_code_mode_protocol::CellId;
use codex_code_mode_protocol::CodeModeNestedToolCall;
use codex_code_mode_protocol::CodeModeOutputTruncation;
use codex_code_mode_protocol::CodeModeSession;
use codex_code_mode_protocol::CodeModeSessionCellExecutionLimits;
use codex_code_mode_protocol::CodeModeSessionDelegate;
@@ -11,8 +12,8 @@ use codex_code_mode_protocol::CodeModeToolKind;
use codex_code_mode_protocol::DEFAULT_EXEC_YIELD_TIME_MS;
use codex_code_mode_protocol::ExecuteRequest;
use codex_code_mode_protocol::ExecuteToPendingOutcome;
#[cfg(test)]
use codex_code_mode_protocol::FunctionCallOutputContentItem;
use codex_code_mode_protocol::ImageDetail;
use codex_code_mode_protocol::RuntimeResponse;
use codex_code_mode_protocol::StartedCell;
use codex_code_mode_protocol::TerminateRequest;
@@ -319,6 +320,7 @@ impl runtime::SessionRuntimeDelegate for ProtocolDelegate {
fn runtime_request(request: ExecuteRequest) -> runtime::CreateCellRequest {
runtime::CreateCellRequest {
output_truncation: CodeModeOutputTruncation::Client,
tool_call_id: request.tool_call_id,
enabled_tools: request
.enabled_tools
@@ -358,7 +360,9 @@ fn pending_outcome(
pending_tool_call_ids,
} => Ok(ExecuteToPendingOutcome::Pending {
cell_id: cell_id.clone(),
content_items: content_items.into_iter().map(output_item).collect(),
content_items: content_items
.into_collected_output()
.into_untruncated_items(),
pending_tool_call_ids,
}),
event => Ok(ExecuteToPendingOutcome::Completed(runtime_response(
@@ -374,7 +378,9 @@ fn runtime_response(
match event {
runtime::CellEvent::Yielded { content_items } => Ok(RuntimeResponse::Yielded {
cell_id: cell_id.clone(),
content_items: content_items.into_iter().map(output_item).collect(),
content_items: content_items
.into_collected_output()
.into_untruncated_items(),
code_mode_host_duration: None,
}),
runtime::CellEvent::Completed {
@@ -383,14 +389,18 @@ fn runtime_response(
result_buffered_for,
} => Ok(RuntimeResponse::Result {
cell_id: cell_id.clone(),
content_items: content_items.into_iter().map(output_item).collect(),
content_items: content_items
.into_collected_output()
.into_untruncated_items(),
error_text,
result_buffered_for,
code_mode_host_duration: None,
}),
runtime::CellEvent::Terminated { content_items } => Ok(RuntimeResponse::Terminated {
cell_id: cell_id.clone(),
content_items: content_items.into_iter().map(output_item).collect(),
content_items: content_items
.into_collected_output()
.into_untruncated_items(),
code_mode_host_duration: None,
}),
runtime::CellEvent::Pending { .. } => {
@@ -399,26 +409,6 @@ fn runtime_response(
}
}
fn output_item(item: runtime::OutputItem) -> FunctionCallOutputContentItem {
match item {
runtime::OutputItem::Text { text } => FunctionCallOutputContentItem::InputText { text },
runtime::OutputItem::Image { image_url, detail } => {
FunctionCallOutputContentItem::InputImage {
image_url,
detail: detail.map(|detail| match detail {
runtime::ImageDetail::Auto => ImageDetail::Auto,
runtime::ImageDetail::Low => ImageDetail::Low,
runtime::ImageDetail::High => ImageDetail::High,
runtime::ImageDetail::Original => ImageDetail::Original,
}),
}
}
runtime::OutputItem::Audio { audio_url } => {
FunctionCallOutputContentItem::InputAudio { audio_url }
}
}
}
fn missing_cell_response(cell_id: CellId) -> RuntimeResponse {
RuntimeResponse::Result {
error_text: Some(format!("exec cell {cell_id} not found")),
@@ -7,6 +7,7 @@ use std::sync::Arc;
use std::sync::atomic::AtomicU64;
use std::sync::atomic::Ordering;
pub(crate) use codex_code_mode_protocol::FunctionCallOutputContentItem as OutputItem;
use opentelemetry::context::FutureExt;
use serde_json::Value as JsonValue;
use tokio::sync::Mutex;
@@ -17,11 +18,9 @@ pub(crate) use self::types::CellEvent;
pub(crate) use self::types::CellId;
pub(crate) use self::types::CreateCellRequest;
pub(crate) use self::types::Error;
pub(crate) use self::types::ImageDetail;
pub(crate) use self::types::NestedToolCall;
pub(crate) use self::types::Observation;
pub(crate) use self::types::ObserveMode;
pub(crate) use self::types::OutputItem;
pub(crate) use self::types::SessionRuntimeDelegate;
pub(crate) use self::types::ToolDefinition;
pub(crate) use self::types::ToolKind;
@@ -286,19 +285,20 @@ impl<D: SessionRuntimeDelegate> CellHost for RuntimeCellHost<D> {
async fn commit_completion(
&self,
stored_value_writes: HashMap<String, Arc<JsonValue>>,
event: CellEvent,
event: impl FnOnce() -> CellEvent + Send,
cell_state: Arc<CellState>,
) -> CompletionCommit {
let cancellation_token = cell_state.cancellation_token();
let mut stored_values = tokio::select! {
biased;
_ = cancellation_token.cancelled() => {
return CompletionCommit::Rejected(event);
return CompletionCommit::Rejected;
}
stored_values = self.inner.stored_values.lock() => stored_values,
};
cell_state.commit_completion(event, || {
stored_values.extend(stored_value_writes);
drop(stored_values);
})
}
@@ -6,12 +6,14 @@ use std::task::Poll;
use std::task::Waker;
use std::time::Duration;
use codex_code_mode_protocol::CodeModeOutputTruncation;
use pretty_assertions::assert_eq;
use serde_json::Value as JsonValue;
use tokio_util::sync::CancellationToken;
use super::*;
use crate::cell_actor::CompletionCommit;
use crate::output::OutputCollector;
struct RecordingDelegate;
@@ -80,9 +82,9 @@ async fn reports_cell_actor_panics_to_the_owner() {
assert_eq!(
started.initial_event().await,
Ok(CellEvent::Completed {
content_items: vec![OutputItem::Text {
content_items: OutputCollector::Client(vec![OutputItem::InputText {
text: "done".to_string(),
}],
}]),
result_buffered_for: Duration::ZERO,
error_text: None,
})
@@ -104,21 +106,13 @@ async fn termination_rejects_a_waiting_store_commit_before_the_next_cell_can_loa
inner: Arc::clone(&runtime.inner),
execution_context: opentelemetry::Context::new(),
};
let completion = CellEvent::Completed {
content_items: vec![OutputItem::Text {
text: "uncommitted output".to_string(),
}],
result_buffered_for: Duration::ZERO,
error_text: None,
};
let stored_values = runtime.inner.stored_values.lock().await;
let commit = host.commit_completion(
HashMap::from([(
"candidate".to_string(),
Arc::new(JsonValue::String("lost".to_string())),
)]),
completion.clone(),
|| panic!("terminated cell must not construct completion output"),
Arc::clone(&cell_state),
);
tokio::pin!(commit);
@@ -128,9 +122,9 @@ async fn termination_rejects_a_waiting_store_commit_before_the_next_cell_can_loa
let termination = cell_state.request_termination();
drop(stored_values);
assert_eq!(commit.await, CompletionCommit::Rejected(completion));
assert_eq!(commit.await, CompletionCommit::Rejected);
let terminated = CellEvent::Terminated {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
};
assert_eq!(
cell_state.finish_termination(terminated.clone()),
@@ -149,6 +143,7 @@ async fn termination_rejects_a_waiting_store_commit_before_the_next_cell_can_loa
let reader = runtime
.execute(
CreateCellRequest {
output_truncation: CodeModeOutputTruncation::Client,
tool_call_id: "reader".to_string(),
enabled_tools: Vec::new(),
source: r#"text(String(load("candidate")));"#.to_string(),
@@ -161,9 +156,9 @@ async fn termination_rejects_a_waiting_store_commit_before_the_next_cell_can_loa
assert_eq!(
reader.initial_event().await,
Ok(CellEvent::Completed {
content_items: vec![OutputItem::Text {
content_items: OutputCollector::Client(vec![OutputItem::InputText {
text: "undefined".to_string(),
}],
}]),
result_buffered_for: Duration::ZERO,
error_text: None,
})
@@ -173,6 +168,7 @@ async fn termination_rejects_a_waiting_store_commit_before_the_next_cell_can_loa
fn execute_request(source: &str) -> CreateCellRequest {
CreateCellRequest {
output_truncation: CodeModeOutputTruncation::Client,
tool_call_id: "call-1".to_string(),
enabled_tools: Vec::new(),
source: source.to_string(),
@@ -255,7 +251,7 @@ async fn drop_terminates_cells_when_the_registry_is_locked() {
assert_eq!(
started.initial_event().await,
Ok(CellEvent::Yielded {
content_items: Vec::new(),
content_items: OutputCollector::Client(Vec::new()),
})
);
@@ -5,6 +5,8 @@ use std::time::Duration;
use serde_json::Value as JsonValue;
use tokio_util::sync::CancellationToken;
use crate::output::OutputCollector;
/// Identifies one execution cell within a session runtime.
#[derive(Clone, Debug, Eq, Hash, PartialEq)]
pub(crate) struct CellId(String);
@@ -54,54 +56,32 @@ impl From<ObserveMode> for Observation {
}
/// An observable cell lifecycle event.
#[derive(Clone, Debug, PartialEq)]
#[derive(Clone, Debug)]
#[cfg_attr(test, derive(PartialEq))]
pub(crate) enum CellEvent {
Yielded {
content_items: Vec<OutputItem>,
content_items: OutputCollector,
},
Pending {
content_items: Vec<OutputItem>,
content_items: OutputCollector,
pending_tool_call_ids: Vec<String>,
},
Completed {
content_items: Vec<OutputItem>,
content_items: OutputCollector,
error_text: Option<String>,
/// Time the completed result spent waiting for an observer.
result_buffered_for: Duration,
},
Terminated {
content_items: Vec<OutputItem>,
content_items: OutputCollector,
},
}
/// Output emitted by a cell since its preceding observation.
#[derive(Clone, Debug, PartialEq)]
pub(crate) enum OutputItem {
Text {
text: String,
},
Image {
image_url: String,
detail: Option<ImageDetail>,
},
Audio {
audio_url: String,
},
}
/// Requested image fidelity for an output image.
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub(crate) enum ImageDetail {
Auto,
Low,
High,
Original,
}
/// Transport-neutral input for creating a cell.
///
/// The owning session assigns the cell ID when it admits the request.
pub(crate) struct CreateCellRequest {
pub(crate) output_truncation: codex_code_mode_protocol::CodeModeOutputTruncation,
pub(crate) tool_call_id: String,
pub(crate) enabled_tools: Vec<ToolDefinition>,
pub(crate) source: String,